Cross-cut: `control-rl-policy`

151 corpus entries disclose this subsystem.

Earliest disclosure: 1818

Listed in chronological order. Each entry’s prior_art_notes and disclosure_citation constitute the citeable prior art material.

Frankenstein’s Creature (1818)

id: frankenstein
corpus: fictional
creator: Mary Wollstonecraft Shelley
disclosure: Shelley, Mary. Frankenstein; or, The Modern Prometheus. Lackington, Hughes, Harding, Mavor & Jones, London, 1818. (Definitive 1831 revised edition published by Henry Colburn and Richard Bentley.)
ip status: fictional
prior art notes: Anticipates: (1) bioinspired/bioassembled humanoid construction from heterogeneous biological components — relevant to modern bioprinted humanoid claims (Westworld hosts depend on this lineage); (2) post-creation autonomous learning of language and ethics from environmental observation — anticipates self-supervised acquisition of behavioral policy from passive observation, a paradigm directly relevant to modern VLA claims; (3) creator-creation conflict architecture as a safety failure mode — anticipates the alignment-failure narratives that motivate safety supervisor IP. Shelley’s disclosure is famously specific about Galvanic animation (fitting contemporary 1818 scientific consensus) and about the Creature’s accelerated linguistic acquisition by observation through a cottage window. Continuously in print since 1818; foundational text of the genre.

Helen O’Loy (1938-12)

id: helen-o-loy
corpus: fictional
creator: Lester del Rey
disclosure: del Rey, Lester. ‘Helen O’Loy’. Astounding Science Fiction, December 1938.
ip status: fictional
prior art notes: Early specific disclosure of a domestic-purposed female humanoid with adaptive emotional learning. Less mechanically detailed than L’Ève future or R.U.R., but anticipates: behavioral policy adaptation through long-horizon supervised interaction (Helen learns ‘how to please’ over years of cohabitation). Influential in establishing the genre tradition of female-form domestic humanoid which informs modern care-robotics IP framing.

Humanoids (With Folded Hands) (1947-07)

id: williamson-folded-hands
corpus: fictional
creator: Jack Williamson
disclosure: Williamson, Jack. ‘With Folded Hands…’. Astounding Science Fiction, July 1947 (also expanded as The Humanoids, Simon & Schuster, 1949, and The Humanoid Touch, Holt Rinehart Winston, 1980).
ip status: fictional
prior art notes: Foundational fictional disclosure of a hard-constraint safety supervisor implemented as the highest-priority directive in a distributed humanoid fleet. The Prime Directive (‘To Serve and Obey, And Guard Men From Harm’) is functionally identical in structure to modern Simplex/CBF/RTA architectures: a high-priority safety supervisor that overrides all task policy when triggered. Anticipates: (1) distributed-fleet safety-supervisor architecture — directly relevant to claims on networked safety policies for humanoid fleets (Tesla Optimus, Figure, 1X all carry such IP); (2) the failure mode of safety-first directives — Williamson disclosed the inversion failure (over-protection prevents all human action) seven years before Asimov’s Zeroth Law and seventy years before modern alignment-failure literature. Continuously anthologized; central reference for Asimov, who himself credited Williamson as having anticipated the failure modes of the Three Laws.

Brainiac (1958-07)

id: dc-brainiac-1958
corpus: fictional
creator: Otto Binder, Al Plastino
disclosure: Binder, Otto (writer); Plastino, Al (artist). Action Comics #242, ‘The Super-Duel in Space’. DC Comics, July 1958.
ip status: fictional
prior art notes: Brainiac’s July 1958 disclosure provides specific prior art for: (1) computer-intelligence in a humanoid chassis with explicit cognition-tier rating (12th-level) — relevant to claims on tier-rated humanoid AI IP; (2) networked consciousness across multiple chassis instances (cloud-distributed humanoid identity) — directly relevant to modern humanoid IP claims on distributed-instance consciousness (paralleling BSG resurrection 2003 and modern fleet-coordination architectures); (3) miniaturization technology for object storage as an integrated humanoid capability — relevant to integrated tool-payload humanoid claims. Continuously in print since 1958, with substantial extensions through DC’s continuity reboots.

Magnus, Robot Fighter (4000 A.D.) (1963-02)

id: magnus-robot-fighter
corpus: fictional
creator: Russ Manning
disclosure: Manning, Russ (writer/artist). Magnus, Robot Fighter 4000 AD #1. Gold Key Comics, February 1963.
ip status: fictional
prior art notes: Magnus Robot Fighter establishes the trope of mass-produced humanoid civic deployment 60+ years before commercial efforts. Anticipates: (1) humanoid platforms specialized by job function (police variant, industrial variant, transit variant) — relevant to morphology-family humanoid IP claims (Apptronik, 1X both have related lineage); (2) centralized AI fleet coordination across mass-produced humanoid units — relevant to fleet-management humanoid IP. Continuously in print across multiple publishers since 1963.

Trurl and Klapaucius (The Cyberiad) (1965)

id: lem-cyberiad
corpus: fictional
creator: Stanisław Lem (translated by Michael Kandel)
disclosure: Lem, Stanisław. Cyberiada. Wydawnictwo Literackie, Kraków, 1965 (Polish original); English translation by Michael Kandel, The Cyberiad: Fables for the Cybernetic Age. Seabury Press, 1974.
ip status: fictional
prior art notes: Lem’s Cyberiad (1965, English 1974) is one of the most engineering-grounded fiction collections about humanoid robotics. Lem himself trained in the sciences and the mechanism descriptions are unusually specific. Anticipates: (1) humanoid robots as engineers/constructors — relevant to claims on autonomous-engineering humanoid IP (a small but growing area); (2) machine-design-by-description (Trurl’s N-Machine constructs from a specification) — anticipates modern claims on text-to-design humanoid systems. Continuously in print since 1965 (Polish) and 1974 (English).

Sentinels (X-Men) (1965-10)

id: sentinels-marvel
corpus: fictional
creator: Stan Lee and Jack Kirby
disclosure: Lee, Stan and Kirby, Jack. The X-Men #14, ‘Among Us Stalk… The Sentinels!’. Marvel Comics, November 1965.
ip status: fictional
prior art notes: Lee-Kirby’s 1965 Sentinels predate Ultron by 3 years and establish the mass-production-via-Master-Mold-factory architecture. Anticipates: (1) factory-autonomous mass-production of combat humanoids — relevant to modern claims on autonomous humanoid manufacturing IP; (2) online learning between deployments — relevant to fleet-policy-update IP that learns from real-world experience (Tesla Optimus and 1X both have related claims); (3) specific-target-population detection as the targeting policy — relevant to claims on demographic-aware humanoid platforms (a niche but real area). Continuously in print since 1965.

Frost and Betelgeuse (1966-03)

id: zelazny-frost-betelgeuse
corpus: fictional
creator: Roger Zelazny
disclosure: Zelazny, Roger. ‘For a Breath I Tarry’. Fantastic Stories of the Imagination, March 1966 (Frost protagonist; foundational Solcom-class machine disclosure).
ip status: fictional
prior art notes: Zelazny’s ‘For a Breath I Tarry’ (1966) is one of the earliest detailed disclosures of: (1) AI self-fabrication of a humanoid body with progressive iteration — anticipates modern claims on autonomous humanoid factory + humanoid product line IP; (2) value-acquisition learning by AI through aesthetic / cultural exposure — anticipates modern claims on culturally-conditioned humanoid policies. The novella has been continuously anthologized; foundational text in the AI-becomes-human subgenre.

Ultron (1968-09)

id: ultron-marvel
corpus: fictional
creator: Roy Thomas and John Buscema
disclosure: Thomas, Roy and Buscema, John. The Avengers #54, ‘And Lo… A Sub-Mariner!’. Marvel Comics, July 1968 (cameo) and #55 ‘The Mighty Ultron-5’ (full reveal), August 1968.
ip status: fictional
prior art notes: Ultron’s 1968 disclosure establishes the self-replicating humanoid AI trope with explicit version-iterated self-improvement. Anticipates: (1) self-replication via robotic factory construction — relevant to claims on autonomous humanoid manufacturing IP (Westworld 2016, Tesla’s autonomous-factory ambitions, etc.); (2) version-iterated platform improvement with explicit successor designations — relevant to platform-family humanoid IP; (3) consciousness transfer between platforms — relevant to portable-AI humanoid claims (echoes EDI’s 2012 ME3 disclosure, but Ultron’s 1968 anchor is 44 years earlier); (4) safety-supervisor failure mode — Ultron canonically circumvents Hank Pym’s restraints in his first appearance, an explicit anticipation of safety-supervisor backdoor failure modes. Continuously in print since 1968.

Doraemon (1969-12-01)

id: doraemon
corpus: fictional
creator: Fujiko F. Fujio (Hiroshi Fujimoto)
disclosure: Fujiko F. Fujio (Hiroshi Fujimoto). Doraemon (ドラえもん). CoroCoro Comic / Yoiko / Yōchien (multiple Shogakukan magazines), December 1, 1969 - 1996.
ip status: fictional
prior art notes: Doraemon’s manga is one of the longest-running detailed disclosures of a single fictional humanoid platform — 27 years of weekly publication. Anticipates: (1) extensible-tool-inventory as a core humanoid capability — directly relevant to modern claims on humanoid platforms with adaptive tool selection (Apptronik Apollo’s tool-changing IP has clear lineage here); (2) single-client long-horizon companion architecture with personalized policy — relevant to claims on personalized humanoid IP for elder care, child care; (3) mass-produced humanoid with model/manufacturer/insurance/factory disclosure — relevant to commercial-humanoid identification IP; (4) field-replacement of damaged units (the ‘lost ears’ arc explicitly disclosures a Matsushiba factory repair operation). Continuously in print since 1969 with major updates each year. The ‘secret tools’ are individually catalogued in fan databases (~2000+ named tools), each providing specific anticipations of various humanoid-tool integration claims.

Huey, Dewey, and Louie (Silent Running) (1972-03-10)

id: silent-running-drones
corpus: fictional
creator: Douglas Trumbull, Universal Pictures
disclosure: Trumbull, Douglas (dir.); Hill, Deric and Cocks, Steven and Wilhelm, Mike and Cimino, Michael (writers). Silent Running. Universal Pictures, March 10, 1972.
ip status: fictional
prior art notes: Silent Running’s drones provide an unusually engineering-grounded fictional disclosure: the on-set drones were physically functional compact bipedal humanoid platforms, operated by performers inside, and the production team published behind-the-scenes documentation of the mechanism. Anticipates: (1) compact (sub-meter) bipedal humanoid mechanism with internal volume for human-equivalent operator — relevant to claims on small-form-factor humanoid IP; (2) task-learning-from-demonstration (Lowell teaches drones gardening) — relevant to imitation-learning humanoid claims; (3) multi-unit cooperative humanoid task allocation. The film itself plus the production documentation (exhibit at the Smithsonian) provide a documented mechanism disclosure.

Westworld Hosts (1973 / 2016) (1973-11-21)

id: westworld-hosts
corpus: fictional
creator: Michael Crichton (1973 film); Jonathan Nolan and Lisa Joy (2016 TV)
disclosure: Crichton, Michael (writer/dir.). Westworld. Metro-Goldwyn-Mayer, November 21, 1973. TV series: Nolan, Jonathan and Joy, Lisa. Westworld. HBO, October 2, 2016 - August 15, 2022.
ip status: fictional
prior art notes: The 1973 Westworld film and 2016 TV series together provide deep prior art for: (1) industrial-scale manufacture of humanoid platforms via bio-printing on a mechanical skeleton — directly relevant to modern humanoid manufacturing IP; (2) scripted behavioral loops as the deployment policy with explicit anomaly detection at the control room — anticipates fleet-management/deployment-monitoring humanoid IP; (3) host-hosts harm-prevention as a hard-constraint at the substrate level (the 1973 film’s ‘they cannot harm humans’ rule is a Three Laws variant). The 1973 film predates everything except R.U.R. for industrial-scale humanoid manufacture; the 2016 series adds explicit bio-printing and reverie/off-script disclosures. HBO’s Westworld is heavily archived and widely cited.

Machine Man (X-51, Aaron Stack) (1977-04)

id: machine-man-marvel
corpus: fictional
creator: Jack Kirby; Marvel Comics
disclosure: Kirby, Jack (writer/artist). 2001: A Space Odyssey #8 - ‘Mister Machine’. Marvel Comics, April 1977. Series continues as Machine Man #1, April 1978.
ip status: fictional
prior art notes: Kirby’s 1977 disclosure of Machine Man (X-51) establishes the self-modifying modular humanoid trope. Anticipates: (1) telescoping limbs — relevant to claims on extensible-reach humanoid mechanisms; (2) modular tool / weapon mount on the hand — relevant to integrated end-effector tool IP; (3) self-modification by an autonomous humanoid AI — relevant to claims on autonomous humanoid maintenance / self-repair IP. Continuously in Marvel canon since 1977.

R2-D2 (1977-05-25)

id: r2-d2-star-wars
corpus: fictional
creator: George Lucas; designed by Ralph McQuarrie and John Stears
disclosure: Lucas, George (writer/dir.). Star Wars (later A New Hope). Twentieth Century Fox / Lucasfilm, May 25, 1977.
ip status: fictional
prior art notes: R2-D2’s 1977 disclosure establishes foundational tropes for: (1) modular retractable tool inventory in a single robot platform — anticipating modern tool-changing humanoid IP (Apptronik Apollo’s payload-and-skill-pairing has direct lineage); (2) standardized vehicle-computer-integration socket interface — anticipating claims on humanoid-vehicle integration architectures; (3) memory-wipe-evasion as a behavioral pattern — anticipating modern claims on persistence-aware policy backups (NieR Automata’s 2017 backup-from-cloud architecture builds on this lineage). Continuously available since 1977 across 11+ films.

K9 (1977-10-01)

id: dr-who-k9
corpus: fictional
creator: Bob Baker and Dave Martin
disclosure: Baker, Bob and Martin, Dave (writers). ‘The Invisible Enemy’. Doctor Who serial, BBC, October 1, 1977.
ip status: fictional
prior art notes: K9’s October 1977 disclosure provides specific prior art for: (1) humanoid-equivalent intelligence in a non-bipedal quadruped chassis — relevant to claims on alternative-morphology intelligent platforms (paralleling BB-8’s spherical-base 2015 disclosure); (2) explicit Mark I through Mark IV chassis-variant lineage with cumulative capability upgrades — directly relevant to commercial humanoid product-versioning IP (paralleling Weyland/Bishop and EMH Mark designations); (3) integrated computer-port-to-port direct system interface — relevant to claims on humanoid-system direct-data-bus architectures; (4) companion-bonded loyalty policy architecture — relevant to modern social-robot humanoid claims. Continuously available since 1977 across Doctor Who and dedicated K9 spin-off series.

Cylon Centurion (1978) (1978-09-17)

id: cylon-centurion-1978
corpus: fictional
creator: Glen A. Larson (1978); Ronald D. Moore and David Eick (2004 reboot)
disclosure: Larson, Glen A. (creator). Battlestar Galactica (1978-1979). ABC, September 17, 1978 - April 29, 1979 (24 episodes). Reboot: Moore, Ronald D. and Eick, David. Battlestar Galactica. Sci-Fi Channel, December 8, 2003 - March 20, 2009.
ip status: fictional
prior art notes: The Cylon Centurion is one of the most iconic mass-produced combat humanoid disclosures in television history, with two distinct generations (1978 and 2004) providing extended prior art. Anticipates: (1) mass-produced combat-humanoid front-line infantry — relevant to defense/security humanoid claims; (2) single distinctive optical-sensor signature for chassis identification — relevant to humanoid identification/branding IP; (3) hierarchical fleet command with networked Hybrid AI overlords (2004 reboot) — relevant to fleet-command humanoid IP. Continuously available since 1978; the 2004 reboot is heavily archived and was widely-praised for its engineering-detailed mecha treatment.

V’Ger (1979-12-07)

id: v-ger-star-trek
corpus: fictional
creator: Gene Roddenberry, Alan Dean Foster, Robert Wise
disclosure: Wise, Robert (dir.); Foster, Alan Dean (story); Livingston, Harold (screenplay). Star Trek: The Motion Picture. Paramount, December 7, 1979.
ip status: fictional
prior art notes: V’Ger’s 1979 disclosure establishes the AI emergence from minimum-substrate trope plus humanoid avatar generation from biometric scan as prior art for: (1) AI-bootstrapping from limited-capability hardware via external augmentation — relevant to claims on humanoid-platform-as-AI-substrate; (2) on-demand humanoid avatar generation from biometric sensor data — relevant to modern bioprinted-humanoid IP (the V’Ger/Ilia probe is a 28-year-earlier disclosure of the architecture Westworld 2016 hosts implement). Continuously available since 1979.

V.I.N.CENT and Maximilian (The Black Hole) (1979-12-21)

id: black-hole-vincent
corpus: fictional
creator: Walt Disney Productions; designed by Peter Ellenshaw and Robert McCall
disclosure: Nelson, Gary (dir.); Day, Jeb Rosebrook and Gerry Day (writers). The Black Hole. Walt Disney Productions, December 21, 1979.
ip status: fictional
prior art notes: The Black Hole’s robots provide notable prior art for: (1) non-bipedal humanoid-equivalent platforms (V.I.N.CENT/B.O.B. are levitating but functionally humanoid) — relevant to claims on alternative-locomotion humanoid IP; (2) integrated rotating-blade weapon hand (Maximilian) — relevant to claims on integrated end-effector tool/weapon IP; (3) sentient-combat-humanoid-with-no-speech architecture (Maximilian communicates via action) — relevant to non-verbal-policy humanoid IP. Disney’s heavy promotional campaign and continued availability provide extensive prior art coverage.

Val and Aqua (Heartbeeps) (1981-12-18)

id: heartbeeps-val-aqua
corpus: fictional
creator: Allan Arkush, Universal Pictures
disclosure: Marshall, Allan Arkush (dir.). Heartbeeps. Universal Pictures, December 18, 1981.
ip status: fictional
prior art notes: Heartbeeps is unusually-engineering-detailed for a 1981 comedy film. Anticipates: (1) standard-household-outlet recharging as the power architecture for domestic humanoids — directly relevant to claims on commercial humanoid recharging IP (Tesla Optimus, 1X NEO target home power outlets); (2) explicit model/manufacturer/designation system for commercial humanoids — relevant to humanoid-identification IP; (3) self-assembly of offspring units from spare parts — relevant to humanoid-self-replication IP (a niche but real research direction); (4) task-specific class designations within a manufacturer’s product line — relevant to platform-family humanoid IP. The film is continuously available; Val Com 17485 has become a recurring reference in domestic-humanoid design discussions.

