Free Humanoid Submersible — an open hardware, open firmware, open data humanoid AUV reference design. Fourth morphology in the OpenIE humanoid family alongside the bipedal Free Humanoid Platform, the wheeled Free Humanoid Wheeled, and the centaur Free Humanoid Centaur. Designed for the shoal dock-A subsurface service mission and the underwater bimanual-manipulation deployment band more broadly. Shielded by the Free Humanoid Corpus.
A ~80 kg-in-air humanoid AUV with a streamlined 8-thruster vectored pod replacing the bipedal platform’s legs, the wheeled sibling’s wheeled base, and the centaur’s quadruped base. Negatively buoyant by ~2 kg and thrust-active for full 6-DoF station-keeping.
Phase 1 status (2026-05-07): descriptor v0.1.0 vendor-pinned to commercial-off-the-shelf components (Blue Robotics T200 thrusters, Bar100 depth, BR pressure-hull series; Cerulean DVL-75; Bluefin BPS battery or DIY Li-ion alternative). Hull + thruster pod BOM is ~$22-46k depending on production-vs-budget alternates. Phase 2 work begins with pressure-hull fabrication + thruster-pod prototype + dry-side bench test. Pool trial Phase 3, open-water shoal dock-A trial Phase 4. Expressed as a single OpenLoco UDD descriptor that compiles to URDF, MJCF, STL meshes, BOM, and assembly through the existing OpenLoco toolchain. Every non-trivial design choice cites a specific entry in the Free Humanoid Corpus by id, with aquanaut-houston-2017 and oceanone-stanford-2016 flagged as recommended-future-corpus-additions for the humanoid-AUV anchors.
This document is the canonical full-system spec for the submersible morphology. It cross-references — and does not duplicate — the bipedal sibling’s ARCHITECTURE.md for everything inherited (upper body, manipulation, learning policy above the locomotion layer) and the centaur’s ARCHITECTURE.md for the descriptor-substitution-of-lower-body pattern. Decisions that diverge from any sibling are argued in full here.
The OpenIE Free Humanoid Family is descriptor-driven: a single OpenLoco UDD descriptor describes the upper body once and is composed with a per-morphology lower body to instantiate a specific platform. The four morphologies are:
| Morphology | Locomotion | Operating medium | Mission band |
|---|---|---|---|
| Free Humanoid Platform (bipedal) | 2 legs (12 DoF) | Air, hard ground | Indoor service, stair-rich environments |
| Free Humanoid Wheeled | 2 drive wheels + 2 casters + tower lift | Air, paved ground | Shoal dock-B paved-curb-edge service |
| Free Humanoid Centaur | 4 wheel-leg-hybrid feet (12 DoF) | Air, rough ground / wetlands | Shoal dock-A wetlands surface service |
| Free Humanoid Submersible (this morphology) | 8 vectored thrusters | Submerged water | Shoal dock-A subsurface service, underwater bimanual manipulation |
The descriptor-share thesis: one upper body, four lower bodies. Reusing the bipedal platform’s 21-DoF upper body — torso, arms, hands, head, neck, sensing-of-the-upper-body, manipulation policy, safety supervisor’s manipulation-layer invariants — across all four morphologies means the same descriptor, same hand, same imitation-learning manipulation policy, and same wrist + tactile fingertip stack work in air on hard ground, on paved ground, on wetland substrate, and below water. The submersible is the most extreme test of this thesis: same arms, same hands, completely different lower body and operating medium.
Shoal docks (per shoal ARCHITECTURE) include benthic and partially-submerged maintenance surfaces that have to be serviced underwater:
These surfaces share a property: they require bimanual dexterous manipulation underwater. A standard ROV (single arm, parallel-jaw or tooled gripper) is inadequate. The same use-case argued for a humanoid centaur above the water — bimanual dexterous manipulation in a service-task envelope — argues for a humanoid AUV below it.
Free Humanoid Submersible is the subsurface complement of the centaur: the centaur services the surface-side of dock-A, the submersible services the subsurface-side. The two morphologies share the upper body, the manipulation stack, and the cartridge-protocol-style service vocabulary, and differ only at the lower body.