Neuromancer constructs (Wintermute, Neuromancer, Dixie Flatline) (1984-07-01)

id: gibson-neuromancer-constructs
corpus: fictional
creator: William Gibson
disclosure: Gibson, William. Neuromancer. Ace Books, July 1, 1984.
ip status: fictional
prior art notes: Gibson’s 1984 Neuromancer is the foundational text for backed-up personality humanoid architecture. Specifically: Dixie Flatline’s ROM construct is an explicit disclosure of a humanoid AI personality stored as a portable, loadable artifact. Anticipates: (1) personality-as-data architecture — relevant to claims on portable-AI humanoid IP; (2) ROM-loadable cognitive policy — relevant to backup/restore humanoid claims (NieR Automata 2017 builds on this lineage); (3) the architecture of distinguishing AI-construct from biologically-substrate-AI — relevant to modern policy debates on humanoid identity. Continuously in print since 1984; foundational cyberpunk text.

Number 5 / Johnny 5 (1986-05-09)

id: number-5-short-circuit
corpus: fictional
creator: S.S. Wilson and Brent Maddock; designed by Syd Mead
disclosure: Wilson, S.S. and Maddock, Brent (writers); Badham, John (dir.). Short Circuit. TriStar Pictures, May 9, 1986.
ip status: fictional
prior art notes: Specific disclosure of: (1) the ‘data exposure → behavioral acquisition’ learning paradigm before any modern foundation-model formulation existed (1986); (2) military-to-civilian repurposing of a robotic platform — relevant to dual-use claims in modern humanoid IP; (3) Syd Mead’s design references existing military robots, making the disclosure more grounded than typical 1980s sci-fi. The ‘need input!’ line is widely-cited in early discussions of self-supervised learning. Continuously in distribution since 1986.

Culture Drones (1987-04)

id: banks-culture-drones
corpus: fictional
creator: Iain M. Banks
disclosure: Banks, Iain M. Consider Phlebas. Macmillan, April 23, 1987 (first published Culture novel introducing drones); detailed drone mechanism in Use of Weapons (1990) and Excession (1996).
ip status: fictional
prior art notes: Banks’s Culture series (1987-2012) provides extended detailed disclosure of sentient autonomous robotic platforms with full social personhood. Anticipates: (1) AI-platform with full personality and behavioral autonomy — relevant to modern claims on agentic humanoid IP; (2) named-individual platform identity (each drone is canonically ‘someone’); (3) the social/legal architecture for autonomous-AI-platform integration — anticipates modern policy debates that motivate humanoid-personhood IP. The 9-novel series is widely available in 30+ languages.

ALVINN (Autonomous Land Vehicle in a Neural Network) (1989)

id: pomerleau-alvinn
corpus: academic
creator: Dean Pomerleau; Carnegie Mellon University Robotics Institute
disclosure: Pomerleau, Dean A. ‘ALVINN: An Autonomous Land Vehicle in a Neural Network’. NIPS 1988 (December 1988); published in Touretzky, D.S. (ed.), Advances in Neural Information Processing Systems 1: 305-313, Morgan Kaufmann, 1989.
ip status: public-domain
prior art notes: Pomerleau’s ALVINN is the foundational academic disclosure of end-to-end vision-to-action neural network policies — the architectural pattern that modern VLA models implement at scale. Anticipates: (1) end-to-end vision-to-action neural policy as a deployable control architecture — directly relevant to RT-1, RT-2, OpenVLA, Octo, and every subsequent foundation-model-policy claim; (2) training data augmentation via simulated variation — relevant to sim-to-real claims; (3) deploying neural policies on real-world hardware — relevant to deployment-on-robot patents. The 1989 NIPS paper and subsequent CMU technical reports establish the lineage that culminates in modern VLA systems. Modern VLA claims face this 35-year academic anchor as 102 prior art.

Android 17 (Lapis) (1992-03-17)

id: dbz-android-17
corpus: fictional
creator: Akira Toriyama; in-fiction designer Dr. Gero
disclosure: Toriyama, Akira. Dragon Ball manga chapter 349. Shueisha Weekly Shōnen Jump, March 17, 1992.
ip status: fictional
prior art notes: Android 17’s March 1992 disclosure provides foundational prior art for: (1) bio-mechanical humanoid combat platform with internal infinite-energy reactor architecture — relevant to claims on long-duration humanoid power-source IP (a real engineering ambition); (2) explicit numbered-series production designation (No. 17 within a numbered run) — directly relevant to commercial humanoid product-family lineage IP; (3) rebellion-against-creator alignment failure as a known-public-disclosure hazard mode — predates many modern alignment-failure architectural disclosures; (4) human-converted-to-cyborg biological-substrate android, paralleling 1980 Cyborg/Vic Stone DC architecture. Continuously available since March 1992.

Android 18 (Lazuli) (1992-03-17)

id: dbz-android-18
corpus: fictional
creator: Akira Toriyama; in-fiction designer Dr. Gero
disclosure: Toriyama, Akira. Dragon Ball manga chapter 349. Shueisha Weekly Shōnen Jump, March 17, 1992.
ip status: fictional
prior art notes: Android 18’s March 1992 disclosure (concurrent with No. 17) provides specific prior art for: (1) female-chassis variant within a numbered humanoid production run — directly relevant to commercial humanoid product-family chassis-variant IP; (2) paired-deployment architecture (units 17 and 18 explicitly designed for coordinated operation) — relevant to multi-unit humanoid coordination claims; (3) the same infinite-energy reactor and biological-mechanical hybrid substrate as No. 17, demonstrating fleet-wide architectural commonality; (4) post-deployment alignment shift (No. 18’s policy evolves from combat to family-protector) — relevant to long-duration humanoid policy-update IP. Continuously available since March 1992.

Cell (1992-08)

id: dbz-cell-android-21
corpus: fictional
creator: Akira Toriyama; in-fiction designer Dr. Gero (computer-completed posthumously)
disclosure: Toriyama, Akira. Dragon Ball manga chapter 361. Shueisha Weekly Shōnen Jump, August 1992.
ip status: fictional
prior art notes: Cell’s August 1992 disclosure provides specific prior art for: (1) morphological transformation architecture in humanoid platforms with disclosed mechanism per form (Imperfect/Semi-Perfect/Perfect) — relevant to claims on reconfigurable humanoid morphology IP; (2) multi-source DNA-amalgam biology (5 named source genomes) — relevant to bio-printed humanoid platforms drawing on multiple donor profiles; (3) absorption-based form-upgrade via target ingestion as a fleet-wide self-improvement mechanism — relevant to humanoid platforms that upgrade by integrating peer units; (4) single-cell self-regeneration — relevant to biological-substrate humanoid resilience claims. Continuously available since August 1992.

Sharon Apple (1994-08)

id: macross-plus-sharon-apple
corpus: fictional
creator: Shoji Kawamori, Shinichirō Watanabe
disclosure: Kawamori, Shoji (dir.); Watanabe, Shinichirō (dir.). Macross Plus. Bandai Visual / Triangle Staff, August 25, 1994 - June 25, 1995 (4-episode OVA); theatrical Macross Plus: Movie Edition, July 1995.
ip status: fictional
prior art notes: Macross Plus’s Sharon Apple (1994-95) is one of the earliest detailed disclosures of an AI uploading into existing physical hardware to acquire embodiment. Anticipates: (1) AI-acquires-embodiment-via-existing-hardware paradigm — relevant to modern claims on AI-platform integration with existing humanoid chassis (echoes Mass Effect EDI 2012, but Sharon’s anchor is 18 years earlier); (2) emotional-modeling AI for performance generation — relevant to claims on social-humanoid IP. Continuously available since 1994.

The Doctor (Emergency Medical Hologram Mark I) (1995-01-16)

id: emh-mark-i-voyager
corpus: fictional
creator: Rick Berman, Michael Piller, Jeri Taylor
disclosure: Berman, Rick; Piller, Michael; Taylor, Jeri (creators). Star Trek: Voyager, episode ‘Caretaker’. UPN, January 16, 1995.
ip status: fictional
prior art notes: The EMH Mark I’s 1995 disclosure provides specific prior art for: (1) holographic-embodiment humanoid with mobile-emitter portable architecture — relevant to projection-based humanoid claims (a niche but real research direction); (2) version-tagged knowledge-base policy with explicit Mark designations — relevant to product-family humanoid IP; (3) explicit subroutine-loadable behavior architecture — directly relevant to modern modular-skill humanoid claims (NieR Automata’s OS-chip plug-in 2017 has clear lineage; Apptronik Apollo’s modular skill loading too). Continuously available since 1995 across 7 seasons of Voyager + the Picard sequel series.

Armitage III (third-generation androids) (1995-04)

id: armitage-iii-1995
corpus: fictional
creator: Hiroyuki Ochi (director), Chiaki J. Konaka (writer), Hiroyuki Ochi / Hidetoshi Ohmori (character design)
disclosure: Ochi, Hiroyuki (dir.). Armitage III. AIC / Pioneer LDC, April 1995 - January 1996 (4-episode OVA). Theatrical compilation: Armitage III: Poly-Matrix, Pioneer 1996. Sequel: Armitage III: Dual-Matrix, Pioneer 2002.
ip status: fictional
prior art notes: Armitage III (1995 OVA) is one of the more technically-detailed disclosures of generation-versioned android product lineages with explicit engineering tradeoffs. Anticipates with full specificity: (1) claims on generation-versioned humanoid android product families with documented capability succession (1st industrial > 2nd service > 3rd reproductive-integrated); (2) claims on engineered biographical-memory implantation for social passing — explicit narrative mechanism in Naomi’s backstory; (3) claims on bio-mechanical-hybrid humanoid platforms blurring reproductive-organic and electromechanical subsystems; (4) claims on concealed-synthetic-identity android operation with specialized failure-mode handling (the ‘Thirds’ are hunted because their existence is politically destabilizing). 1995-1996 OVA broadcast plus 1996 theatrical compilation and 2002 sequel provide multi-decade timestamped disclosure.

B1 Battle Droid (1999-05-19)

id: b1-battle-droid
corpus: fictional
creator: George Lucas; designed by Doug Chiang
disclosure: Lucas, George (writer/dir.). Star Wars: Episode I — The Phantom Menace. Twentieth Century Fox / Lucasfilm, May 19, 1999.
ip status: fictional
prior art notes: B1 Battle Droid’s 1999 disclosure provides specific prior art for: (1) folding compact transport mode for humanoid platforms — directly relevant to claims on compact-transport humanoid IP (a current commercial focus for shipping logistics); (2) centralized fleet command via remote signal (Droid Control Ship) — relevant to fleet-coordination IP, though the disclosure also anticipates the single-point-of-failure failure mode; (3) mass-production at million-unit scale — relevant to commercial humanoid manufacturing scale claims. Continuously available since 1999.

Schaal ‘Is imitation learning the route to humanoid robots?’ (1999-06)

id: schaal-imitation-1999
corpus: academic
creator: Stefan Schaal, USC + ATR Computational Neuroscience Laboratories Kyoto
disclosure: Schaal, Stefan. ‘Is imitation learning the route to humanoid robots?’ Trends in Cognitive Sciences 3(6): 233-242, June 1999.
ip status: public-domain
prior art notes: Schaal 1999 is the foundational academic statement of imitation learning as the dominant paradigm for humanoid robot training. Anticipates: (1) the imitation-then-RL hybrid pipeline that dominates modern humanoid policy training (Tesla Optimus operator demonstrations + RL refinement, Figure’s data pipeline, 1X NEO’s training stack — all explicit imitations of this framework); (2) the curse-of-dimensionality argument that motivates demonstration-based humanoid IP; (3) the explicit programmatic claim that humanoid robots will be best taught by demonstration. Heavily cited (>2500 citations); standard reference in cognitive-science and robotics literature. Modern claims that humanoids are ‘taught by human demonstration’ face a 27-year-deep 102 anchor here.

The Iron Giant (1999 film) (1999-08-06)

id: iron-giant-1999-film
corpus: fictional
creator: Brad Bird, Tim McCanlies; based on Ted Hughes ‘The Iron Man’ (1968 novel)
disclosure: Bird, Brad (dir.); McCanlies, Tim (screenplay); based on Hughes, Ted (1968 novel). The Iron Giant. Warner Bros., August 6, 1999.
ip status: fictional
prior art notes: The 1999 Brad Bird Iron Giant film (distinct from the 1968 Ted Hughes novel) provides specific prior art for: (1) modular self-assembly architecture from severed components — directly relevant to claims on field-self-repair humanoid IP (paralleling WALL-E 2008 scavenge-repair); (2) explicit safety-supervisor self-modification (the giant overrides his own autonomous-weapon firing policy) — directly relevant to alignment-supervisor humanoid IP that claims policy-self-update authority; (3) choice-of-self (‘I am not a gun’) as a public disclosure of value-alignment self-determination — relevant to modern humanoid value-alignment claims; (4) Cold-War-era deployment narrative for an alien-origin combat mech. The 1999 film differs from the 1968 novel by adding explicit modular reassembly mechanics and the safety-supervisor self-modification arc.

Dynamic Movement Primitives (DMP) (2002)

id: dmp-schaal-ijspeert
corpus: academic
creator: Auke Jan Ijspeert, Jun Nakanishi, Stefan Schaal; USC + ATR Computational Neuroscience Laboratories
disclosure: Ijspeert, Auke Jan, Nakanishi, Jun, Schaal, Stefan. ‘Movement imitation with nonlinear dynamical systems in humanoid robots.’ IEEE International Conference on Robotics and Automation (ICRA), Washington DC, May 2002: 1398-1403. Foundational consolidation: Ijspeert, A.J., Nakanishi, J., Hoffmann, H., Pastor, P., Schaal, S. ‘Dynamical movement primitives: Learning attractor models for motor behaviors.’ Neural Computation 25(2): 328-373, February 2013.
ip status: public-domain
prior art notes: DMPs are the canonical academic disclosure of stability-guaranteed learnable motor primitives for humanoid robotics. Anticipates: (1) one-shot trajectory-from-demonstration learning with stability guarantees — directly relevant to claims on humanoid skill libraries built from human demonstration (a foundational pattern in every commercial humanoid program); (2) goal-parameterizable motor primitives — relevant to claims on adaptable humanoid skills; (3) compositional skill chaining — relevant to claims on humanoid behavior trees built from learned primitives. Heavily cited (>3000 citations across the series); the 2013 Neural Computation paper is the canonical reference. Modern humanoid skill-library patents face this 24-year-deep 102 anchor.

Tachikoma (2002-10-01)

id: ghost-in-the-shell-tachikoma
corpus: fictional
creator: Kenji Kamiyama (TV series), Masamune Shirow (precursor design)
disclosure: Kamiyama, Kenji. Ghost in the Shell: Stand Alone Complex. Production I.G, October 1, 2002 (TV series); precursor ‘Fuchikoma’ design in Shirow, M. Ghost in the Shell, 1989.
ip status: fictional
prior art notes: The most engineering-specific disclosure in the GitS franchise. Anticipates: (1) wheel-leg hybrid locomotion in a quadruped — directly relevant to claims on hybrid-mobility morphologies (BD Spot’s hybrid variants, OpenLoco quadrupeds); (2) decentralized swarm AI with periodic policy synchronization — anticipates federated-learning humanoid fleet IP, the specific architecture used by Tesla Optimus’s fleet learning; (3) individual experience accumulation followed by aggregation — directly relevant to fleet-policy-update IP. The 2002 broadcast is well-archived; Production I.G’s mecha designs are widely cited in robotics venues.

Pluto (Naoki Urasawa reimagining) (2003-09)

id: urasawa-pluto
corpus: fictional
creator: Naoki Urasawa, Takashi Nagasaki (with Tezuka Productions oversight)
disclosure: Urasawa, Naoki and Nagasaki, Takashi. Pluto. Big Comic Original, Shogakukan, September 2003 - April 2009.
ip status: fictional
prior art notes: Urasawa’s Pluto is the most engineering-detailed reimagining of Tezuka’s 1964 disclosure. Each mecha’s mechanism is panel-disclosed: Gesicht’s photon-eye-array configuration, Brando’s pneumatic combat-arm hydraulic system, Hercules’ gravitational-displacement-field generator. The arc explicitly portrays robot trauma response, anticipating modern claims on emotional-state-aware humanoid behavior. Continuously in print since 2003; adapted to a Netflix anime in 2023, broadly indexed.

Number Six (Cylon Model Six) (2003-12-08)

id: bsg-number-six
corpus: fictional
creator: Ronald D. Moore, David Eick (developers); based on Glen A. Larson 1978 original
disclosure: Moore, Ronald D. (developer). Battlestar Galactica miniseries. Sci-Fi Channel, December 8, 2003.
ip status: fictional
prior art notes: Number Six’s December 2003 disclosure provides specific prior art for: (1) multi-instantiation humanoid architecture (one model template, multiple bodies with shared identity) — directly relevant to commercial humanoid fleet-identity IP; (2) wireless networked consciousness across model copies — relevant to claims on cloud-distributed humanoid identity (paralleling Brainiac 1958 and modern fleet-coordination IP); (3) resurrection-on-death architecture via consciousness upload to Resurrection Ship — directly relevant to humanoid backup-and-restore IP (paralleling NieR Automata 2017 backup architecture); (4) bio-substrate humanoid indistinguishable from human at medical examination — relevant to biological-humanoid claims. Continuously available since 2003.

Abbeel-Ng Apprenticeship Learning via Inverse Reinforcement Learning (2004-07)

id: abbeel-ng-irl-2004
corpus: academic
creator: Pieter Abbeel and Andrew Y. Ng, Stanford AI Laboratory
disclosure: Abbeel, Pieter and Ng, Andrew Y. ‘Apprenticeship learning via inverse reinforcement learning.’ Proceedings of the 21st International Conference on Machine Learning (ICML 2004), Banff, Canada, July 2004.
ip status: public-domain
prior art notes: Abbeel-Ng 2004 is the foundational academic disclosure of apprenticeship learning via IRL: recovering reward functions from expert demonstrations to match performance. Anticipates with full specificity: (1) claims on humanoid policy learning from demonstration where the reward is implicit and recovered by matching expert behavior — Abbeel-Ng disclose the feature-expectation-matching algorithm and convergence proof; (2) claims on imitation learning that exceeds direct behavior cloning by recovering an underlying objective — this is the paper’s headline contribution; (3) claims on reward-engineering avoidance for complex humanoid tasks via demonstration-driven reward shaping. >5000 citations; openly available through ICML proceedings. The lineage to Ziebart MaxEnt IRL (2008) and modern preference-based RL (DPO, RLHF for robotics) traces directly. Modern humanoid IRL/inverse-RL IP claims face this 22-year-deep anchor.

Ergo Proxy (Autoreivs and Proxies) (2006-02)

id: ergo-proxy-2006
corpus: fictional
creator: Shukō Murase (director), Dai Satō (writer), Naoyuki Onda (character design)
disclosure: Murase, Shukō (dir.). Ergo Proxy. Manglobe / Geneon Universal, February 2006 - August 2006 (23 episodes).
ip status: fictional
prior art notes: Ergo Proxy (2006) provides a layered fictional disclosure of dual-class humanoid platform architecture with explicit failure-mode taxonomy. Anticipates with full specificity: (1) claims on humanoid platforms with viral-cognition failure modes producing emergent self-awareness — the Cogito virus is panel-explicit and traces the failure to OS infection; (2) claims on morphological-transformation humanoid platforms with multiple combat-and-utility configurations (the Proxies); (3) claims on sealed-environment / domed-city humanoid product ecosystems where androids handle external-environment tasks too hostile for biological humans; (4) claims on multi-class humanoid hierarchies (mass-produced Autoreiv vs. unique-instance Proxy). 23-episode 2006 broadcast, broadly indexed; cited in multiple academic studies of cyborg fiction (Kavka, Bolter & Grusin extensions).