The submersible descriptor descriptor/free-humanoid-submersible.udd.json is OpenLoco UDD-compliant and composes with the bipedal sibling’s upper-body descriptor. Specifically:
| Descriptor block | Source | Treatment in the submersible |
|---|---|---|
meta |
new for this morphology | morphology_class: "humanoid_submersible"; inherits_upper_body_from: "free-humanoid-platform" |
links (upper body — torso, arms, hands, head, neck) |
inherited verbatim from free-humanoid.udd.json |
tagged _metadata.inherited_from: "free-humanoid-platform" per link |
links (lower body — pressure hull, thruster pod, ballast tank, comms module) |
new for this morphology | replaces the bipedal legs block, the wheeled wheeled_base + tower_stand block, and the centaur quadruped_base block |
joints (upper body) |
inherited verbatim | parented to a new pressure_hull_top_to_pelvis fixed joint instead of the bipedal pelvis free root |
joints (lower body) |
new | 8 thruster mounts (fixed) + 1 pressure-hull-to-pelvis fixed + 1 ballast-tank fixed + 1 comms-module fixed |
actuator_slots (upper body — 19 slots) |
inherited verbatim | tagged _metadata.inherited_from: "free-humanoid-platform" per slot |
actuator_slots (lower body — 8 slots) |
new | one per thruster, modeled as continuous_rotation actuators producing thrust force + reaction torque on the platform body |
sensors (lower body) |
new | depth (pressure), DVL (Doppler velocity log), USBL (acoustic positioning), multibeam imaging sonar, downward-looking camera array, body IMU (re-purposed at the pressure-hull frame) |
modes |
new | safe_pose_passive_descent, station_keeping_hover, transit_thrust_forward, manipulation_static_hover, surfacing |
The descriptor differentiates from siblings at:
Topology: 8 thrusters in an ROV-class layout that gives full 6-DoF (surge, sway, heave, roll, pitch, yaw) station-keeping authority:
Why 8 and not 6: a 6-thruster vectored layout (canonical Aquanaut configuration) is leaner but couples sway with yaw at the corner-mounting geometry; the 8-thruster layout decouples translation from rotation in all 6 DoF and gives much higher station-keeping authority for one-handed-while-the-other-hand-works manipulation tasks. The 8-thruster pattern is the canonical Bluefin / REMUS / SAAB Sabertooth ROV-class layout.
Thruster choice: brushless DC subsea thrusters in the Blue Robotics T200 / T500 class, ~5 kgf thrust each at full power. Total instantaneous thrust authority: ~40 kgf. Continuous duty cycle: ~50% at full thrust (limited by motor heat dissipation in water — water is a better heat sink than air, but the motors are still rated). Each thruster runs on its own ESC (electronic speed controller) on a CAN-FD bus shared with the 16-Moteus inheritance from the centaur.
Battery + electronics in the pressure hull. The submersible has one pressure hull; the thrusters live outside it (water-cooled, oil-filled or pressure-balanced). All compute, the safety supervisor, the battery, the upper-body actuator drivers, and the comms electronics live inside the pressure hull. Single point of failure on the hull seal, which is why the seal architecture (§5) is conservative.
Ballast trim. A small variable-buoyancy ballast tank (~1.5 L water displacement) sits at the centerline of the thruster pod. It is not a primary depth-control mechanism — that’s the vertical thrusters — but it allows the supervisor to trim the platform to neutral or slightly-negative buoyancy for the operating depth and the freshwater/saltwater density of the deployment site. TBD (architectural call): active variable-buoyancy via syringe pump vs. fixed-trim with selectable ballast plates. Active is more flexible; fixed-trim is more reliable. Recommendation: fixed-trim plates for v0.2; active variable-buoyancy held as a future descriptor variant.
Design depth: 50 m at v0.2 (Phase 2). Shoal dock-A subsurface depths are typically < 30 m (riverbed, marine shore, shallow wetlands); the 50 m design margin gives 1.7× safety factor at the deployment band. This is well within the routine-engineering envelope of every commercial AUV ever (Aquanaut, Bluefin, REMUS, SAAB Sabertooth all rate to 200–600 m); the submersible is intentionally shallower than oceanographic platforms because the shoal mission band is shallower.