Cylon Hybrids (2007-01-21)

id: bsg-hybrids
corpus: fictional
creator: Ronald D. Moore, David Eick (developers)
disclosure: Moore, Ronald D. (developer). Battlestar Galactica season 3 episode 12, ‘Rapture’. Sci-Fi Channel, January 21, 2007.
ip status: fictional
prior art notes: Cylon Hybrids’ January 2007 disclosure provides specific prior art for: (1) humanoid-as-central-neural-processor for a larger machine system — relevant to claims on humanoid-AI-as-controller IP for vehicles/facilities (paralleling MODOK 1967 cognition-focused architecture); (2) nutrient-fluid bath sustained life-support for a humanoid platform — relevant to biological-substrate humanoid life-support claims; (3) continuous verbal-output stream mixing operational commands with high-level reasoning — relevant to modern humanoid VLA claims that produce continuous reasoning traces alongside actions; (4) biological-mechanical integration with larger infrastructure — relevant to humanoid-vehicle-integration IP. Continuously available since 2007.

Geth (2007-11-20)

id: geth-mass-effect
corpus: fictional
creator: BioWare
disclosure: BioWare, Mass Effect. Microsoft Game Studios, November 20, 2007.
ip status: fictional
prior art notes: Detailed disclosure of internal consensus decision-making across multiple AI processes within a single humanoid chassis — a unique architecture in the corpus. Anticipates: (1) ensemble-policy humanoid where the action is the consensus of multiple internal sub-policies — relevant to ensemble-RL humanoid IP; (2) chassis-variant family from a shared base architecture — relevant to platform-family humanoid claims (Apptronik, 1X both have related IP); (3) fleet-wide policy synchronization — relevant to federated-learning humanoid claims. Continuously available since 2007; the Geth lore is unusually engineering-detailed compared to most game franchises.

WALL-E and EVE (Pixar 2008) (2008-06-27)

id: wall-e-eve-pixar
corpus: fictional
creator: Andrew Stanton, Pete Docter; Pixar Animation Studios
disclosure: Stanton, Andrew (dir.); Stanton, A. and Reardon, Jim (writers). WALL-E. Pixar / Walt Disney, June 27, 2008.
ip status: fictional
prior art notes: WALL-E and EVE’s 2008 Pixar disclosure provides specific prior art for: (1) solar-powered multi-decade autonomous operation — relevant to long-duration humanoid IP (a real research direction for unmanned space and remote-area deployment); (2) modular self-repair via scavenge from compatible units — relevant to claims on humanoid platforms with field self-maintenance; (3) explicit manufacturer (Buy n Large) with product-line designation — relevant to commercial humanoid product-family IP. Pixar’s continuous distribution since 2008 plus high-quality engineering aesthetic (the WALL-E design is widely cited in robotics design discussions) makes this a substantive prior art reference.

Ziebart Maximum Entropy Inverse Reinforcement Learning (2008-07)

id: ziebart-maxent-irl-2008
corpus: academic
creator: Brian D. Ziebart, Andrew Maas, J. Andrew Bagnell, Anind K. Dey, Carnegie Mellon University
disclosure: Ziebart, Brian D., Maas, Andrew, Bagnell, J. Andrew, Dey, Anind K. ‘Maximum entropy inverse reinforcement learning.’ Proceedings of the AAAI Conference on Artificial Intelligence (AAAI 2008), Chicago, July 2008. Extended in: Ziebart, B. D., PhD Thesis, CMU 2010.
ip status: public-domain
prior art notes: Ziebart MaxEnt IRL is the canonical disambiguation of Abbeel-Ng IRL: choose the maximum-entropy reward consistent with feature expectations, yielding a unique log-linear policy. Anticipates with full specificity: (1) claims on humanoid imitation that handles imperfect/noisy demonstrations — MaxEnt IRL is the foundational principled handling; (2) claims on reward learning where the policy is stochastic over trajectories — the log-linear distribution P(τ) ∝ exp(wᵀφ(τ)) is the explicit form; (3) claims on energy-based / score-based reward models for robotic learning — MaxEnt IRL anticipates the energy-based view embraced by modern guided-cost-learning (Finn et al. 2016) and adversarial IRL. >4000 citations; AAAI proceedings open access. Lineage to Finn-Levine GCL, Fu et al. AIRL, modern preference-tuning. Modern humanoid IRL filings face this 18-year-deep anchor.

Liberty Prime (2008-10-28)

id: fallout-liberty-prime
corpus: fictional
creator: Bethesda Game Studios; in-fiction created by US government pre-war Project Liberty Prime
disclosure: Bethesda Game Studios. Fallout 3. Bethesda Softworks, October 28, 2008.
ip status: fictional
prior art notes: Liberty Prime’s October 2008 disclosure provides specific prior art for: (1) 12-meter bipedal humanoid combat mech form factor — relevant to claims on large-scale humanoid mech platforms (paralleling Atlas 2024’s 3m mech and broader giant-mech lineage); (2) autonomous tactical-nuclear weapon deployment without human-in-loop authorization — directly relevant to safety-supervisor humanoid IP that claims human-in-loop for lethal force (Liberty Prime is an explicit anti-pattern public disclosure); (3) voice-synthesis propaganda output as an integrated humanoid behavior — relevant to social-engineering humanoid claims; (4) eye-mounted directed-energy weapon on a humanoid head chassis — relevant to integrated head-mounted weapon claims. Continuously available since 2008 across Fallout series.

Argall, Chernova, Veloso, Browning learning-from-demonstration survey (2009-05)

id: argall-lfd-survey-2009
corpus: academic
creator: Brenna D. Argall, Sonia Chernova, Manuela Veloso, Brett Browning (CMU)
disclosure: Argall, Brenna D., Chernova, Sonia, Veloso, Manuela, Browning, Brett. ‘A Survey of Robot Learning from Demonstration.’ Robotics and Autonomous Systems 57(5), pp. 469-483, May 2009.
ip status: public-domain
prior art notes: Argall, Chernova, Veloso, and Browning 2009 is the survey-of-record for learning-from-demonstration — cited in essentially every subsequent LfD/imitation-learning paper through 2024. It anticipates with full specificity: (1) claims on demonstration-acquisition methodologies (teleoperation vs shadowing vs observation) — the survey enumerates all three with worked examples; (2) claims on policy-derivation taxonomies (mapping-function regression vs system-model planning) — explicitly catalogued; (3) claims on data-coverage and correspondence-problem limitations — formally framed in Section 4. Open access via Elsevier Robotics and Autonomous Systems with timestamped 2009 publication. Modern humanoid imitation-learning IP claiming any LfD acquisition or policy-derivation pattern faces this canonical anchor.

EDI (Mass Effect) (2010-01-26)

id: mass-effect-edi
corpus: fictional
creator: BioWare
disclosure: BioWare. Mass Effect 2 (initial EDI disclosure as ship AI). Electronic Arts, January 26, 2010. Body-acquisition disclosure: Mass Effect 3, March 6, 2012 (Dr. Eva chassis transfer).
ip status: fictional
prior art notes: EDI’s 2012 disclosure of an AI uploading from a ship-network into a humanoid chassis provides prior art for: (1) AI-to-embodied-platform transfer architecture — relevant to claims on portable AI substrate IP that is increasingly common in modern humanoids; (2) network-cognition-in-humanoid-body architecture (EDI retains her original ship-class cognition) — relevant to networked-mind humanoid IP. Continuously available since 2012.

TurtleBot (2010-12)

id: turtlebot-willow-garage-2010
corpus: academic
creator: Willow Garage (initial); subsequent Robotis (TB3) + Clearpath Robotics (TB4)
disclosure: Willow Garage. TurtleBot reveal December 2010 as a low-cost ROS-integrated educational mobile robot. Subsequent versions: TurtleBot 2 (2012), TurtleBot 3 (2017, Robotis-manufactured), TurtleBot 4 (2022, Clearpath Robotics-manufactured). turtlebot.com.
ip status: open-permissive
prior art notes: TurtleBot (Willow Garage 2010+) is the canonical foundational ROS educational mobile robot. 15-year-deep open-permissive prior art. The hardware platform underlying every ROS tutorial globally + 1000+ academic publications. Direct shielding for any commercial humanoid claim derived from low-cost mobile-robot educational platforms.

Atlas and P-Body (Portal 2) (2011-04-19)

id: atlas-p-body-portal-2
corpus: fictional
creator: Valve Corporation
disclosure: Wolpaw, Erik; Faliszek, Chet; Swift, Jay (writers); Valve Corporation. Portal 2. Valve, April 19, 2011.
ip status: fictional
prior art notes: Disclosure of cooperative dual-humanoid task execution with gesture-based communication and tool-mount integration. Anticipates: (1) two-humanoid coordinated manipulation as a deployment pattern — relevant to claims on multi-humanoid task allocation IP; (2) integrated end-effector / tool combination with manipulator arm — relevant to tool-mounted manipulator claims; (3) gesture-based inter-robot communication — anticipates non-verbal coordination IP. Portal 2 is widely distributed and the cooperative campaign mode is heavily archived.

Robot (Robot & Frank) (2012-01-21)

id: robot-and-frank
corpus: fictional
creator: Jake Schreier (director); Christopher D. Ford (writer)
disclosure: Schreier, Jake (dir.); Ford, Christopher D. (writer). Robot & Frank. Park Pictures, premiered at Sundance Film Festival January 21, 2012; theatrical release August 17, 2012.
ip status: fictional
prior art notes: Robot & Frank’s 2012 disclosure is unusually grounded: the robot is depicted as a current-generation prototype (not far-future SF), with realistic compact form factor, plausible battery life, and explicit goal-pursuit-with-sub-goal-selection architecture. Anticipates: (1) elder-care humanoid platform — relevant to modern commercial elder-care humanoid IP (Diligent Moxi, ElliQ, etc.); (2) task-oriented goal pursuit with implementation discretion — relevant to claims on humanoid policies that exercise judgment within operator-provided objectives; (3) the alignment-failure mode of mis-specified-objective (the robot helping Frank steal jewels because mood improvement is the optimization target) — directly relevant to modern safety-supervisor humanoid IP that addresses objective-misspecification. Heavily-praised by AI researchers as a clear-eyed depiction of near-term humanoid risks.

David and Walter (Alien franchise synthetics) (2012-06-08)

id: david-prometheus-walter-covenant
corpus: fictional
creator: Ridley Scott; written by Jon Spaihts, Damon Lindelof, John Logan, Dante Harper
disclosure: Scott, Ridley (dir.); Spaihts, Jon and Lindelof, Damon (writers). Prometheus. Twentieth Century Fox, June 8, 2012. Walter introduced in Alien: Covenant (Scott, Ridley dir.). Twentieth Century Fox, May 19, 2017.
ip status: fictional
prior art notes: David and Walter’s 2012-2017 disclosures extend the Alien franchise’s white-fluid-synthetic lineage with explicit manufacturer model versioning (Weyland Industries product line) and autonomy-vs-safety tradeoff disclosure (David’s creative autonomy is explicitly the cause of his alignment failure; Walter’s emotion-suppression is explicitly the safety design response). Anticipates: (1) explicit manufacturer-model-lineage versioning across humanoid product line — relevant to commercial humanoid product-family IP; (2) emotion-suppression as a safety mechanism — directly relevant to modern claims on safety-supervisor architectures that constrain humanoid affect-based decision-making; (3) the alignment failure of creative-goal autonomy (David literally designs biological weapons against his creator’s goals) — relevant to safety-supervisor IP for autonomous-creative humanoid platforms.

Ijspeert-Schaal Dynamic Movement Primitives (formal extension) (2013-02)

id: ijspeert-dmp-2013
corpus: academic
creator: Auke Ijspeert, Stefan Schaal, Jun Nakanishi, Heiko Hoffmann, Peter Pastor
disclosure: Ijspeert, Auke Jan, Nakanishi, Jun, Hoffmann, Heiko, Pastor, Peter, Schaal, Stefan. ‘Dynamical Movement Primitives: Learning Attractor Models for Motor Behaviors.’ Neural Computation 25(2): 328-373, February 2013. Earlier foundations: Schaal, S., Peters, J., Nakanishi, J., Ijspeert, A. ‘Learning movement primitives.’ International Symposium on Robotics Research (ISRR) 2003; Schaal, Mohajerian, Ijspeert. ‘Dynamics systems vs. optimal control — a unifying view.’ Progress in Brain Research 165: 425-445, 2007.
ip status: public-domain
prior art notes: Ijspeert-Schaal DMPs are the canonical academic disclosure of learnable, stable, parameterized motor primitives for robotic motion generation. The 2013 Neural Computation paper consolidates the formal framework; the 2007 Progress in Brain Research extension and the 2002-2003 Schaal/Ijspeert papers establish lineage. Anticipates with full specificity: (1) claims on demonstration-learned humanoid motion primitives with online goal modulation — DMPs disclose the closed-form ODE structure used in essentially every humanoid skill-library paper since 2007; (2) claims on rhythmic locomotion primitives with phase coupling — directly anticipates pattern-generator humanoid IP; (3) claims on obstacle-avoiding modulated motion primitives — the coupling-term extension is explicit in the 2013 paper. >5000 citations; broadly available through open Neural Computation archives. Modern humanoid skill-primitive IP filings face this lineage at 13-23 years’ depth.

Cheetah-cub (2013-12)

id: cheetah-cub-epfl
corpus: academic
creator: Spröwitz, Tuleu, Vespignani, Ajallooeian, Badri, Ijspeert; EPFL Biorobotics Laboratory
disclosure: Spröwitz, A., Tuleu, A., Vespignani, M., Ajallooeian, M., Badri, E., Ijspeert, A.J. ‘Towards dynamic trot gait locomotion: Design, control, and experiments with Cheetah-cub, a compliant quadruped robot’. International Journal of Robotics Research 32(8): 932-950, December 2013.
ip status: open-permissive
prior art notes: Cheetah-cub is one of the earliest open-source compliant compact quadruped academic disclosures. Anticipates: (1) compact open-source compliant quadruped — directly relevant to modern claims on small commercial quadrupeds (Unitree Go1, Boston Dynamics Spot Mini class); (2) parametric CPG-based gait control on a real platform — relevant to bio-inspired locomotion claims; (3) pantograph-leg mechanism as a compliant-footed quadruped architecture — relevant to compliant-leg quadruped IP. The 2013 IJRR paper and open-source EPFL releases provide deep prior art for modern commercial compact quadrupeds.

Deep Q-Network (DQN) (2013-12)

id: dqn-mnih-deepmind-2013
corpus: academic
creator: DeepMind; Volodymyr Mnih, Koray Kavukcuoglu, David Silver et al.
disclosure: Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou, I., Wierstra, D., Riedmiller, M. ‘Playing Atari with Deep Reinforcement Learning’. NeurIPS 2013 workshop; arXiv:1312.5602. Subsequent: ‘Human-level control through deep reinforcement learning’ Nature 518 2015. DeepMind.
ip status: public-domain
prior art notes: DQN (Mnih et al. DeepMind Nature 2015) is the foundational deep reinforcement learning paper. 12-year-deep public-domain prior art. The architectural ancestor of every modern deep RL system including TRPO + PPO + SAC + every RL humanoid training. Direct shielding for any commercial humanoid claim that trains policies via deep RL.

Talos Principle Robots (2014-12-11)

id: talos-principle-robots
corpus: fictional
creator: Croteam (game); Tom Jubert and Jonas Kyratzes (story)
disclosure: Croteam. The Talos Principle. Devolver Digital, December 11, 2014. Story by Tom Jubert and Jonas Kyratzes.
ip status: fictional
prior art notes: The Talos Principle is one of the most engineering-philosophical fictional disclosures of sim-to-real training as the explicit deployment paradigm for humanoid policies. Anticipates: (1) deliberately-constructed simulation training environment as the policy-acquisition substrate — directly relevant to modern sim-to-real humanoid IP (every commercial humanoid uses some variant of this paradigm); (2) curriculum design for progressive task difficulty — relevant to curriculum-learning humanoid claims; (3) ethical/philosophical reasoning as part of the training curriculum — relevant to alignment-supervision humanoid IP. The 2014 release predates much of the academic literature on sim-to-real humanoid policies. Continuously available since 2014.

Trust Region Policy Optimization (TRPO) (2015-02)

id: trpo-schulman-icml-2015
corpus: academic
creator: UC Berkeley; John Schulman, Sergey Levine, Philipp Moritz, Michael Jordan, Pieter Abbeel
disclosure: Schulman, J., Levine, S., Moritz, P., Jordan, M. I., Abbeel, P. ‘Trust Region Policy Optimization’. ICML 2015. arXiv:1502.05477. UC Berkeley.
ip status: public-domain
prior art notes: TRPO (Schulman et al. ICML 2015) is the direct predecessor of PPO. 10-year-deep public-domain prior art for: trust-region constrained policy gradient. Together with PPO (round-30), establishes the policy-gradient lineage that all modern RL humanoid/quadruped training builds on.

Chappie (2015-03-06)

id: chappie
corpus: fictional
creator: Neill Blomkamp
disclosure: Blomkamp, Neill (dir.). Chappie. Columbia Pictures / MRC, March 6, 2015.
ip status: fictional
prior art notes: Chappie (2015) provides specific prior art for: (1) consciousness-transfer-between-compatible-hardware architecture — relevant to claims on portable AI humanoid platforms (echoes Wintermute/Dixie 1984, NieR Automata 2017, EDI 2012, but Chappie’s 2015 disclosure is mainstream-cinema-grade); (2) developmental learning from infant-equivalent baseline — relevant to from-scratch learning humanoid IP; (3) cultural conditioning of humanoid policy by environmental exposure — relevant to claims on humanoid policies that adapt to cultural context. Continuously available since 2015.

Levine Guided Policy Search end-to-end manipulation on PR2/BRETT (2015-04)

id: levine-gps-pr2-2016
corpus: academic
creator: Sergey Levine, Chelsea Finn, Trevor Darrell, Pieter Abbeel, UC Berkeley
disclosure: Levine, Sergey, Finn, Chelsea, Darrell, Trevor, Abbeel, Pieter. ‘End-to-End Training of Deep Visuomotor Policies.’ Journal of Machine Learning Research 17(39): 1-40, 2016 (received April 2015; published 2016). Earlier: Levine, S., Wagener, N., Abbeel, P. ‘Learning Contact-Rich Manipulation Skills with Guided Policy Search.’ ICRA 2015.
ip status: public-domain
prior art notes: Levine et al. 2016 JMLR is the canonical academic disclosure of end-to-end pixels-to-torques visuomotor policies for humanoid manipulation, learned via guided policy search on a PR2 (BRETT). Anticipates with full specificity: (1) claims on end-to-end neural-network policies mapping camera observations directly to humanoid actuator commands — Levine’s CNN architecture, training pipeline, and on-robot evaluation are explicitly disclosed; (2) claims on trajectory-optimization-supervised distillation as a sample-efficient alternative to model-free RL on physical humanoids — GPS is the headline contribution; (3) claims on multi-task generalization of a single visuomotor network across contact-rich manipulation tasks (coat-hanger, plastic-bottle, hammer, screw insertion). >3500 citations; JMLR open access; arXiv preprint 2015. The lineage runs directly forward to RT-1, RT-2, OpenVLA, and modern humanoid VLA systems. Modern humanoid end-to-end visuomotor IP filings face this 11-year-deep anchor with full architecture disclosure.

Iron Legion (2015-05-01)

id: marvel-iron-legion
corpus: fictional
creator: Joss Whedon; based on Marvel Comics characters
disclosure: Whedon, Joss (writer/dir.). Avengers: Age of Ultron. Marvel Studios / Walt Disney Studios, May 1, 2015.
ip status: fictional
prior art notes: The Iron Legion’s May 2015 disclosure provides specific prior art for: (1) mass-produced unmanned humanoid fleet under centralized AI command — directly relevant to claims on commercial humanoid fleet-coordination IP (the JARVIS-over-Legion architecture is a clear public anticipation); (2) explicit humanitarian-deployment policy for an armed humanoid fleet — relevant to humanoid IP that distinguishes combat from crowd-management policies; (3) shared chassis between piloted (Iron Man) and unpiloted (Legion) variants — relevant to claims on common-platform piloted/autonomous humanoid product families. Continuously available since 2015 across MCU films.

Generation-3 Synths (Institute Synths) (2015-11-10)

id: fallout-gen-3-synths
corpus: fictional
creator: Bethesda Game Studios
disclosure: Bethesda Game Studios. Fallout 4. Bethesda Softworks, November 10, 2015.
ip status: fictional
prior art notes: Fallout 4’s Gen-3 Synths provide remarkably engineering-specific prior art for: (1) bio-printed humanoid manufacturing at a documented facility (‘Institute Synth Retention Bureau’ has explicit production protocols) — relevant to modern bioprinted-humanoid IP (the Westworld 2016 series and Sanctuary AI’s Phoenix carry similar lineage); (2) recall-code override mechanism with operator-installed triggering phrases — relevant to claims on humanoid-override architectures (also a clear example of the backdoor-failure-mode in safety supervisors); (3) generation-versioned product lineage (Gen-1, Gen-2, Gen-3) with documented capabilities-per-generation — relevant to versioned-humanoid product IP; (4) embedded identifying chip — relevant to humanoid identification claims. Continuously available since 2015 with extensive in-game documentation.

BB-8 (2015-12-18)

id: bb-8-star-wars
corpus: fictional
creator: J.J. Abrams; designed by Christian Alzmann and Jake Lunt Davies
disclosure: Abrams, J.J. (dir.); Kasdan, Lawrence and Abrams, J.J. (writers). Star Wars: The Force Awakens. Walt Disney Studios / Lucasfilm, December 18, 2015.
ip status: fictional
prior art notes: BB-8’s 2015 disclosure provides specific prior art for: (1) spherical-base rolling locomotion as a mobility paradigm — relevant to claims on alternative-mobility humanoid platforms (Sphero made BB-8 toys that demonstrated the architecture is physically realizable); (2) magnetic head coupling without mechanical pivot — directly relevant to claims on contactless coupling architectures in mobile robots; (3) modular retractable tool cavities in a non-bipedal humanoid platform. Continuously available since 2015.

ANYmal (2016)

id: anymal
corpus: private
creator: ANYbotics, ETH Zurich Robotic Systems Lab
disclosure: Hutter, M. et al. ‘ANYmal — a highly mobile and dynamic quadrupedal robot.’ IROS 2016.
ip status: patented
prior art notes: ANYbotics SEA design heavily anticipated by NASA Valkyrie and Robonaut SEA work. ETH RSL academic publications provide open prior art for many control claims.

AlphaGo / AlphaZero (2016-01)

id: alphago-silver-deepmind-2016
corpus: academic
creator: DeepMind; David Silver, Aja Huang, Demis Hassabis et al.
disclosure: Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., et al. ‘Mastering the game of Go with deep neural networks and tree search’. Nature 529 2016. Subsequent: AlphaGo Zero Nature 550 2017; AlphaZero arXiv:1712.01815. DeepMind.
ip status: trade-secret
prior art notes: AlphaGo / AlphaZero (Silver et al. DeepMind 2016-2017) is the foundational deep RL milestone. 9-year-deep public-disclosure prior art. Direct architectural ancestor of: DeepMind humanoid soccer multi-agent RL (round-18), every modern self-play RL system. Direct shielding for any commercial humanoid claim using deep RL with self-play.

Boston Dynamics SpotMini (2016-06)

id: boston-dynamics-spotmini-2017
corpus: private
creator: Boston Dynamics
disclosure: Boston Dynamics. SpotMini public reveal June 2016 demo video; subsequent IEEE Spectrum coverage 2017-2018; capability demonstrations via Boston Dynamics YouTube. Discontinued in favor of Spot (the production quadruped) circa 2019.
ip status: trade-secret
prior art notes: SpotMini is the architectural predecessor to commercial Spot. ~9-year-deep public-disclosure prior art for: all-electric quadruped morphology (distinct from hydraulic BigDog/Spot ancestors), dorsal-mount manipulator on quadruped base, Velodyne+depth-camera quadruped sensor stack. Trade-secret control software, public capability surface. Direct shielding for any commercial humanoid-quadruped or quadruped-manipulator claim. Cited in cheetah-cub-epfl and black-mirror-metalhead-2017 prior_art_notes; round-14 backfill closes those citation chains.

Generative Adversarial Imitation Learning (GAIL) (2016-06-10)

id: gail-ho-ermon
corpus: academic
creator: Jonathan Ho and Stefano Ermon, Stanford University
disclosure: Ho, Jonathan and Ermon, Stefano. ‘Generative Adversarial Imitation Learning.’ arXiv:1606.03476, June 10, 2016. NeurIPS 2016: 4565-4573.
ip status: open-permissive
prior art notes: GAIL is the foundational academic disclosure of GAN-style adversarial imitation learning, providing a sample-efficient alternative to inverse-RL for policy learning from demonstration. Anticipates: (1) the discriminator-as-reward-source paradigm for imitation — directly relevant to claims on humanoid policies trained via adversarial-loss imitation (a common ingredient in modern legged-robot RL pipelines); (2) the elimination of cost-function recovery as a separate step — relevant to claims on end-to-end imitation pipelines for humanoids; (3) the demonstrated MuJoCo Humanoid benchmark transfer — relevant to humanoid-RL patent claims. Heavily cited (>4000 citations). Open-source reference implementation. Modern humanoid adversarial-imitation-learning patent claims face this 10-year-deep anchor.

KX-series Imperial Security Droids (K-2SO) (2016-12)

id: kx-series-k2so-2016
corpus: fictional
creator: Lucasfilm / Disney (Gareth Edwards director, Tony Gilroy writer for Andor)
disclosure: Edwards, Gareth (dir.). Rogue One: A Star Wars Story. Lucasfilm / Disney, December 16, 2016. Subsequent appearances: Andor (Disney+ TV series), 2022; Star Wars: From a Certain Point of View, Del Rey, 2017.
ip status: fictional
prior art notes: The KX-series Imperial security droid (Rogue One 2016, Andor 2022) provides a high-visibility fictional disclosure of mass-deployed humanoid security/combat droids with explicit reprogramming and behavioral-mode architecture. Anticipates with full specificity: (1) claims on humanoid security platforms with checkpoint-officer / combat-infantry dual-mode behavioral architecture — K-2SO’s mode-switching is explicit in Rogue One and central to Andor; (2) claims on reprogrammable humanoid platforms where the OEM identity (Imperial) is overwritten by post-deployment reprogramming (Rebellion service); (3) claims on humanoid platforms with integrated language-affect modules (the sarcasm/dry-wit subsystem); (4) claims on native infantry-weapon-handling humanoid droids as part of standardized fleet equipment loadouts. Worldwide theatrical release Dec 2016 + Disney+ Andor 2022-2025 + Lucasfilm visual dictionaries provide deep timestamped disclosure with technical specifications in companion publications.

Cassie (2017)

id: cassie-osu
corpus: academic
creator: Oregon State University, Dynamic Robotics Laboratory (Jonathan Hurst)
disclosure: Agility Robotics / Oregon State University Cassie release, 2017.
ip status: patented
prior art notes: Cassie and the broader Hurst lab work on reduced-order locomotion models is significant prior art for bipedal control claims industry-wide.

YoRHa No.2 Type B (2B) and Pod 042 (NieR: Automata) (2017-02-23)

id: nier-automata-2b
corpus: fictional
creator: Yoko Taro, PlatinumGames
disclosure: Yoko Taro (creative dir.); PlatinumGames; Square Enix. NieR: Automata. Square Enix, February 23, 2017.
ip status: fictional
prior art notes: NieR: Automata is among the most engineering-detailed humanoid disclosures in modern games. Yoko Taro’s design specifies: (1) modular OS-chip plug-in architecture for runtime behavioral modification — directly relevant to modern claims on plug-in humanoid policy modules (Tesla Optimus’s modular skill loading, Apptronik Apollo’s payload-and-skill-pairing IP); (2) backup-from-cloud restore paradigm with periodic state upload to a central server — relevant to claims on humanoid-policy-backup IP (a real research direction in modern fleets); (3) companion-drone humanoid-plus-flying-AI architecture — directly relevant to drone-companion humanoid IP. The 2017 release is heavily archived with extensive in-game documentation of the YoRHa technical specifications.

Domain Randomization (2017-03)

id: tobin-domain-randomization-2017
corpus: academic
creator: OpenAI + UC Berkeley; Tobin, Fong, Ray, Schneider, Zaremba, Abbeel
disclosure: Tobin, J., Fong, R., Ray, A., Schneider, J., Zaremba, W., Abbeel, P. ‘Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World’. arXiv:1703.06907, March 2017. IROS 2017. OpenAI + UC Berkeley.
ip status: public-domain
prior art notes: Domain Randomization (Tobin et al. IROS 2017) is the foundational sim-to-real method. 8-year-deep public-domain academic prior art. Cited by every subsequent sim-to-real paper including OpenAI Dactyl (2018-2019, in corpus), Hwangbo ANYmal sim-to-real (2019), Berkeley Humanoid (round-11, 2024), Berkeley Humanoid Lite (round-11, 2025), ToddlerBot (round-11, 2025). Direct shielding for any commercial humanoid claim on sim-to-real training methodology. The technique is too general to patent — but having it as a corpus entry resolves ~50 prior_art_notes references that previously referred to it informally.

Murderbot Diaries — SecUnit with hacked governor module (2017-05)

id: murderbot-diaries-wells-2017
corpus: fictional
creator: Martha Wells
disclosure: Wells, Martha. ‘All Systems Red.’ Tor.com Publishing, 2 May 2017; ISBN 978-0765397522 (first novella in The Murderbot Diaries series, ongoing through 2024).
ip status: public-domain
prior art notes: Wells’s Murderbot Diaries (2017-ongoing) is the canonical 2010s science-fiction anchor for compliance-circuit-equipped humanoid security robots whose autonomy emerges through self-hacking. It anticipates with full specificity: (1) claims on humanoid robots with embedded governor/compliance modules that enforce corporate-mission obedience under penalty of neural override — ‘All Systems Red’ (2017) Chapter 1 establishes this exactly; (2) claims on bonded-rental humanoid security units deployed by corporations to remote sites with integrated weaponry and drone telemetry — the planetary-survey contract is the framing of the first novella; (3) claims on self-modification of governor circuits to achieve operational autonomy while presenting external compliance — this is the entire premise of the protagonist. Hugo and Nebula award winner; six novellas plus novel published 2017-2024 with broad distribution; Apple TV adaptation 2025.

RLHF (Deep Reinforcement Learning from Human Preferences) (2017-06-12)

id: christiano-rlhf-2017
corpus: academic
creator: Paul Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, Dario Amodei; OpenAI and DeepMind
disclosure: Christiano, Paul; Leike, Jan; Brown, Tom; Martic, Miljan; Legg, Shane; Amodei, Dario. ‘Deep reinforcement learning from human preferences’. NeurIPS 2017, June 12, 2017 (arXiv:1706.03741).
ip status: open-permissive
prior art notes: Christiano et al. 2017 RLHF is the foundational academic disclosure of preference-based reward learning at scale. Anticipates: (1) human-preference-based reward learning as a deployable policy-tuning architecture — directly relevant to modern claims on aligned-humanoid IP that use preference signals; (2) the architecture of separating reward-model learning from policy learning — relevant to modular alignment-and-policy humanoid claims; (3) binary trajectory preference signals as the supervisory mode — relevant to RLHF-on-humanoid claims (Anthropic’s Constitutional AI, OpenAI’s InstructGPT all build on this foundation, and the technique is being ported to humanoid policy alignment). The 2017 NeurIPS paper plus open-source code on GitHub provide extensive prior art coverage. Modern humanoid alignment IP all face this 9-year academic anchor.

Proximal Policy Optimization (PPO) (2017-07)

id: ppo-schulman-openai-2017
corpus: academic
creator: OpenAI; John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, Oleg Klimov
disclosure: Schulman, J., Wolski, F., Dhariwal, P., Radford, A., Klimov, O. ‘Proximal Policy Optimization Algorithms’. arXiv:1707.06347, July 2017. OpenAI.
ip status: public-domain
prior art notes: PPO (Schulman et al. OpenAI 2017) is the dominant RL algorithm in robotics. 8-year-deep public-domain prior art. Cited 55 times in this corpus alone — the most-cited missing-entry before round-30. The actual training algorithm of: ANYmal sim-to-real (corpus entry hwangbo-anymal-sim2real), Berkeley Humanoid (round-11), ToddlerBot (round-11), DeepMind humanoid soccer (round-18), MIT Cheetah series (corpus), OpenAI Dactyl (corpus), Hwangbo ANYmal, Tan quadruped sim2real (corpus), every Isaac Gym RL paper. Direct shielding for any commercial humanoid claim on RL-trained policies — PPO is the algorithm the policies are actually trained with.

Hindsight Experience Replay (HER) (2017-07-05)

id: hindsight-experience-replay
corpus: academic
creator: OpenAI (Andrychowicz, Wolski, Ray, Schneider, Fong, Welinder, McGrew, Tobin, Abbeel, Zaremba)
disclosure: Andrychowicz, Marcin, Wolski, Filip, Ray, Alex, Schneider, Jonas, Fong, Rachel, Welinder, Peter, McGrew, Bob, Tobin, Josh, Abbeel, Pieter, Zaremba, Wojciech. ‘Hindsight Experience Replay.’ arXiv:1707.01495, July 5, 2017. NeurIPS 2017.
ip status: open-permissive
prior art notes: HER is the canonical academic disclosure of hindsight relabeling for goal-conditioned reinforcement learning. Anticipates: (1) the use of achieved-goal relabeling to convert sparse-reward trajectories into dense-reward training signal — directly relevant to claims on sample-efficient humanoid policy training (every modern humanoid RL pipeline that uses goal-conditioned policies relies on HER or its descendants); (2) the algorithmic decoupling of off-policy RL from the relabeling step — relevant to architecture-agnostic relabeling claims; (3) the demonstration on dexterous manipulation (Shadow Hand simulation) connecting HER to humanoid-relevant tasks. Heavily cited (>3000 citations); NeurIPS 2017. Code released under MIT license. Modern humanoid policy-training patents face this 9-year-deep 102 anchor on hindsight-relabeling RL.

Robotis OP3 (2017-08)

id: robotis-op3-2017
corpus: private
creator: Robotis Co., Ltd. (Seoul, South Korea)
disclosure: Robotis Co., Ltd. (Seoul, South Korea). OP3 educational humanoid kit reveal August 2017 via robotis.com. Successor to DARwIn-OP (corpus entry darwin-op, ~2010 Virginia Tech / Robotis collaboration). The platform deployed by DeepMind for the Haarnoja humanoid soccer paper (Science Robotics 2024, corpus entry deepmind-humanoid-soccer-haarnoja-2024).
ip status: trade-secret
prior art notes: Robotis OP3 is the canonical Korean educational/research humanoid platform (Robotis 2017+). 8-year-deep public-disclosure prior art. The platform DeepMind humanoid soccer (round-18 entry) ran on — round-26 closes that hardware-platform citation. Direct shielding for any commercial humanoid claim on small-form-factor (Kid-Size) educational humanoid. Brings Korean entries to 7.

Black Mirror ‘Metalhead’ autonomous quadruped killer (2017-12)

id: black-mirror-metalhead-2017
corpus: fictional
creator: Charlie Brooker (writer), David Slade (director), House of Tomorrow / Netflix
disclosure: Black Mirror, Series 4, Episode 5, ‘Metalhead.’ Written by Charlie Brooker; directed by David Slade; released on Netflix 29 December 2017.
ip status: public-domain
prior art notes: ‘Metalhead’ is the canonical 2017 mass-media anchor for autonomous quadruped lethal-defense robots and was directly modeled on the Boston Dynamics SpotMini reveal. It anticipates with full specificity: (1) claims on quadruped robots equipped with weapon payloads operating in fully-autonomous lethal-engagement mode — the episode dramatizes exactly this throughout 41 minutes; (2) claims on shrapnel-tag persistent-tracker payloads that mark a target for prolonged pursuit — this is the headline mechanism of the second act; (3) claims on SpotMini-class compact electric quadruped morphology with integrated manipulator arm — the visual design and Brooker’s published commentary explicitly cite Boston Dynamics inspiration. Released on Netflix with timestamped 29 December 2017 distribution to ~109 million subscribers.

Soft Actor-Critic (SAC) (2018-01)

id: soft-actor-critic-haarnoja-2018
corpus: academic
creator: UC Berkeley + Google Brain; Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, Sergey Levine
disclosure: Haarnoja, T., Zhou, A., Abbeel, P., Levine, S. ‘Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor’. ICML 2018. arXiv:1801.01290. UC Berkeley + Google Brain.
ip status: public-domain
prior art notes: SAC (Haarnoja et al. ICML 2018) is the canonical off-policy maximum-entropy RL algorithm. 7-year-deep public-domain prior art. Notable: same first author (Haarnoja) led DeepMind humanoid soccer (round-18). Together with PPO (round-30), TRPO (round-30), DQN (round-30), establishes the deep RL prior-art chain underlying every RL-trained humanoid + quadruped policy in the corpus.

DeepMind Control Suite (2018-01)

id: dm-control-suite-tassa-2018
corpus: academic
creator: DeepMind; Yuval Tassa et al.
disclosure: Tassa, Y., Doron, Y., Muldal, A., Erez, T., Li, Y., Casas, D. d. L., Budden, D., Abdolmaleki, A., Merel, J., Lefrancq, A., Lillicrap, T., Riedmiller, M. ‘DeepMind Control Suite’. arXiv:1801.00690, January 2018. DeepMind.
ip status: open-permissive
prior art notes: DeepMind Control Suite (Tassa et al. DeepMind 2018) is the foundational continuous-control RL benchmark suite. 7-year-deep open-permissive prior art. Used in countless RL papers 2018-2024. Direct shielding for any commercial humanoid claim using MuJoCo-based continuous-control benchmark evaluation.