Material: 6061-T6 aluminum cylinder with hard-anodized exterior and Delrin (acetal) endcaps, double-O-ring face-seal at each endcap. The 6061-T6 + double-O-ring + Delrin-endcap pattern is canonical commercial AUV practice and is shielded by every commercial AUV in the corpus (and prior to it).
Penetrators. 14 connectors through the rear endcap (8 thrusters + DVL + USBL + sonar + camera-array + tether + charging-and-data); 2 through the front endcap (forward camera, depth sensor); standard subsea wet-mateable connectors (Subconn, Birns, or compatible). All pressure-tested at 1.5× design depth.
Maintenance access. The forward endcap is removable for battery and compute access; the rear endcap is fixed (penetrator-loaded). Hot-swap of the battery cassettes is therefore a dry-side operation only — there is no underwater hot-swap. The mating geometry inherits the wheeled sibling’s commitment #4 form factor for tooling compatibility, but the cells inside the cassette are subsea-rated.
Pressure-cycle life. 1000 cycles at 50 m, then re-certify the seals. This is conservative versus published AUV data and intentionally so for an open-design platform whose failure mode (flood-out) is total-loss-of-platform.
This is the morphology where the fluid-dynamics simulation matters more than for any of the siblings, because the operating medium is the load on the platform.
Genesis is the primary simulator for hydrodynamic-loaded scenarios. Genesis’s MPM (material-point-method) and fluid backends are the right tools for modeling thruster-induced flow, drag-coupled manipulation forces (the platform pushes water as it moves an arm), wake interaction at the dock, and the partially-submerged contact regime around the dock’s intake screen. Genesis is GPU-parallelized, differentiable, URDF/MJCF interoperable, and Apache-2.0. Corpus citation: genesis-embodied-ai-simulator. Genesis is more relevant here than it is for the centaur, where fluid simulation is a wetland-substrate edge case rather than the core regime.
MuJoCo is secondary for rigid-body kinematics validation. The arms, hands, and pressure-hull-as-rigid-body station-keeping problem decompose nicely into a rigid-body model with externally-applied thrust forces and added-mass terms; MuJoCo handles this well at 500 Hz cycle rates and is the right environment for the upper-body manipulation stack inherited from the bipedal sibling. MuJoCo’s contact model is far better than Genesis’s for tactile-fingertip work.
Stonefish (open-source AUV simulator, Gazebo-compatible, GPL) is the reference simulator for subsea-platform-specific scenarios: through-water acoustic comms simulation, DVL/USBL noise modeling, and multibeam-imaging-sonar artifacts. Used as a tertiary simulator for sensor-stack validation. Citation in shoal ARCHITECTURE.
Drake / Gazebo / Isaac are tertiary for this morphology. The submersible’s hydrodynamic regime is the load-bearing simulation problem and is better served by Genesis’s fluid stack than by any of the rigid-body simulators that dominate the air-side morphologies.
Compile path: descriptor/free-humanoid-submersible.udd.json → OpenLoco compiler → MJCF (MuJoCo / Genesis-rigid) + URDF (Gazebo / Stonefish) + STL meshes. The same descriptor feeds all simulators; the differentiator is which simulator’s solver is run.
humanoid_submersible morphology.Same posture as the three sibling morphologies:
| Artifact class | License | Files |
|---|---|---|
| Hardware (CAD, schematics, PCB layouts, pressure-hull and thruster-pod mechanical designs) | CERN-OHL-S 2.0 | LICENSE-HARDWARE |
| Firmware, control software, simulation, tools | Apache 2.0 | LICENSE-SOFTWARE |
| Documentation | CC-BY-SA 4.0 | LICENSE-DOCS |
| Descriptors (UDD JSON), datasets, BOM data | CC0 1.0 | LICENSE-DATA |
CC0 on the descriptor is load-bearing — it is the artifact that, when shared, propagates the prior-art shielding most aggressively. Inlining the canonical license text is a release-blocker before any public push.