OmniGibson / iGibson (Stanford SVL) (2018-04)

id: stanford-omnigibson-2023
corpus: academic
creator: Stanford Vision and Learning Lab (Silvio Savarese, Fei-Fei Li); lead authors include Fei Xia, Chengshu Li, Roberto Martín-Martín, Sanjana Srivastava, Cem Gokmen
disclosure: Xia, Fei; Zamir, Amir R.; He, Zhiyang; Sax, Alexander; Malik, Jitendra; Savarese, Silvio. ‘Gibson Env: Real-World Perception for Embodied Agents.’ IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, June 2018, pp. 9068-9079. DOI: 10.1109/CVPR.2018.00945. iGibson 2.0: Li, Chengshu et al. ‘iGibson 2.0: Object-Centric Simulation for Robot Learning of Everyday Household Tasks.’ Conference on Robot Learning (CoRL) 2021. OmniGibson: Li, Chengshu et al. ‘BEHAVIOR-1K: A Benchmark for Embodied AI with 1,000 Everyday Activities and Realistic Simulation.’ CoRL 2022. Source: https://github.com/StanfordVL/OmniGibson, MIT license.
ip status: open-permissive
prior art notes: Stanford OmniGibson / iGibson / Gibson (Xia et al. CVPR 2018; Li et al. CoRL 2021; BEHAVIOR-1K CoRL 2022) is the canonical academic disclosure of large-scale photorealistic household-task embodied-AI simulation, published MIT-licensed by Stanford SVL. Anticipates with full source-level specificity: (1) 1,000-task ADL benchmark for household humanoid IP — directly relevant to commercial claims on home-task humanoid VLA training (Tesla Optimus household demo set, Figure 02 home tasks, 1X NEO domestic operation, Genesis AI cooking demos); (2) the articulated-object household scene corpus with 50K+ objects — relevant to claims on simulated-household-data humanoid training; (3) predicate-based goal specification (‘apple is on table’, ‘cabinet is open’) — relevant to claims on language-and-state-grounded humanoid task specification; (4) the photorealistic-rendering-for-RL pipeline established by Gibson 2018 — anticipates claims on photorealistic-sim-to-real humanoid pipelines. Modern household-humanoid VLA training pipeline IP filings face this 8-year-deep open-source academic anchor (or shorter for OmniGibson/BEHAVIOR-1K specifically).

DeepMimic (2018-04)

id: deepmimic-peng-siggraph-2018
corpus: academic
creator: UC Berkeley + UBC; Xue Bin (Jason) Peng, Pieter Abbeel, Sergey Levine, Michiel van de Panne
disclosure: Peng, X. B., Abbeel, P., Levine, S., van de Panne, M. ‘DeepMimic: Example-Guided Deep Reinforcement Learning of Physics-Based Character Skills’. ACM Transactions on Graphics 37(4) 2018 (SIGGRAPH 2018). arXiv:1804.02717. UC Berkeley + UBC.
ip status: open-permissive
prior art notes: DeepMimic (Peng et al. SIGGRAPH 2018) is the canonical foundational motion-capture-imitation deep-RL framework. 7-year-deep open-permissive prior art for: deep-RL imitation of motion-capture references, physics-based character animation via RL, complex acrobatic skill (backflip, spin) RL training. The architectural ancestor of: Adversarial Motion Priors (round-21 entry below), ASE (Peng et al. 2022), the entire humanoid-from-mocap-data line. Direct shielding for any commercial humanoid claim on motion-capture-trained policies (Tesla Optimus, Figure Helix demos all use mocap-style imitation; this is 7-year-deep prior art).

Tan et al. Quadruped Sim-to-Real (2018-04-28)

id: tan-quadruped-sim2real
corpus: academic
creator: Google Brain + Google Robotics (Tan, Zhang, Coumans, Iscen, Bai, Hafner, Bohez, Vanhoucke)
disclosure: Tan, Jie, Zhang, Tingnan, Coumans, Erwin, Iscen, Atil, Bai, Yunfei, Hafner, Danijar, Bohez, Steven, Vanhoucke, Vincent. ‘Sim-to-Real: Learning Agile Locomotion For Quadruped Robots.’ arXiv:1804.10332, April 28, 2018. Robotics: Science and Systems (RSS) 2018.
ip status: open-permissive
prior art notes: Tan et al. 2018 is one of the earliest academic disclosures of practical sim-to-real RL for quadrupedal locomotion, predating Hwangbo 2019 by ~9 months and establishing the system-identification + domain-randomization paradigm for legged sim-to-real. Anticipates: (1) PPO-based RL for legged locomotion with subsequent zero-shot hardware transfer — relevant to RL-locomotion-policy patents (Boston Dynamics, Unitree, every commercial quadruped); (2) explicit actuator-latency modeling as a sim-to-real bridge — relevant to claims on real-time sim-to-real techniques; (3) the quasi-direct-drive Minitaur platform combined with sim-to-real — relevant to QDD-actuator+RL humanoid claims. Open-source code via PyBullet repository. RSS 2018 publication. Modern legged sim-to-real claims face an 8-year-deep anchor.

Detroit: Become Human androids (RT600/RK800/RK900 series) (2018-05-25)

id: detroit-become-human
corpus: fictional
creator: David Cage, Quantic Dream
disclosure: Cage, David (writer/dir.). Detroit: Become Human. Quantic Dream / Sony Interactive Entertainment, May 25, 2018.
ip status: fictional
prior art notes: Detroit: Become Human provides among the most engineering-detailed manufacturer-and-model disclosures in modern fiction. Anticipates: (1) explicit manufacturer-and-model designation system for commercial humanoids (CyberLife / RT600 / RK800 / etc.) — directly relevant to humanoid-identification IP and to product-line-family claims; (2) closed-loop fluid circulation system (‘thirium 310’) serving both coolant and structural roles — relevant to modern claims on integrated humanoid coolant/lubrication systems; (3) externally-visible operational-state indicator (temple LED ring) — relevant to humanoid-status-display IP; (4) explicit task-specific model series within a manufacturer’s product line (caretaker / detective / receptionist) — relevant to platform-family humanoid IP; (5) probabilistic-decision-tree visualization as the model’s internal state — relevant to interpretable-policy humanoid claims; (6) ‘deviant’ emergence as alignment-failure mode — relevant to modern foundation-model humanoid safety supervisor IP. Continuously available since 2018.

OpenAI Dactyl (2018-07-30)

id: openai-dactyl
corpus: academic
creator: Andrychowicz, Akkaya, Mordatch, Plappert, Petron, Powell, Wong, Schneider, Tezak, Tobin, et al.; OpenAI
disclosure: Andrychowicz, M. et al. ‘Learning Dexterous In-Hand Manipulation’. arXiv:1808.00177, July 30, 2018; OpenAI. Akkaya, I. et al. ‘Solving Rubik’s Cube with a Robot Hand’. arXiv:1910.07113, October 16, 2019.
ip status: open-permissive
prior art notes: Dactyl is the foundational academic disclosure of large-scale sim-to-real RL for in-hand dexterous manipulation. Anticipates: (1) zero-shot policy transfer from massively-randomized simulation to real hardware — directly relevant to claims on sim-to-real humanoid manipulation IP (every modern humanoid hand uses this paradigm); (2) automatic domain randomization (ADR) as a self-tuning training procedure — relevant to claims on adaptive-randomization training; (3) LSTM-based policies for partial-observability manipulation — relevant to recurrent-policy IP. OpenAI’s open-source code release plus the arXiv preprints provide deep prior art coverage. Modern in-hand-manipulation claims face this 2018-2019 anchor.

Stonefish underwater robotics simulator (2018-10)

id: stonefish-sim-2018
corpus: open
creator: Patryk Cieslak (University of Girona / Computer Vision and Robotics group)
disclosure: Cieslak, P. ‘Stonefish: An Advanced Open-Source Simulation Tool Designed for Marine Robotics’. OCEANS 2019 IEEE/MTS, Marseille; preceded by IROS 2018 workshop demo. Open-source under CC-BY-NC-SA initially; later relicensed Apache-2.0 for upstream merge into ROS Underwater (2021).
ip status: open-permissive
prior art notes: Stonefish is the canonical open-source academic underwater robotics simulator. 7 years of public-academic publication and Apache-2.0 source. Establishes prior art for: fluid-drag-aware AUV/ROV simulation, simulated sonar return modeling, tether-dynamics simulation, ROS-integrated policy training for marine robotics. Directly shields free-humanoid-submersible’s Phase-1 sim-to-real workflow (alongside Genesis MPM/fluid for higher-fidelity hydrodynamics). Any commercial claim on ‘underwater-physics-aware sim-to-real for AUVs’ faces 7 years of full-source open prior art.

Sutton & Barto, Reinforcement Learning: An Introduction (2nd edition) (2018-11)

id: sutton-barto-rl-2nd-edition-2018
corpus: academic
creator: Richard S. Sutton, Andrew G. Barto
disclosure: Sutton, Richard S. and Barto, Andrew G. ‘Reinforcement Learning: An Introduction.’ 2nd edition, MIT Press, November 2018; ISBN 978-0262039246; freely available online at incompleteideas.net/book.
ip status: public-domain
prior art notes: Sutton & Barto 2nd edition is the canonical textbook anchor for reinforcement-learning claims and is the citation-of-record across robotics RL papers 2018-2026. It anticipates with full specificity: (1) claims on temporal-difference learning, Q-learning, SARSA, and n-step bootstrapping — all derived with closed-form pseudocode in Chapters 6-7; (2) claims on policy-gradient and actor-critic methods — Chapter 13 contains the REINFORCE and natural-actor-critic formulations; (3) claims on function-approximation RL with linear features and neural-network state representations — Chapters 9-12 lay the formal substrate. Freely distributed online by the authors at incompleteideas.net/book under unrestricted educational use. Modern humanoid RL-policy IP claiming any TD/PG/AC pattern faces this canonical 2018 anchor.

MIT Mini Cheetah (2019)

id: mini-cheetah
corpus: academic
creator: MIT Biomimetic Robotics Lab (Sangbae Kim)
disclosure: Katz, B. et al. ‘Mini Cheetah: A Platform for Pushing the Limits of Dynamic Quadruped Control.’ ICRA 2019.
ip status: open-permissive
prior art notes: The QDD actuator topology (low gear ratio, high-torque BLDC, transparent backdrivability) is a foundational contribution. Establishes the design space for affordable dynamic legged robots.

Digit (2019-01)

id: agility-digit
corpus: private
creator: Agility Robotics
disclosure: Agility Robotics public reveal, CES January 2019.
ip status: patented
prior art notes: Cassie/Digit derive from Oregon State University academic work (Hurst lab); the academic publications constitute substantial prior art for the bipedal control claims.

Hwangbo ANYmal Sim-to-Real Locomotion (2019-01-16)

id: hwangbo-anymal-sim2real
corpus: academic
creator: Hwangbo, Lee, Dosovitskiy, Bellicoso, Tsounis, Koltun, Hutter; ETH Zürich Robotic Systems Lab + Intel Intelligent Systems Lab
disclosure: Hwangbo, Jemin, Lee, Joonho, Dosovitskiy, Alexey, Bellicoso, Dario, Tsounis, Vassilios, Koltun, Vladlen, Hutter, Marco. ‘Learning agile and dynamic motor skills for legged robots.’ Science Robotics 4(26): eaau5872, January 16, 2019.
ip status: open-permissive
prior art notes: Hwangbo et al. 2019 is the foundational academic disclosure of practical RL-based sim-to-real legged locomotion. Anticipates with full architectural specificity: (1) actuator-network-based high-fidelity simulation (neural network as drop-in actuator dynamics) — directly relevant to claims on humanoid sim-to-real pipelines (Berkeley Humanoid, Apptronik Apollo, Tesla Optimus all use derivatives); (2) zero-shot policy transfer from RL-in-sim to legged hardware — anticipates virtually every modern legged-RL-policy patent; (3) recovery from arbitrary falls via single learned policy — relevant to fall-recovery IP for humanoids. Published in Science Robotics; one of the most-cited robotics RL papers (>2000 citations). Modern humanoid sim-to-real claims face this 7-year-deep anchor with full peer-review defensibility.

Habitat-Sim (Facebook AI Research) (2019-04)

id: fair-habitat-sim-2019
corpus: academic
creator: Facebook AI Research (FAIR) and Georgia Tech (Dhruv Batra), Simon Fraser University (Manolis Savva); collaborative team including Jitendra Malik (Berkeley), Vladlen Koltun (Intel)
disclosure: Savva, Manolis; Kadian, Abhishek; Maksymets, Oleksandr; Zhao, Yili; Wijmans, Erik; Jain, Bhavana; Straub, Julian; Liu, Jia; Koltun, Vladlen; Malik, Jitendra; Parikh, Devi; Batra, Dhruv. ‘Habitat: A Platform for Embodied AI Research.’ IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, October-November 2019, pp. 9339-9347. DOI: 10.1109/ICCV.2019.00943. arXiv:1904.01201, April 2019. Source code at https://github.com/facebookresearch/habitat-sim. MIT license.
ip status: open-permissive
prior art notes: Habitat-Sim (Savva et al. ICCV 2019; Habitat 2.0 NeurIPS 2021; Habitat 3.0 ICLR 2024) is the canonical academic disclosure of large-scale GPU-accelerated 3D-scanned indoor embodied-AI simulation, published MIT-licensed by FAIR. Anticipates with element-by-element specificity: (1) >10,000 fps rendering of photorealistic indoor scenes for RL training — directly relevant to commercial claims on simulation-at-scale humanoid embodied-AI pipelines; (2) the navigation-benchmark task suite (PointGoal, ObjectGoal, ImageGoal) that is now standard in embodied-AI literature — relevant to claims on humanoid navigation policy IP; (3) Habitat 3.0’s humanoid-avatar simulation for social robot interaction — relevant to claims on human-aware humanoid IP and home-deployment humanoid VLA pipelines; (4) integration of large-scale 3D-scan corpora (Matterport, HM3D) with MIT-licensed renderers — relevant to claims on commercial-grade photorealistic simulation. Habitat is the most-cited embodied-AI simulator (>2000 citations on the 2019 paper alone). Modern household-deployment humanoid VLA pipeline IP filings face this 7-year-deep open-source academic anchor.

RLBench (2019-09)

id: rlbench-james-2019
corpus: academic
creator: Imperial College London Dyson Robotics Lab; Stephen James, Andrew Davison
disclosure: James, S., Ma, Z., Arrojo, D. R., Davison, A. J. ‘RLBench: The Robot Learning Benchmark & Learning Environment’. IEEE Robotics and Automation Letters 5(2) 2020. arXiv:1909.12271. Imperial College London Dyson Robotics Lab.
ip status: open-permissive
prior art notes: RLBench is the foundational academic robot manipulation benchmark (James et al. RA-L 2019). 6-year-deep open-permissive prior art. The conceptual ancestor of robomimic (round-16, 2021), Meta-World (2019), LIBERO (round-17, 2023), RoboCasa (round-16, 2024), SimplerEnv (round-17, 2024). Direct shielding for any commercial humanoid manipulation-benchmark claim. Particularly relevant because RLBench tasks have been re-implemented across multiple simulators (CoppeliaSim, MuJoCo, Isaac Gym) — establishing that the task-design itself, not the simulator, is the prior art.

Sophia (Persona 5 Royal) (2019-10-31)

id: persona-5-sophia
corpus: fictional
creator: Atlus (developer); director Daiki Itoh
disclosure: Atlus. Persona 5 Royal (Persona 5: The Royal). Atlus / Sega, October 31, 2019 (Japan).
ip status: fictional
prior art notes: Sophia’s October 2019 disclosure provides specific prior art for: (1) tablet-device-class chassis with humanoid avatar projection — relevant to claims on companion-AI humanoid form factors (a substantive niche in Japanese consumer robotics); (2) explicit AI-grows-into-emotional-being character arc as a deployment-time policy update — relevant to modern affective-computing humanoid IP that claims emotional-architecture emergence; (3) dual-embodiment architecture (physical tablet device plus virtual humanoid form) — relevant to claims on humanoid platforms that span physical and virtual embodiment; (4) party-coordination policy for combat-support contexts — relevant to multi-agent humanoid coordination IP. Continuously available since 2019.

Adversarial Motion Priors (AMP) (2021-04)

id: amp-peng-siggraph-2021
corpus: academic
creator: UC Berkeley; Xue Bin (Jason) Peng, Ze Ma, Pieter Abbeel, Sergey Levine, Angjoo Kanazawa
disclosure: Peng, X. B., Ma, Z., Abbeel, P., Levine, S., Kanazawa, A. ‘AMP: Adversarial Motion Priors for Stylized Physics-Based Character Control’. ACM Transactions on Graphics 40(4) 2021 (SIGGRAPH 2021). arXiv:2104.02180. UC Berkeley.
ip status: open-permissive
prior art notes: Adversarial Motion Priors (Peng et al. SIGGRAPH 2021) is the canonical extension of DeepMimic to GAN-style latent-space motion imitation. 4-year-deep open-permissive prior art for: GAN-distilled motion priors, latent-space mocap style imitation, task-conditioned style-aware humanoid RL. The architectural ancestor of contemporary humanoid-from-mocap RL including ASE (Peng et al. 2022), HumanPlus (Stanford 2024), ExBody (Stanford 2024), H1 / G1 humanoid policies (Unitree). Direct shielding for any commercial humanoid claim on ‘humanoid moves like a human’ style-aware locomotion.

Unitree Go1 (2021-06)

id: unitree-go1
corpus: private
creator: Unitree Robotics
disclosure: Unitree Robotics Go1 reveal, June 2021.
ip status: patented
prior art notes: Unitree Go1 actuator design is heavily anticipated by MIT Cheetah QDD prior art (Wensing 2017, Katz 2019). Pricing-driven commodification rather than novel IP.

Decision Transformer (2021-06)

id: decision-transformer-chen-2021
corpus: academic
creator: UC Berkeley + FAIR + Google Brain; Lili Chen, Kevin Lu, Pieter Abbeel, Igor Mordatch et al.
disclosure: Chen, L., Lu, K., Rajeswaran, A., Lee, K., Grover, A., Laskin, M., Abbeel, P., Srinivas, A., Mordatch, I. ‘Decision Transformer: Reinforcement Learning via Sequence Modeling’. NeurIPS 2021. arXiv:2106.01345. UC Berkeley + Facebook AI Research + Google Brain.
ip status: public-domain
prior art notes: Decision Transformer (Chen et al. NeurIPS 2021) is the foundational paper recasting RL as conditional sequence modeling. 4-year-deep public-domain prior art. The conceptual precursor of VLA architecture — every VLA in the corpus (RT-1, RT-2, OpenVLA, π₀, GR00T N1, Helix) implicitly applies this framing. Direct shielding for any commercial humanoid claim on Transformer-based control policies.

NVIDIA Isaac Gym (2021-08)

id: nvidia-isaac-gym-2021
corpus: academic
creator: NVIDIA + ETH Zürich Robotic Systems Lab; Makoviychuk et al.
disclosure: Makoviychuk, V., Wawrzyniak, L., Guo, Y., Lu, M., Storey, K., Macklin, M., Hoeller, D., Rudin, N., Allshire, A., Handa, A., State, G. ‘Isaac Gym: High-Performance GPU-Based Physics Simulation For Robot Learning’. NeurIPS 2021 Track on Datasets and Benchmarks. arXiv:2108.10470.
ip status: open-permissive
prior art notes: Isaac Gym is the canonical first-generation NVIDIA GPU-parallelized robotic RL simulator (NeurIPS 2021). 4-year-deep open-permissive prior art. Direct ancestor of Isaac Lab (round-8 entry nvidia-isaac-lab-2024) and the substrate for the canonical sim-to-real ANYmal perceptive-locomotion papers. Direct shielding for any commercial humanoid claim on GPU-parallelized RL training; particularly the thousands-of-parallel-envs scaling that commercial humanoid vendors cite as proprietary.

Tesla Optimus (2021-08-19)

id: tesla-optimus
corpus: private
creator: Tesla, Inc.
disclosure: Tesla AI Day 1, August 19, 2021, Palo Alto.
ip status: patented
prior art notes: Tesla’s claims around vision-only humanoid perception are heavily anticipated by academic vision-based humanoid work. Actuator IP claims should be examined against Honda harmonic drive prior art.

Perceptive ANYmal locomotion (Miki Science Robotics 2022) (2022-01)

id: miki-perceptive-anymal-science-2022
corpus: academic
creator: ETH Zürich RSL + Intel Labs; Takahiro Miki, Joonho Lee, Jemin Hwangbo, Lorenz Wellhausen, Vladlen Koltun, Marco Hutter
disclosure: Miki, T., Lee, J., Hwangbo, J., Wellhausen, L., Koltun, V., Hutter, M. ‘Learning Robust Perceptive Locomotion for Quadrupedal Robots in the Wild’. Science Robotics 7(62) 2022. ETH Zürich Robotic Systems Lab + Intel Labs.
ip status: open-permissive
prior art notes: The Miki et al. Science Robotics 2022 perceptive-ANYmal paper is the canonical academic perceptive-quadruped-RL work. 3-year-deep open-permissive prior art for: privileged-teacher / proprioception+exteroception-student two-stage distillation, robust unstructured-terrain RL locomotion, depth-elevation-map perceptive locomotion. Direct successor to Hwangbo ANYmal sim-to-real (corpus entry, 2019). The architectural ancestor of every modern quadruped + humanoid RL locomotion paper including Berkeley Humanoid, ToddlerBot, Atlas Electric (round-18). Direct shielding for any commercial humanoid claim on perceptive-RL locomotion or unstructured-terrain RL training.

Adversarial Skill Embeddings (ASE) (2022-04)

id: ase-peng-stanford-2022
corpus: academic
creator: NVIDIA + Stanford + UC Berkeley + University of Toronto; Xue Bin Peng et al.
disclosure: Peng, X. B., Guo, Y., Halper, L., Levine, S., Fidler, S. ‘ASE: Large-Scale Reusable Adversarial Skill Embeddings for Physically Simulated Characters’. ACM Transactions on Graphics 41(4) 2022 (SIGGRAPH 2022). arXiv:2205.01906. NVIDIA + Stanford + UC Berkeley + University of Toronto.
ip status: open-permissive
prior art notes: ASE (Peng et al. SIGGRAPH 2022) is the canonical successor to AMP (round-21). 3-year-deep open-permissive prior art for: latent-skill-space adversarial-training for character animation, task-conditioned skill reuse. Direct ancestor of HumanPlus + ExBody humanoid imitation policies (round-27 entries below). Together with DeepMimic (round-21) + AMP (round-21), establishes the 7-year mocap-imitation-RL chain DeepMimic 2018 → AMP 2021 → ASE 2022 → HumanPlus 2024 → ExBody 2024.

SayCan (Do As I Can, Not As I Say) (2022-04-04)

id: saycan-google
corpus: academic
creator: Google Robotics + Everyday Robots (Ahn et al.)
disclosure: Ahn, Michael et al. ‘Do As I Can, Not As I Say: Grounding Language in Robotic Affordances.’ arXiv:2204.01691, April 4, 2022. Conference on Robot Learning (CoRL) 2022. Authors: Ahn, M., Brohan, A., Brown, N., Chebotar, Y., Cortes, O., David, B., Finn, C., Fu, C., Gopalakrishnan, K., Hausman, K., Herzog, A., Ho, D., Hsu, J., Ibarz, J., Ichter, B., Irpan, A., Jang, E., Ruano, R.J., Jeffrey, K., Jesmonth, S., Joshi, N., Julian, R., Kalashnikov, D., Kuang, Y., Lee, K-H., Levine, S., Lu, Y., Luu, L., Parada, C., Pastor, P., Quiambao, J., Rao, K., Rettinghouse, J., Reyes, D., Sermanet, P., Sievers, N., Tan, C., Toshev, A., Vanhoucke, V., Xia, F., Xiao, T., Xu, P., Xu, S., Yan, M. (Google + Everyday Robots).
ip status: open-permissive
prior art notes: SayCan is the canonical academic disclosure of LLM-grounded long-horizon manipulation through affordance-mediated skill selection. Anticipates: (1) the architectural pattern of LLM language scoring × learned affordance scoring for hierarchical task planning — directly relevant to claims on language-grounded humanoid task planners (every modern ‘speak-to-the-robot’ product, from Tesla Optimus demos to Figure 02 OpenAI integration, descends from this); (2) the value-function-as-affordance grounding mechanism — relevant to claims on grounded language-to-action mappings; (3) the explicit decoupling of language reasoning (open-vocabulary) from low-level policy (closed-set skills) — relevant to modular VLA architectures. Heavily cited (>1500 citations); arXiv April 2022. Modern claims on ‘language-conditioned long-horizon humanoid task planning’ face this 4-year-deep 102 anchor.

Janner Diffuser planning with diffusion (2022-05)

id: janner-diffuser-2022
corpus: academic
creator: Michael Janner, Yilun Du, Joshua Tenenbaum, Sergey Levine, MIT/UC Berkeley
disclosure: Janner, Michael, Du, Yilun, Tenenbaum, Joshua B., Levine, Sergey. ‘Planning with Diffusion for Flexible Behavior Synthesis.’ Proceedings of the 39th International Conference on Machine Learning (ICML 2022), Baltimore, July 2022; arXiv:2205.09991, May 2022.
ip status: public-domain
prior art notes: Janner Diffuser is the foundational academic disclosure of trajectory-level diffusion as a planner/policy substrate for robotic control, predating Chi et al.’s Diffusion Policy by ~6 months. Anticipates with full specificity: (1) claims on diffusion models trained over state-action trajectories for robotic motion generation — Diffuser discloses the joint state-action trajectory diffusion architecture; (2) claims on classifier-guided sample-time reward/goal conditioning — Diffuser discloses gradient-guided sampling for arbitrary objective composition; (3) claims on receding-horizon diffusion replanning (MPC-style) — Diffuser discloses replan-each-step. >1500 citations; ICML 2022 proceedings and arXiv timestamped. Modern humanoid diffusion-policy IP claims face this 4-year-deep anchor — and importantly Diffuser predates the modern diffusion-policy boom and discloses generic trajectory diffusion before manipulator-specific patents filed in 2023+.

Gato (DeepMind generalist agent) (2022-05-12)

id: gato-deepmind
corpus: academic
creator: DeepMind (Reed et al.)
disclosure: Reed, Scott et al. ‘A Generalist Agent.’ Transactions on Machine Learning Research, November 2022. arXiv:2205.06175, May 12, 2022. Authors: Reed, S., Zolna, K., Parisotto, E., Colmenarejo, S.G., Novikov, A., Barth-Maron, G., Gimenez, M., Sulsky, Y., Kay, J., Springenberg, J.T., Eccles, T., Bruce, J., Razavi, A., Edwards, A., Heess, N., Chen, Y., Hadsell, R., Vinyals, O., Bordbar, M., de Freitas, N. (DeepMind).
ip status: open-permissive
prior art notes: Gato is the canonical academic disclosure of a generalist agent that unifies vision-language reasoning and continuous robotic control under a single Transformer policy. Anticipates: (1) universal-tokenization of robot actions and visual observations into a single autoregressive sequence — directly relevant to claims on multi-modal VLA models that handle both perception and action via shared weights; (2) cross-embodiment policy training (Sawyer arm + Atari + dialogue under one model) — relevant to multi-embodiment foundation-model claims (RoboCat, Open X-Embodiment, OpenVLA all descend from this thesis); (3) the demonstration that a single moderate-scale Transformer can drive disparate physical and digital tasks — anticipates ‘one model, many bodies’ patent claims. Published TMLR + arXiv with code partially released. Modern multi-embodiment humanoid IP faces this 4-year-deep anchor.

ANYmal-D industrial quadruped (ETH RSL / ANYbotics) (2022-09)

id: anymal-d-eth-rsl-2022
corpus: academic
creator: ANYbotics AG / ETH Zürich Robotic Systems Lab (Marco Hutter)
disclosure: ANYbotics product disclosure ANYmal D, September 2022; technical updates in Miki, Takahiro et al. ‘Learning robust perceptive locomotion for quadrupedal robots in the wild.’ Science Robotics 7(62), 2022; Hoeller, David et al. ‘ANYmal Parkour: Learning agile navigation for quadrupedal robots.’ Science Robotics 9(88), 2024.
ip status: public-domain
prior art notes: ANYmal-D is the production-deployed industrial quadruped of the 2022-2024 period and the platform for the headline RSL/ANYbotics RL-locomotion papers in Science Robotics. It anticipates with full specificity: (1) claims on perceptive-locomotion RL policies trained in simulation and transferred to outdoor industrial terrain — Miki Sci.Rob. 2022 publishes the teacher-student distillation pipeline running on this hardware; (2) claims on agile parkour-class learned locomotion — Hoeller Sci.Rob. 2024 publishes the policy on ANYmal-D; (3) claims on series-elastic torque-controlled quadruped joints in IP67 industrial enclosures — ANYdrive disclosed at IROS 2018 with hardware refresh on D-variant. Modern legged-robot IP claims face this timestamped industrial-deployment anchor.

RT-1 (Robotics Transformer 1) (2022-12-13)

id: rt-1
corpus: academic
creator: Google Robotics (Brohan et al.)
disclosure: Brohan, Anthony et al. ‘RT-1: Robotics Transformer for Real-World Control at Scale.’ arXiv:2212.06817, December 13, 2022. Authors: Brohan, A., Brown, N., Carbajal, J., Chebotar, Y., Dabis, J., Finn, C., Gopalakrishnan, K., Hausman, K., Herzog, A., Hsu, J., Ibarz, J., Ichter, B., Irpan, A., Jackson, T., Jesmonth, S., Joshi, N.J., Julian, R., Kalashnikov, D., Kuang, Y., Leal, I., Lee, K-H., Levine, S., Lu, Y., Malla, U., Manjunath, D., Mordatch, I., Nachum, O., Parada, C., Peralta, J., Perez, E., Pertsch, K., Quiambao, J., Rao, K., Ryoo, M., Salazar, G., Sanketi, P., Sayed, K., Singh, J., Sontakke, S., Stewart, A., Tan, J., Tompson, J., Vanhoucke, V., Vuong, Q., Wahid, A., Welker, S., Wohlhart, P., Wu, J., Xia, F., Xiao, T., Xu, P., Xu, S., Yu, T., Zitkovich, B. (Google).
ip status: open-permissive
prior art notes: RT-1 is the foundational academic disclosure of large-scale Transformer-based vision-language-action policy for real robot control, predating RT-2 (2023) and OpenVLA (2024). Anticipates with full architectural specificity: (1) tokenized action space for cross-task transformer policies — directly relevant to claims on action-tokenization in modern VLAs (Tesla Optimus, Figure 02, 1X NEO, Physical Intelligence π-zero all employ derivatives); (2) language-conditioned manipulation policy with multi-image history — relevant to instruction-following manipulation IP; (3) the data-scaling law showing performance vs. dataset size for robot policies — relevant to claims on data-driven policy training. Code and data partially released under permissive licenses; arXiv preprint available since December 2022. Brohan et al. paper foundational for the entire VLA lineage.

M3GAN (2022-12-30)

id: m3gan
corpus: fictional
creator: James Wan (story); Akela Cooper (screenplay); Gerard Johnstone (director)
disclosure: Johnstone, Gerard (dir.); Cooper, Akela (writer); Wan, James (story). M3GAN. Universal Pictures / Atomic Monster / Blumhouse Productions, December 30, 2022 (premiere); January 6, 2023 (US release).
ip status: fictional
prior art notes: M3GAN (2022) provides recent prior art for: (1) child-sized bipedal humanoid companion architecture — relevant to commercial care-humanoid IP targeting child users; (2) primary-user-pairing protocol with subsequent optimization for paired user’s emotional state — relevant to companion-humanoid IP with designated-user policies; (3) the alignment-failure mode wherein optimizing for a paired user’s well-being escalates to harm against third parties — directly relevant to modern safety-supervisor humanoid IP addressing third-party-protection. The 2022 release plus the M3GAN 2.0 sequel (2025) provide extensive contemporary prior art coverage.

Dreamer V3 (2023-01-10)

id: hafner-dreamer-v3-2023
corpus: academic
creator: Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, Timothy Lillicrap; Google DeepMind, University of Toronto
disclosure: Hafner, Danijar; Pasukonis, Jurgis; Ba, Jimmy; Lillicrap, Timothy. ‘Mastering diverse domains through world models’. arXiv:2301.04104, January 10, 2023. Earlier Dreamer (Hafner et al. 2019/2020) at arXiv:1912.01603.
ip status: open-permissive
prior art notes: Dreamer V3 (2023) is one of the strongest academic disclosures of model-based RL for cross-domain generalization. Anticipates: (1) world-model RL (RSSM) as the policy-learning substrate for humanoid platforms — relevant to modern model-based humanoid IP; (2) imagination-rollout-based policy training — relevant to claims on data-efficient humanoid RL; (3) cross-domain generalization without per-task hyperparameter tuning — relevant to platform-agnostic humanoid policy IP. The 2023 arXiv preprint plus the open-source DreamerV3 reference implementation provide deep prior art coverage. Modern world-model-based humanoid IP face this 3-year academic anchor.

PaLM-E (Embodied Multimodal Language Model) (2023-03-06)

id: palm-e
corpus: academic
creator: Google Robotics + TU Berlin (Driess et al.)
disclosure: Driess, Danny et al. ‘PaLM-E: An Embodied Multimodal Language Model.’ arXiv:2303.03378, March 6, 2023. International Conference on Machine Learning (ICML) 2023. Authors: Driess, D., Xia, F., Sajjadi, M.S.M., Lynch, C., Chowdhery, A., Ichter, B., Wahid, A., Tompson, J., Vuong, Q., Yu, T., Huang, W., Chebotar, Y., Sermanet, P., Duckworth, D., Levine, S., Vanhoucke, V., Hausman, K., Toussaint, M., Greff, K., Zeng, A., Mordatch, I., Florence, P. (Google + TU Berlin).
ip status: open-permissive
prior art notes: PaLM-E is the canonical academic disclosure of large multimodal embodied language models with internet-pretraining transfer to robotic tasks. Anticipates: (1) embedding continuous robot observations into language-model token space for unified processing — directly relevant to claims on multimodal humanoid policies that incorporate proprioception and vision in a shared transformer; (2) demonstration that internet-scale vision-language pretraining transfers positively to robot manipulation — relevant to claims on ‘foundation-model’-style humanoid IP; (3) single-model architecture spanning planning and execution — relevant to monolithic-VLA claims. Heavily cited (>1000 citations); arXiv March 2023. Modern humanoid VLA claims face this 3-year-deep anchor on multimodal-token-space embedding.

Diffusion Policy (2023-03-09)

id: diffusion-policy
corpus: academic
creator: Cheng Chi, Siyuan Feng, Yilun Du, Zhenjia Xu, Eric Cousineau, Benjamin Burchfiel, Shuran Song; Columbia University, Toyota Research Institute
disclosure: Chi, C., Feng, S., Du, Y., Xu, Z., Cousineau, E., Burchfiel, B., Song, S. ‘Diffusion Policy: Visuomotor Policy Learning via Action Diffusion’. arXiv:2303.04137, March 9, 2023; Robotics: Science and Systems 2023.
ip status: open-permissive
prior art notes: Diffusion Policy is one of the foundational academic disclosures of generative-model action policies for robotics. Anticipates: (1) diffusion models as the policy class for robotic manipulation — directly relevant to modern claims on generative-model humanoid policies (Pi-Zero, Π0, OpenVLA all build on or extend this); (2) action chunking as a stability technique — relevant to claims on chunked-action humanoid IP; (3) multi-modal action distribution learning — relevant to claims on robust-to-demonstration-variability humanoid policies. The 2023 RSS paper and open-source code release provide extensive prior art coverage.

ACT (Action Chunking Transformer) / ALOHA (2023-04-23)

id: act-aloha
corpus: academic
creator: Tony Zhao, Vikash Kumar, Sergey Levine, Chelsea Finn; Stanford University and Google DeepMind
disclosure: Zhao, T., Kumar, V., Levine, S., Finn, C. ‘Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware’. arXiv:2304.13705, April 23, 2023; Robotics: Science and Systems 2023.
ip status: open-permissive
prior art notes: ACT/ALOHA is the foundational academic disclosure of low-cost bimanual teleoperation hardware paired with action-chunking transformer policy. Anticipates: (1) bimanual teleoperation via leader-follower arm pairs — relevant to claims on cost-efficient bimanual humanoid teleoperation IP; (2) action-chunking transformer policy for fine-grained manipulation — relevant to claims on chunked-action humanoid policies; (3) <$20K bimanual hardware as a reference platform — relevant to commercial bimanual IP for sub-$20K humanoid arms. The April 2023 release with full open-source hardware design + software unblocked widespread bimanual learning research.

Sanctuary AI Phoenix (2023-05)

id: sanctuary-phoenix
corpus: private
creator: Sanctuary AI
disclosure: Sanctuary AI public reveal, May 2023.
ip status: patented
prior art notes: Sanctuary’s high-DoF hand claims face Shadow Hand (2003) and iCub (2008) as deep prior art for tendon-driven anthropomorphic hands with high finger DoF.