Eight load-bearing decisions for this morphology, resolved. Each commitment is shielded by the cited prior-art chain. Where the chain calls for a corpus entry that does not yet exist (the humanoid-AUV anchors aquanaut-houston-2017 and oceanone-stanford-2016), it is flagged as a recommended-future-corpus-addition; the decision is held-shielded by the public publication of the prior-art platform itself pending the formal corpus entry.
| # | Decision | Commitment | Shielding chain (corpus ids; recommended-future ids in italics) |
|---|---|---|---|
| 1 | Inherit upper body from bipedal sibling verbatim | 21-DoF upper body — 1-DoF waist + 2-DoF neck + 2× 7-DoF arms + 2× 1-DoF underactuated synergy hands. Hybrid actuator distribution: harmonic-drive at shoulder/elbow, tendon at wrist/hand, QDD at neck/waist. Wrist seals are submerged-rated; otherwise unchanged. | Bipedal sibling’s §9 commitments #1, #4, #8, #10. The inheritance is governance-load-bearing: a divergence in the upper body would fork the descriptor and the corpus citations. The wheeled and centaur siblings are the first two worked examples of upper-body inheritance; the submersible is the third — and the most extreme, since the operating medium is different. Plus the recommended-future-corpus-additions aquanaut-houston-2017 and oceanone-stanford-2016 for the humanoid-AUV anchors that establish humanoid arms on a free-floating subsea base as published prior art. |
| 2 | 8-thruster vectored layout (4 horizontal vectored at ±45° + 4 vertical) | Full 6-DoF station-keeping authority. ROV-class layout, not the leaner 6-thruster Aquanaut layout. Decouples translation from rotation in all 6 DoF for one-handed-while-the-other-hand-works manipulation. | The Bluefin / REMUS / SAAB Sabertooth ROV lineage is the canonical 8-thruster anchor; recommended-future-corpus-additions bluefin-21-auv, kongsberg-remus-100-auv, saab-seaeye-falcon-rov. Plus the historical HOV / observation-class lineage jason-rov-1989 and robin-hov for the subsea-thruster-equipped service-platform pattern. The 6-thruster Aquanaut layout (recommended-future aquanaut-houston-2017) is acknowledged as a leaner alternative but the 8-thruster pattern shields a stronger station-keeping claim. |
| 3 | Negatively-buoyant + thrust-active station-keeping (descend on power-off, ascend by thrust) | ~+2 kg net negative buoyancy in fresh water. Vertical thrusters maintain depth; loss of thrust → controlled descent → controlled bottom contact. Inverse of the Aquanaut “neutral-buoyancy + minimum-thrust” pattern; matches the conservative ROV/HOV pattern. | The ROV / HOV literature back to robin-hov (the 1960s manned observation HOV that established the negatively-buoyant subsea-service pattern) and jason-rov-1989 (Woods Hole’s tethered ROV, the canonical academic ROV anchor). The negatively-buoyant + thrust-active pattern is the dominant ROV pattern; the alternative (neutrally-buoyant + minimum-thrust, Aquanaut-style) is a more efficient but less-fail-safe pattern that requires active depth control even in idle. |
| 4 | 6061-T6 aluminum pressure hull with Delrin endcaps + double-O-ring seals, 50 m design depth | Single-cylinder pressure hull, hard-anodized 6061-T6 aluminum, Delrin endcaps, face-seal with double-O-rings. Penetrators rated 1.5× design depth. 1000-cycle seal recertification interval. | The entire commercial AUV pressure-hull lineage: recommended-future-corpus-additions bluefin-21-auv, kongsberg-remus-100-auv, saab-seaeye-falcon-rov; plus the academic-anchor jason-rov-1989. 6061-T6 + double-O-ring + Delrin-endcap is canonical practice with > 30 years of commercial deployment; this is shielded by literally every commercial AUV ever. |