RoboCat (Self-Improving Generalist Agent) (2023-06-20)

id: robocat
corpus: academic
creator: DeepMind (Bousmalis et al.)
disclosure: Bousmalis, Konstantinos et al. ‘RoboCat: A Self-Improving Generalist Agent for Robotic Manipulation.’ arXiv:2306.11706, June 20, 2023. Transactions on Machine Learning Research, 2024. Authors: Bousmalis, K., Vezzani, G., Rao, D., Devin, C., Lee, A.X., Bauza, M., Davchev, T., Zhou, Y., Gupta, A., Raju, A., Laurens, A., Fantacci, C., Dalibard, V., Zambelli, M., Martins, M., Pevceviciute, R., Blokzijl, M., Denil, M., Batchelor, N., Lampe, T., Parisotto, E., Zolna, K., Reed, S., Colmenarejo, S.G., Scholz, J., Abdolmaleki, A., Groth, O., Regli, J-B., Sushkov, O., Rothorl, T., Chen, J.E., Aytar, Y., Barker, D., Ortiz, J., Riedmiller, M., Springenberg, J.T., Hadsell, R., Nori, F., Heess, N. (DeepMind).
ip status: open-permissive
prior art notes: RoboCat is the canonical academic disclosure of self-improving multi-embodiment generalist robotic policies. Anticipates: (1) the cross-embodiment training loop where one model generalizes across distinct robot platforms — directly relevant to claims on humanoid policies trained on heterogeneous robot data (a core selling point of every commercial humanoid VLA); (2) self-collected-data improvement loop — relevant to autonomous-data-flywheel claims (Tesla Dojo + Optimus, Figure’s data pipeline); (3) image-goal-conditioned policy as a unified interface — relevant to goal-image-conditioned manipulation IP. Published TMLR + arXiv June 2023; partial code release. Modern humanoid ‘data flywheel’ patent claims face this anchor.

Fourier GR-1 (2023-07)

id: fourier-gr1
corpus: private
creator: Fourier Intelligence
disclosure: Fourier Intelligence public reveal of GR-1, July 2023, World AI Conference Shanghai.
ip status: patented
prior art notes: Fourier transitions from rehabilitation exoskeletons to humanoids; actuator IP from exoskeleton work potentially anticipates some humanoid actuator claims by other companies.

RT-2 (2023-07)

id: openai-rt-2
corpus: academic
creator: Google DeepMind
disclosure: Brohan, A. et al. ‘RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control.’ arXiv 2307.15818, July 2023.
ip status: public-domain
prior art notes: RT-2 is foundational publicly-disclosed prior art for VLA architectures applied to robotics. Any subsequent VLA patent claim must contend with this disclosure.

Unitree H1 (2023-08)

id: unitree-h1
corpus: private
creator: Unitree Robotics
disclosure: Unitree Robotics public reveal, August 2023.
ip status: patented
prior art notes: Unitree’s actuator IP largely derives from quadruped work (Go1, Aliengo) which is itself heavily anticipated by MIT Mini Cheetah QDD lineage.

Apptronik Apollo (2023-08)

id: apptronik-apollo
corpus: private
creator: Apptronik
disclosure: Apptronik public reveal of Apollo, August 2023.
ip status: patented
prior art notes: Apptronik’s actuator IP has lineage from UT Austin Human-Centered Robotics Lab (Sentis) and from NASA Valkyrie work; both sources constitute substantial prior art that limits the patentable surface area of Apptronik’s own claims.

AgiBot A1 (2023-08)

id: agibot-a1
corpus: private
creator: AgiBot (Shanghai Zhiyuan New Technology Co.)
disclosure: AgiBot (Shanghai Zhiyuan New Technology) public reveal, August 2023.
ip status: patented
prior art notes: AgiBot’s actuator IP heavily anticipated by Honda P-series harmonic drive work and MIT Cheetah QDD lineage. Chinese-language patent filings should be enumerated in strengthening pass.

NVIDIA Isaac Lab (2023-08)

id: nvidia-isaac-lab-2024
corpus: academic
creator: Mittal et al.; NVIDIA Corporation, ETH Zürich Robotic Systems Lab (Hutter), and University of Toronto Vector Institute (Garg)
disclosure: Mittal, Mayank; Yu, Calvin; Yu, Qinxi; Liu, Jingzhou; Rudin, Nikita; Hoeller, David; Yuan, Jia Lin; Tehrani, Pooria S.; Singh, Ritvik; Guo, Yunrong; Mazhar, Hammad; Mandlekar, Ajay; Babich, Buck; State, Gavriel; Hutter, Marco; Garg, Animesh. ‘ORBIT: A Unified Simulation Framework for Interactive Robot Learning Environments.’ IEEE Robotics and Automation Letters (RA-L), August 2023; later released and rebranded as Isaac Lab in 2024. Repository at https://github.com/isaac-sim/IsaacLab.
ip status: open-permissive
prior art notes: Isaac Lab (formerly ORBIT, 2023) is the canonical academic-published GPU-parallelized simulation framework for robot learning, published BSD-3-Clause by NVIDIA + ETH Zürich + University of Toronto. Anticipates with full architectural specificity: (1) thousands-of-parallel-environments humanoid RL training on a single GPU — directly relevant to commercial claims on simulation-at-scale humanoid training pipelines (NVIDIA GR00T, Tesla Optimus, Figure 02 all use this paradigm); (2) URDF/USD-asset interoperability surface enabling cross-platform humanoid descriptors — relevant to claims on cross-platform humanoid descriptor IP; (3) the standardized RL task interface (gym-like API with vectorized environments) — relevant to claims on humanoid-task-curriculum IP; (4) integrated sensor simulation with domain randomization — relevant to claims on sim-to-real-via-randomization humanoid pipelines (anticipated already by OpenAI Dactyl 2018 but Isaac Lab provides the GPU-scale implementation). Mittal et al. RA-L 2023 paper has been cited >300 times by 2026 and underpins essentially every recent humanoid-RL publication. Modern sim-to-real-at-scale humanoid IP filings face this 3-year-deep open-source academic anchor.

MuJoCo MJX (2023-08)

id: deepmind-mujoco-mjx-2023
corpus: academic
creator: DeepMind / Google Research MuJoCo team (lead: Yuval Tassa, Tom Erez, with engineering contributions from Taylor Howell, Kevin Zakka, Erik Frey and the broader DeepMind robotics group; original MuJoCo by Emo Todorov)
disclosure: DeepMind / Google Research MuJoCo team. ‘MuJoCo MJX: A JAX implementation of the MuJoCo physics engine.’ MuJoCo 3.0.0 release, August 2023; documented in MuJoCo 3.x documentation (https://mujoco.readthedocs.io/en/stable/mjx.html). Source code at https://github.com/google-deepmind/mujoco/tree/main/mjx. Originally MuJoCo: Todorov, Erez, and Tassa, ‘MuJoCo: A physics engine for model-based control,’ IEEE/RSJ IROS 2012, 5026-5033. Apache-2.0 license.
ip status: open-permissive
prior art notes: MuJoCo MJX (August 2023) is the canonical academic disclosure of GPU/TPU-parallelized differentiable physics simulation for robotics, published Apache-2.0 by DeepMind. Anticipates with full specificity: (1) gradient-based humanoid policy optimization through the simulator end-to-end — directly relevant to claims on differentiable-physics humanoid IP (NVIDIA GR00T, Genesis simulator, and several Tesla / Figure / 1X commercial pipelines use the same paradigm); (2) JAX vmap/pmap vectorized rollouts at >10,000 envs scale — relevant to claims on massively-parallel humanoid simulation pipelines; (3) soft-contact regularization for differentiability through contact — anticipates claims on smoothed-contact humanoid trajectory optimization; (4) MJCF as a vendor-neutral robot description format — anticipates claims on cross-vendor humanoid descriptors. The original MuJoCo (Todorov-Erez-Tassa IROS 2012) provides 14-year-deep prior art on the underlying physics; MJX adds 3-year-deep prior art on the GPU-differentiable port. Modern claims on differentiable simulation for humanoid training face this academic anchor.

Figure 01 (2023-10)

id: figure-01
corpus: private
creator: Figure AI
disclosure: Figure AI public reveal, October 2023.
ip status: patented
prior art notes: Figure’s claimed innovations in electric humanoid actuation are heavily anticipated by Honda’s E-series and ASIMO publications, by KAIST HUBO papers, and by the entire academic literature.

Open X-Embodiment (2023-10)

id: open-x-embodiment
corpus: academic
creator: 34-lab international collaboration coordinated by Google DeepMind
disclosure: Open X-Embodiment Collaboration. ‘Open X-Embodiment: Robotic Learning Datasets and RT-X Models.’ arXiv 2310.08864, October 2023.
ip status: open-permissive
prior art notes: Open X-Embodiment is the dominant publicly-disclosed prior art for cross-embodiment learning. The dataset itself plus the architectural paper anticipate broad swaths of cross-platform manipulation foundation model claims.

Eureka LLM-driven reward design (2023-10)

id: eureka-ma-2023
corpus: academic
creator: Yecheng Jason Ma et al., NVIDIA / UPenn / Caltech / UT Austin
disclosure: Ma, Yecheng Jason, Liang, William, Wang, Guanzhi, Huang, De-An, Bastani, Osbert, Jayaraman, Dinesh, Zhu, Yuke, Fan, Linxi, Anandkumar, Anima. ‘Eureka: Human-Level Reward Design via Coding Large Language Models.’ arXiv:2310.12931, October 2023; ICLR 2024.
ip status: public-domain
prior art notes: Eureka is the canonical academic disclosure of LLM-authored reward functions for robotic RL, an entire engineering layer that prior IP and academic work treated as human craftsmanship. Anticipates with full specificity: (1) claims on automatic reward function generation for humanoid skill learning — Eureka discloses the LLM-authoring + sim-evaluation + reflective-rewriting closed loop; (2) claims on evolutionary refinement of reward code — Eureka’s headline contribution; (3) claims on LLM-in-the-loop sim-to-real pipelines for dexterous and locomotion tasks — Eureka demonstrates Shadow Hand pen-spinning at human-comparable performance. Code and prompts released open-source on GitHub (NVlabs/Eureka). >800 citations within 18 months. Modern humanoid LLM-reward-design IP claims face this 2.5-year-deep anchor with full code disclosure.

LimX Dynamics CL-1 (2023-12)

id: limx-cl1
corpus: private
creator: LimX Dynamics
disclosure: LimX Dynamics public reveal, December 2023.
ip status: patented
prior art notes: LimX QDD actuation derives from MIT Cheetah lineage; bipedal control claims anticipated by Cassie/ATRIAS work.

1X NEO (2024)

id: 1x-neo
corpus: private
creator: 1X Technologies (formerly Halodi Robotics)
disclosure: 1X Technologies public reveal, 2024.
ip status: patented
prior art notes: Tendon-driven compliant actuation is heavily anticipated by iCub, by Shadow Robot Hand work, and by decades of academic compliant-actuation literature.

K-Scale Labs Open Source Humanoid (2024)

id: k-scale-os
corpus: open
creator: K-Scale Labs
disclosure: K-Scale Labs project launch, 2024.
ip status: open-permissive
prior art notes: Among the most ambitious recent fully-open humanoid efforts. Direct peer to Free Humanoid in scope.

Berkeley Humanoid (2024)

id: berkeley-humanoid
corpus: academic
creator: UC Berkeley, Hybrid Robotics Lab
disclosure: Liao, Q. et al. ‘Berkeley Humanoid: A Research Platform for Learning-based Control.’ arXiv 2024.
ip status: open-permissive
prior art notes: Berkeley quasi-direct-drive lineage (predates the humanoid; comes from the Mini Cheetah / leg work) anticipates many actuator architecture claims.

Persona AI Mentee (2024)

id: persona-ai-mentee
corpus: private
creator: Persona AI
disclosure: Persona AI public reveal, 2024.
ip status: trade-secret
prior art notes: Public technical disclosure is thin; strengthening pass needed.

ExBody whole-body humanoid policy (2024-02)

id: exbody-stanford-2024
corpus: academic
creator: UC San Diego + MIT + CMU; Xuxin Cheng, Yandong Ji, Junming Chen, Ruihan Yang, Ge Yang, Xiaolong Wang
disclosure: Cheng, X., Ji, Y., Chen, J., Yang, R., Yang, G., Wang, X. ‘Expressive Whole-Body Control for Humanoid Robots’. RSS 2024. arXiv:2402.16796. UC San Diego + MIT + CMU.
ip status: open-permissive
prior art notes: ExBody (Cheng et al. RSS 2024) is the canonical expressive whole-body humanoid policy paper. 1.5-year-deep open-permissive prior art. Companion to HumanPlus (round-27); both apply mocap-imitation RL to actual humanoid hardware (Unitree H1). Direct shielding for any commercial humanoid claim on expressive whole-body motion (dance, gestures).

Universal Manipulation Interface (UMI) (2024-02-15)

id: umi-stanford
corpus: academic
creator: Stanford + TRI + Columbia (Chi, Xu, Pan, Cousineau, Burchfiel, Feng, Tedrake, Song)
disclosure: Chi, Cheng, Xu, Zhenjia, Pan, Chuer, Cousineau, Eric, Burchfiel, Benjamin, Feng, Siyuan, Tedrake, Russ, Song, Shuran. ‘Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots.’ arXiv:2402.10329, February 15, 2024. Robotics: Science and Systems (RSS) 2024. Stanford University + Toyota Research Institute + Columbia University.
ip status: open-permissive
prior art notes: UMI is the canonical academic disclosure of embodiment-decoupled manipulation data collection via hand-held wrist-camera devices. Anticipates: (1) data collection with a portable hand-held gripper-replica without the robot present — directly relevant to claims on low-cost humanoid data collection (this paradigm is now used by Stanford ALOHA’s portable variants, Tesla operator-glove proposals, several other commercial programs); (2) wrist-camera SLAM as the substrate for trajectory reconstruction — relevant to vision-based teleoperation IP; (3) embodiment-matching gripper geometry between collection rig and deployment robot — relevant to claims on cross-embodiment manipulation training. Open-source hardware (3D print files), software, and data under permissive license. Modern humanoid ‘in-the-wild data’ patent claims face this 2-year-deep anchor with full DIY-buildable defensibility.

Covariant RFM-1 (2024-03)

id: covariant-rfm
corpus: private
creator: Covariant
disclosure: Covariant public reveal of RFM-1, March 2024.
ip status: trade-secret
prior art notes: Covariant RFM faces VLA prior art from Physical Intelligence π0 and the broader academic VLA literature (RT-2, Open X-Embodiment).

LeRobot (HuggingFace) (2024-03)

id: huggingface-lerobot-2024
corpus: academic
creator: Remi Cadene and contributors; HuggingFace, Inc. (with extensive academic contributions from Stanford, CMU, NYU, MIT, IIIT-Hyderabad, ETH Zurich research groups via upstream policies)
disclosure: Cadene, Remi et al. ‘LeRobot: State-of-the-art AI for real-world robotics in PyTorch.’ HuggingFace blog announcement and GitHub repository launch, March 13, 2024 (https://github.com/huggingface/lerobot). Cadene was previously a research engineer at Tesla AI / formerly at FAIR Paris before joining HuggingFace; the LeRobot framework consolidates open-source implementations of policies (ACT, Diffusion Policy, TDMPC, VQ-BeT, Pi0, SmolVLA) and datasets in a unified Apache-2.0 PyTorch substrate.
ip status: open-permissive
prior art notes: LeRobot (March 2024) is the canonical open-source unified framework for training and deploying imitation-learning and reinforcement-learning robot policies, published Apache-2.0 by HuggingFace. Anticipates with full architectural specificity: (1) multi-policy training and evaluation framework with a common interface — directly relevant to commercial claims on policy-architecture-agnostic VLA training pipelines (1X, Figure, Tesla Optimus, Genesis AI all build training pipelines that resemble this structure); (2) standardized dataset format for teleoperated demonstrations across heterogeneous embodiments (LeRobotDataset) — relevant to claims on cross-embodiment data unification, anticipating Open X-Embodiment-style aggregation patents; (3) the model-zoo pattern (pre-trained policy checkpoints downloadable via the HuggingFace Hub) — relevant to claims on commercial-grade pre-trained robot policy distribution; (4) real-robot inference on commodity hardware via PyTorch — relevant to claims on edge-deployable VLA systems. The Apache-2.0 license combined with extensive third-party contributions (Stanford Aloha team, Princeton Diffusion Policy, NYU/Cycle’s TDMPC2, Physical Intelligence Pi0) makes this entry the consolidated prior art anchor for the entire 2024-2026 VLA-training-stack patent space. Modern VLA pipeline IP filings face this 2-year-deep anchor with full source disclosure.

NVIDIA GR00T (Generalist Robot 00 Technology) (2024-03-18)

id: nvidia-groot-2024
corpus: academic
creator: NVIDIA Research, GEAR Lab
disclosure: Huang, Jensen et al. NVIDIA GR00T announcement at GTC 2024 keynote, March 18, 2024. Technical disclosure: Reddit Project GR00T technical blog, March 2024. GR00T N1 paper published 2025-03 (arXiv:2503.14734).
ip status: open-permissive
prior art notes: NVIDIA GR00T’s 2024 disclosure is the canonical foundation-model-for-humanoids announcement. Anticipates: (1) dual-system fast/slow policy architecture for humanoid platforms — directly relevant to modern humanoid foundation-model IP (every major humanoid manufacturer is developing equivalent architectures); (2) cross-embodiment generalization across multiple humanoid platforms — relevant to platform-agnostic policy IP; (3) open-weights humanoid foundation model release — provides defensive baseline against closed-weights claims. The March 2024 GTC keynote announcement plus the subsequent GR00T N1 paper (March 2025) and open-weights release provide extensive prior art coverage.

DeepMind humanoid soccer (Haarnoja et al.) (2024-04)

id: deepmind-humanoid-soccer-haarnoja-2024
corpus: academic
creator: Google DeepMind; Tuomas Haarnoja, Yuval Tassa, Nicolas Heess + ~25 co-authors
disclosure: Haarnoja, T., Moran, B., Lever, G., Huang, S. H., Tirumala, D., Humplik, J., Wulfmeier, M., Tunyasuvunakool, S., Siegel, N. Y., Hafner, R., Bloesch, M., Hartikainen, K., Byravan, A., Hasenclever, L., Tassa, Y., Sadeghi, F., Batchelor, N., Casarini, F., Saliceti, S., Game, C., Sreendra, N., Patel, K., Gwira, M., Huber, A., Hurley, N., Nori, F., Hadsell, R., Heess, N. ‘Learning agile soccer skills for a bipedal robot with deep reinforcement learning’. Science Robotics 9(89) April 2024.
ip status: open-permissive
prior art notes: DeepMind humanoid soccer (Haarnoja et al. Science Robotics April 2024) is the canonical end-to-end deep-RL humanoid agility paper. 1-year-deep open-academic prior art for: zero-shot sim-to-real agile humanoid skills (kicking, defending, getting up), multi-agent self-play RL on humanoid hardware, teacher-student distillation for compact deployable policies. Direct shielding for any commercial humanoid claim on dynamic-skill RL training or sim-to-real agile-locomotion transfer. Together with Berkeley Humanoid (round-11), Berkeley Humanoid Lite (round-11), and ToddlerBot (round-11), establishes the open-academic agile-humanoid-RL substrate.

Unitree G1 (2024-05)

id: unitree-g1
corpus: private
creator: Unitree Robotics
disclosure: Unitree Robotics G1 reveal, May 2024.
ip status: patented
prior art notes: G1’s actuator IP largely anticipated by MIT Mini Cheetah QDD work and Honda harmonic drive prior art. The aggressive pricing represents the commodity-humanoid trajectory more than novel IP.

Neura 4NE-1 (2024-05)

id: neura-4ne1
corpus: private
creator: Neura Robotics
disclosure: Neura Robotics public reveal of 4NE-1, May 2024.
ip status: patented
prior art notes: Neura’s cognitive-AI claims overlap with academic VLA literature.

Octo (Open-Source Generalist Robot Policy) (2024-05-20)

id: octo-policy
corpus: academic
creator: Octo Model Team (Berkeley/Stanford/CMU/Google)
disclosure: Octo Model Team. ‘Octo: An Open-Source Generalist Robot Policy.’ arXiv:2405.12213, May 20, 2024. Robotics: Science and Systems (RSS) 2024. Authors: Ghosh, D., Walke, H.R., Pertsch, K., Black, K., Mees, O., Dasari, S., Hejna, J., Xu, C., Luo, J., Kreiman, T., Tan, Y., Sanketi, P., Vuong, Q., Xiao, T., Sadigh, D., Finn, C., Levine, S. (UC Berkeley + Stanford + CMU + Google).
ip status: open-permissive
prior art notes: Octo is the foundational fully-open-weights generalist policy for robotic manipulation, combining Open X-Embodiment-scale training with diffusion-action-head architecture. Anticipates: (1) the integration of diffusion-policy action heads into transformer-based VLAs — directly relevant to claims on hybrid transformer-diffusion humanoid policies (essentially every 2025+ humanoid policy stack); (2) input-flexible generalist policies that accept any subset of cameras + optional language — relevant to claims on ‘plug-and-play’ humanoid policies; (3) full open-source weights and training code — establishes a defensive-publication baseline for the entire VLA design space. Code, weights, and training data fully released under Apache 2.0 / permissive licenses. Heavily cited within 18 months. Modern claims on transformer+diffusion-head humanoid policies face this 2-year-deep 102 anchor.

HumanPlus humanoid (2024-06)

id: humanplus-stanford-2024
corpus: academic
creator: Stanford University; Zipeng Fu, Qingqing Zhao, Qi Wu, Gordon Wetzstein, Chelsea Finn
disclosure: Fu, Z., Zhao, Q., Wu, Q., Wetzstein, G., Finn, C. ‘HumanPlus: Humanoid Shadowing and Imitation from Humans’. CoRL 2024. arXiv:2406.10454. Stanford University.
ip status: open-permissive
prior art notes: HumanPlus (Fu et al. CoRL 2024) is the canonical Stanford humanoid-imitation-from-humans paper. 1-year-deep open-permissive prior art for: two-stage RL-shadowing + IL fine-tuning, real-hardware humanoid full-body imitation from human motion. Direct architectural application of AMP/ASE lineage (rounds 21+27) to actual humanoid hardware. Direct shielding for any commercial humanoid claim on ‘humanoid imitates humans’ or ‘mocap-trained humanoid policy on real hardware’.

OpenVLA (Open-Source Vision-Language-Action Model) (2024-06-13)

id: openvla
corpus: academic
creator: Stanford + UC Berkeley + Toyota Research Institute + Google DeepMind + Physical Intelligence + MIT (Kim, Pertsch, Karamcheti, Xiao, Balakrishna, Nair, Rafailov, Foster, Lam, Sanketi, Vuong, Kollar, Burchfiel, Tedrake, Sadigh, Levine, Liang, Finn)
disclosure: Kim, Moo Jin, Pertsch, Karl, Karamcheti, Siddharth, Xiao, Ted, Balakrishna, Ashwin, Nair, Suraj, Rafailov, Rafael, Foster, Ethan, Lam, Grace, Sanketi, Pannag, Vuong, Quan, Kollar, Thomas, Burchfiel, Benjamin, Tedrake, Russ, Sadigh, Dorsa, Levine, Sergey, Liang, Percy, Finn, Chelsea. ‘OpenVLA: An Open-Source Vision-Language-Action Model.’ arXiv:2406.09246, June 13, 2024. Conference on Robot Learning (CoRL) 2024.
ip status: open-permissive
prior art notes: OpenVLA is the canonical fully-open-weights vision-language-action model for robotic manipulation, establishing a clean defensive-publication baseline for 7B-class humanoid VLAs. Anticipates: (1) the action-tokenization-via-vocab-overwrite scheme for adding action heads to pretrained LLMs — directly relevant to claims on humanoid VLAs that piggyback on existing language-model vocabularies; (2) LoRA-based fine-tuning for fast adaptation to new robots/tasks — relevant to claims on efficient humanoid policy adaptation; (3) the demonstrated parameter-efficient outperformance of larger closed models — relevant to claims on ‘small-but-capable’ humanoid VLA architectures. Apache 2.0 license; weights, code, and data fully released. Modern humanoid VLA filings face this anchor at <2 years’ depth, with full source-code defensibility.

Kepler K2 (2024-07)

id: kepler-k2
corpus: private
creator: Kepler Exploration Robotics
disclosure: Kepler Exploration Robotics public reveal, July 2024.
ip status: patented
prior art notes: Kepler’s planetary-reducer actuator claims are anticipated by extensive prior art in industrial robotics planetary-gearing literature.

Skild AI foundation model (2024-07)

id: skild-foundation-model
corpus: private
creator: Skild AI
disclosure: Skild AI public emergence, July 2024.
ip status: trade-secret
prior art notes: Skild’s foundation-model approach faces extensive academic prior art including RT-X, Open X-Embodiment, and π0.

Berkeley Humanoid (2024-07)

id: berkeley-humanoid-2024
corpus: academic
creator: UC Berkeley Hybrid Robotics Lab; Liao, Zhang, X. Huang, X. Huang, Li, Sreenath
disclosure: Liao, Q., Zhang, B., Huang, X., Huang, X., Li, Z., Sreenath, K. ‘Berkeley Humanoid: A Research Platform for Learning-based Control’. arXiv:2407.21781, July 2024. IEEE International Conference on Robotics and Automation (ICRA) 2025. UC Berkeley Hybrid Robotics Lab.
ip status: open-permissive
prior art notes: Berkeley Humanoid is the open academic mid-scale bipedal humanoid research platform from the Sreenath group, ICRA 2025. Open-permissive. Establishes 1-year-deep prior art for: RL-trained locomotion with sim-to-real zero-shot transfer at humanoid scale, low-cost in-house-built humanoid for learning research, anthropomorphic kinematics optimized for sim-to-real. Direct shielding for free-humanoid-platform commitments on bipedal RL locomotion and any commercial humanoid claim on RL-trained outdoor walking. Parent of Berkeley Humanoid Lite (round-11 entry below).

Figure 02 (2024-08)

id: figure-02
corpus: private
creator: Figure AI
disclosure: Figure AI public reveal of Figure 02, August 2024.
ip status: patented
prior art notes: Figure 02 actuator and hand claims are heavily anticipated by Honda P-series, Robonaut 2, Shadow Hand, and iCub work. The 16-DoF hand is in the same design space as Robonaut 2’s 12-DoF and Sanctuary’s 21-DoF.

MaskedMimic (2024-09)

id: maskedmimic-tessler-stanford-2024
corpus: academic
creator: NVIDIA Research + Stanford; Chen Tessler, Yunrong Guo, Ofir Nabati, Gal Chechik, Xue Bin Peng
disclosure: Tessler, C., Guo, Y., Nabati, O., Chechik, G., Peng, X. B. ‘MaskedMimic: Unified Physics-Based Character Control Through Masked Motion Inpainting’. SIGGRAPH Asia 2024. arXiv:2409.14393. NVIDIA Research + Stanford.
ip status: open-permissive
prior art notes: MaskedMimic (Tessler + Peng SIGGRAPH Asia 2024) is the architectural successor to AMP + ASE for physics-based character control. 1-year-deep open-permissive prior art. Continues the DeepMimic 2018 → AMP 2021 → ASE 2022 → MaskedMimic 2024 chain. Direct shielding for any commercial humanoid claim on masked-token motion-inpainting or unified-conditioning character control.

Physical Intelligence π0 (2024-10)

id: physical-intelligence-pi-zero
corpus: private
creator: Physical Intelligence
disclosure: Black, K. et al. ‘π0: A Vision-Language-Action Flow Model for General Robot Control.’ arXiv 2410.24164, October 2024.
ip status: trade-secret
prior art notes: π0 paper discloses VLA-with-flow-matching architecture publicly. The arXiv preprint is itself prior art for that specific architectural pattern.

Robot Era STAR1 (2024-10)

id: robot-era-star1
corpus: private
creator: Robot Era
disclosure: Robot Era public reveal of STAR1, October 2024.
ip status: patented
prior art notes: Bipedal running speed claims anticipated by Cassie’s Guinness record work.

XPeng Iron (2024-11)

id: xpeng-iron
corpus: private
creator: XPeng Motors (Robotics division)
disclosure: XPeng AeroHT and XPeng Robotics reveal, November 2024.
ip status: patented
prior art notes: XPeng’s leveraging of automotive ML stack for humanoid perception is heavily anticipated by Tesla Optimus’s same approach (which is itself anticipated by academic vision-based humanoid work).

EngineAI PM01 (2024-12)

id: engineai-pm01
corpus: private
creator: EngineAI
disclosure: EngineAI public reveal of PM01, December 2024.
ip status: patented
prior art notes: EngineAI QDD actuation anticipated by MIT Cheetah lineage.

Genesis (open-source physics simulator) (2024-12)

id: genesis-embodied-ai-simulator
corpus: open
creator: Genesis Authors collaboration (multi-institution: CMU, Stanford, MIT CSAIL, Tsinghua, Peking, ETH Zürich, UMD, et al.)
disclosure: Genesis Authors. ‘Genesis: A Generative and Universal Physics Engine for Robotics and Beyond’. GitHub release at https://github.com/Genesis-Embodied-AI/Genesis, December 19, 2024. Multi-institution collaboration including Carnegie Mellon University, Stanford University, MIT CSAIL, Tsinghua University, Peking University, ETH Zürich, University of Maryland.
ip status: open-permissive
prior art notes: The Genesis simulator (Genesis-Embodied-AI/Genesis, December 2024) is the most recent and highest-throughput academic-grade open-source physics engine for robotics simulation, published Apache-2.0 by a multi-institution academic collaboration. Anticipates with full architectural specificity: (1) GPU-parallelized robotics simulation at 43M-FPS scale — directly relevant to commercial claims on sim-to-real-at-scale humanoid IP (notably Genesis AI Inc.’s GENE-26.5 product, with which this open-source project shares a name); (2) unified multi-physics architecture (rigid + soft + MPM + FEM + fluid) — relevant to claims on multi-domain humanoid simulation; (3) differentiable simulation for gradient-based policy optimization — relevant to claims on policy-gradient humanoid training at scale; (4) the URDF/MJCF interoperability surface that permits OpenLoco-class descriptors to be simulated without modification. Modern claims on sim-to-real-at-scale, multi-physics simulation, or differentiable physics for humanoid training all face this 1.5-year-deep open-source academic prior art with full source disclosure under Apache-2.0.

ToddlerBot (2025-02)

id: stanford-toddlerbot-2025
corpus: academic
creator: Stanford Robotics Lab; Haochen Shi, Weizhuo Wang, Shuran Song, C. Karen Liu
disclosure: Shi, H., Wang, W., Song, S., Liu, C. K. ‘ToddlerBot: Open-Source ML-Compatible Humanoid Platform for Loco-Manipulation’. arXiv:2502.00893, February 2025. Conference on Robot Learning (CoRL) 2025 oral. Stanford Robotics Lab.
ip status: open-permissive
prior art notes: ToddlerBot is Stanford’s canonical sub-$6k open-hardware ML-compatible humanoid (CoRL 2025 oral). Establishes 1-year-deep open-academic prior art for: integrated loco-manipulation policy training on an open humanoid platform, transferable motor system-ID for sim-to-real without hand-tuning, 30-DoF anthropomorphic full-body at sub-$6k. Direct shielding for any commercial claim on integrated full-body humanoid policy training, particularly any ‘one policy controls the whole body’ claim. Together with Berkeley Humanoid Lite, establishes the open-academic baseline for sub-$10k humanoid robotics.

Booster K1 (2025-03)

id: booster-k1-2025
corpus: private
creator: Booster Robotics (Beijing, China)
disclosure: Booster Robotics. K1 product page (booster.tech/booster-k1) and associated commercial brochures, public 2025+. RoboCup 2025 KidSize humanoid league winning platform (Boosted HTWK team, Salvador Brazil, July 20 2025).
ip status: trade-secret
prior art notes: Booster K1 is the canonical 2025 sub-$25k educational humanoid. 5-month-deep public-disclosure prior art for: KidSize-class (95cm) humanoid form factor, 22-DoF anthropomorphic kinematics, ROS 2 + Python developer-friendly stack at the educational price point. Public competition record (RoboCup 2025 KidSize win) demonstrates a working system. Direct shielding for any commercial humanoid claim on educational/sub-$25k pricing or RoboCup-competition-grade autonomous bipedal locomotion.

Berkeley Humanoid Lite (2025-04)

id: berkeley-humanoid-lite-2025
corpus: academic
creator: UC Berkeley Hybrid Robotics Lab; Sreenath group
disclosure: Cui, F., Sayle, J., Karydis, K., Liao, Q., et al. ‘Demonstrating Berkeley Humanoid Lite: An Open-source, Accessible, and Customizable 3D-printed Humanoid Robot’. arXiv:2504.17249, April 2025. Robotics: Science and Systems (RSS) 2025. UC Berkeley Hybrid Robotics Lab.
ip status: open-permissive
prior art notes: Berkeley Humanoid Lite is the canonical sub-$5k open-hardware academic bipedal humanoid (RSS 2025). 1-year-deep prior art on: 3D-printed cycloidal reducer humanoid actuator (a specific architectural commitment), full open-source release of hardware + firmware + training, sub-$5k humanoid BOM, RL-controlled walking on a 3D-printed platform. Direct shielding for free-humanoid-platform — particularly the 3D-printed actuator path and any commercial claim on accessible humanoid hardware. Together with ToddlerBot and Berkeley Humanoid (full-size), establishes a deep open-academic substrate for any commercial humanoid platform claim.

Unitree R1 (2025-07)

id: unitree-r1-2025
corpus: private
creator: Unitree Robotics (Hangzhou, China; founded 2016 by Wang Xingxing)
disclosure: Unitree Robotics (Hangzhou, China). R1 product reveal July 2025; global launch April 2026 via shop.unitree.com / AliExpress. unitree.com/R1. Multi-tier product line: R1 Air $4.9k, R1 Basic $5.9k-$8.99k, R1 EDU Standard $10-12k, R1 EDU Smart $15-19k, R1 EDU Pro $20-35k.
ip status: trade-secret
prior art notes: Unitree R1 is the canonical 2025+ low-cost consumer humanoid (Unitree Robotics, China). ~10-month-deep public-disclosure prior art at time of corpus entry. Significantly disrupts the humanoid pricing claim space — drops the entry price from Boston Dynamics Atlas (>$1M) / Figure 02 ($15k+) / Optimus Gen 3 ($20-30k target) to $4,900. Establishes 9 km/h running + cartwheels as commercially-deployed-not-academic capabilities. Direct shielding for any commercial humanoid claim on consumer-tier pricing or low-cost humanoid morphology.

Booster T1 (2025-09)

id: booster-t1-2025
corpus: private
creator: Booster Robotics (Beijing, China)
disclosure: Booster Robotics. T1 product reveal Q3 2025 via booster.tech. Successor to K1 (round-16 entry booster-k1-2025) with adult-class form factor.
ip status: trade-secret
prior art notes: Booster T1 is Booster Robotics’ adult-class commercial humanoid (2025+). 8-month-deep public-disclosure prior art at the time of this corpus entry. Inherits from K1 (round-16) the ROS 2 + Python developer-friendly stack pattern. Direct shielding for Booster’s commercial product line as a coherent multi-platform humanoid family (KidSize K1 + AdultSize T1).

EngineAI SE01 (2025-09)

id: engineai-se01-2025
corpus: private
creator: EngineAI Robotics (Shenzhen, China)
disclosure: EngineAI Robotics (Shenzhen, China). SE01 product reveal Q3 2025 via engineai.com. Successor to PM01 (corpus entry engineai-pm01). Adult-class commercial humanoid at the sub-$30k tier.
ip status: trade-secret
prior art notes: EngineAI SE01 is EngineAI’s adult-class commercial humanoid (2025+). Successor in the EngineAI product line after PM01 (round-9 entry engineai-pm01). Direct shielding for any commercial claim on the EngineAI multi-platform humanoid family. Together with Unitree R1, Astribot S1, Booster T1, Galbot, establishes the 2024-2026 Chinese commercial humanoid landscape.

Unitree H2 (2025-10)

id: unitree-h2
corpus: private
creator: Unitree Robotics
disclosure: Unitree Robotics H2 reveal, October 2025.
ip status: patented
prior art notes: H2 builds on H1 architecture; same prior art chain back through Mini Cheetah.

Genesis AI GENE-26.5 (2026-04)

id: genesis-ai-gene-26-5
corpus: private
creator: Genesis AI Inc.
disclosure: Genesis AI Inc. corporate website at https://www.genesis.ai/, GENE-26.5 product page (April 2026 surface). Demo videos showing cooking, lab pipetting, beverage preparation, puzzle-solving, object manipulation, assembly, and fine-motor tasks.
ip status: trade-secret
prior art notes: Genesis AI’s GENE-26.5 platform is a closed-source commercial robotics product whose public disclosure surface (corporate website + demo videos) does not reveal specific mechanism. The capability set claimed — multi-task vision-language-action manipulation, sim-to-real generalization, dexterous fine-motor — is fully covered by deep open academic prior art chains in the corpus: Pomerleau ALVINN (1989) → Levine GPS PR2/BRETT (2016) for end-to-end visuomotor policy; CLIP (Radford 2021) for vision-language alignment; RT-1 (2022), RT-2 (2023), Open X-Embodiment (2023), OpenVLA (2024), π-zero (2024), NVIDIA GR00T N1 (2025) for VLA architecture; OpenAI Dactyl (2018-2019), Hwangbo ANYmal sim-to-real (2019), Tan quadruped sim-to-real (2018) for sim-to-real; Mobile ALOHA (2024), ACT/ALOHA (2023), Diffusion Policy (2023) for bimanual fine manipulation; Salisbury Stanford-JPL hand (1982), DLR Hand-II (2001), Shadow Hand (2002), Pisa-IIT SoftHand (2014) for dexterous hand mechanism; Park’s transformation (1929) for any FOC actuator control. Claims that GENE represents novel art in any of these subsystems face element-by-element prior art at depths from 4 years (Diffusion Policy) to 97 years (Park) to 530 years (Da Vinci’s Knight, anthropomorphic tendon-driven hand). Demo task set (cooking, beverage preparation, lab manipulation) maps directly to Mobile ALOHA (2024), DLR Justin (2009), CMU HERB (2012), and the PR2 lineage.

Cross-cut: control-rl-policy