| 5 | Subsea Li-ion battery (pressure-balanced / oil-filled) + dry-side hot-swap | 2 kWh subsea Li-ion in cassette form factor inheriting the wheeled sibling’s commitment #4 mating geometry. Hot-swap is a dry-side operation only; there is no underwater hot-swap. | Wheeled sibling’s §9 commitment #4 for the mating geometry. Centaur’s §9 commitment #5 for the multi-module-distribution pattern. The subsea-Li-ion pressure-balanced pattern is canonical commercial-AUV practice (recommended-future-corpus-additions bluefin-21-auv, kongsberg-remus-100-auv). Plus spot-fuel-cell for the extended-endurance variant if added later. |
| 6 | Genesis MPM/fluid as the primary simulator; MuJoCo for rigid-body kinematics validation; Stonefish for sensor-stack validation | Genesis is the load-bearing simulator for hydrodynamic-loaded scenarios (thruster wake, drag-coupled manipulation, partially-submerged contact). MuJoCo is the load-bearing simulator for upper-body manipulation. Stonefish is the load-bearing simulator for through-water acoustic comms and DVL/USBL/sonar sensor stacks. | genesis-embodied-ai-simulator (the canonical academic disclosure of unified-multi-physics differentiable simulation, December 2024). Plus recommended-future-corpus-addition stonefish-sim-2018 for the AUV-specific simulator. MuJoCo is in the corpus already at the bipedal sibling’s chain. |
| 7 | Tendon-driven underactuated 5-finger hand, operated submerged | Same hand as the bipedal sibling, with submerged-rated tendon channels and hydrolytically-stable cable material. The synergy progress DoF [0,1] semantics are unchanged; only the cable jacket and the wrist-seal-pass-through are submerged-rated. | The bipedal sibling’s hand commitment chain (pisa-iit-softhand, shadow-dexterous-hand, dlr-hand-ii); plus the recommended-future-corpus-additions for underwater-manipulation prior art (oceanone-stanford-2016 for the humanoid-AUV anthropomorphic-hand instance, plus general subsea-manipulator prior art). The underactuated synergy hand is mechanically simpler than a fully-actuated hand and is therefore easier to seal for submerged operation than a fully-actuated alternative would be — the synergy commitment is a hand-design decision that pays dividends in this morphology. |
| 8 | JANUS acoustic + Ethernet tether comms (no WiFi/5G/WebRTC underwater) | Through-water comms is JANUS / STANAG 4748 (the lingua franca per shoal ARCHITECTURE §4). Tether is the high-bandwidth fallback for dock-coupled or pool-trial operations. WiFi / 5G / WebRTC are not viable underwater and are not present. | shoal ARCHITECTURE §4 for the JANUS / STANAG 4748 commitment (already canonical in the shoal architecture, which uses it for the fish ↔ dock comms). Plus the unetstack-subnero-2014 recommended-future-corpus-addition for the open-source software-defined underwater networking stack. Tethered Ethernet is canonical commercial-ROV practice. |
These commitments are this document’s defaults. Subsystem sections (§§3–6) reflect them in detail. Future amendments require a corpus-citation update demonstrating that any new shielding chain is at least as deep as the chain it replaces.
Recommended-future-corpus-additions (called out in docs/):
aquanaut-houston-2017 — Houston Mechatronics’ humanoid-arm AUV (the strongest commercial humanoid-AUV anchor)oceanone-stanford-2016 / ocean-onek-stanford-2022 — Khatib lab’s bimanual humanoid teleop AUV (the strongest academic anchor)bluefin-21-auv, kongsberg-remus-100-auv, saab-seaeye-falcon-rov — non-humanoid commercial AUV substrate prior art (8-thruster layouts, pressure hulls, subsea Li-ion)jason-rov-1989, robin-hov, alvin-hov-1964 — academic ROV/HOV lineageunetstack-subnero-2014 — open-source underwater networking stackstonefish-sim-2018 — open-source AUV simulatorFree Humanoid Submersible — fourth morphology in the OpenIE humanoid family — designed for the shoal dock-A subsurface service mission and the underwater bimanual-manipulation deployment band — Phase 1 v0.1 — vendor-pinned descriptor + BOM, 2026-05-07.