Add robot profile: unitree/go2/unitree.go2.mujoco-pybullet-sim.v1

[EslaM-X]

🤖 Unitree Go2 · Tier 1 Simulator Submission

Profile: unitree/go2/unitree.go2.mujoco-pybullet-sim.v1
Scope: Simulator-only · 8 priced skills · MuJoCo + PyBullet sim-to-sim

A paid RoboPay action hitting the Zenoh topic robot/tunnel/action drives
the official Go2 model from mujoco_menagerie through real,
policy-triggered skill episodes — joint-space PD trajectory control, body
weight compensated, never a replayed animation or a built-in demo motion.

---

🎯 What's in the box

Path	Purpose
registry/vendors/unitree/go2/unitree.go2.mujoco-pybullet-sim.v1/	Full robot action profile (robot · skills · functions · payment policy · execution mapping · example envelope · skill-contract tests · validation report)
simulation/go2/	Go2 controller, x402 payment gate, Zenoh action link, 5 test suites
simulation/pybullet/	Sim-to-sim validation + machine-readable report
simulation/setup.sh	Pinned, reproducible fetch of the official Go2 model assets

🦿 The 8 skills (all priced, all measurable)

wave · sit · stand · stop (safe stop) · bow · nod · turn_to_face · hold

Every skill runs through a joint-space trajectory controller and reports
real simulator metrics — paw lift height, sit depth, torso pitch, body
height, achieved heading — then returns to the home stance
(|bodyZ − 0.283| < 0.02) so paid actions can run back to back. stop is the
fail-safe skill: it halts motion and returns the robot to its stable home pose.

⚙️ The Go2 specifics (what's not a copy of the Spot demo)

• Go2 exposes torque motor actuators (no native position actuators), so
  the controller wraps them in a small PD position servo — while keeping
  the torque limits in the model intact.
• Base frame is base_link, robot id unitree-go2; the hip-yaw chain is
  calibrated (pitch −0.6 rad / roll −1.5s) so the robot actually settles.
• turn_to_face yaws the body with a bounded applied yaw torque
  (xfrc) plus a differential hip-abduction shuffle, then reports the
  achieved yaw and the remaining error honestly — a partial turn is never
  faked as a complete one.

✅ Success criteria — cleared

• End-to-end paid flow — paid action → Zenoh → skill episode → correlated result on robot/tunnel/result
• Zenoh bridge — subscribes on robot/tunnel/action, republishes with actionId, responds on the documented result topic
• Real action requirement — 8 non-trivial embodied skills, visually demonstrable (screen recording: simulation/docs/go2.gif)
• Payment gate — unpaid ⇒ 402 + PAYMENT-REQUIRED; forged/expired receipts ⇒ 402; replayed idempotencyKey/txHash ⇒ 409; only status: success may settle
• Success & failure semantics — success + 6 deliberately testable failure paths (UNPAID, INVALID_PARAMS, UNKNOWN_SKILL, WRONG_ROBOT, DUPLICATE, tampered paramsHash)
• Reproducible — clean checkout + 5 commands, under 30 minutes, on Windows or ubuntu-latest (CI)
• Safe stop — stop halts motion and returns to the stable home stance
• Code quality — no secrets, configurable topics/robotId/model path, JSON logs, 5 test suites
• Sim-to-sim — same joint configurations recomputed in MuJoCo and PyBullet agree to 0.01 cm across all salient poses (simulation/pybullet/sim2sim_report.json)

🔁 End-to-end flow

paid action (x402) → tunnel → Zenoh robot/tunnel/action
    → robopay_link.py → validate envelope + payment gate
    → joint-space Go2 controller on mujoco_menagerie
    → metrics → result on robot/tunnel/result (correlated by actionId)

🎬 Evidence

• Screen recording (simulated action): simulation/docs/go2.gif (1.6 MB) — the 8 paid skills executed on the official MuJoCo Go2 model
• Terminal logs (paid request → Zenoh → execution → result): simulation/docs/go2-ci-logs.txt
• Per-skill screenshots: simulation/docs/go2-shots/ (wave peak, sit crouch, bow, nod, turn, stance)
• Sim-to-sim report: simulation/pybullet/sim2sim_report.json
• CI: go2-simulation-tests workflow runs the full suite headless on ubuntu-latest (green on the fork)

🚀 Reproduce in < 30 minutes

git clone https://github.com/fabricfoundation/RoboPay.git
cd RoboPay/simulation
./setup.sh                       # pinned official Go2 model assets
cd go2
python3 test_go2_control.py      # every skill's physics actually happen
python3 test_payment_gate.py     # 402/409, no-settle-on-failure
python3 test_result_semantics.py # success/error semantics, replay, tampering
python3 test_link.py             # paid action → Zenoh → episode → result
cd ../pybullet
python3 test_sim2sim_go2.py      # MuJoCo ⇄ PyBullet kinematics agreement

⚠️ Known limitations (honest scope)

• Simulator-only submission — settlement is exercised end to end against the
  same local facilitator the robot link trusts; no on-chain settlement.
• The Go tunnel binary is not run on Windows; the Python payment gate
  reimplements the exact x402 decisions the tunnel's middleware makes before
  actuation, and the wire contract is exercised through peer-mode Zenoh exactly
  as the tunnel would publish it.
• turn_to_face uses an applied yaw torque as a documented stand-in for the
  ground interaction a real Go2 shuffle produces — the only skill (besides
  wave's body-weight compensation) that applies an external force.

---

Built with care by EslaM-X 🇪🇬 — open to review feedback and happy to
iterate on any criterion.

[EslaM-X]

Closing - duplicate targeting wrong base (main). Correct PR is #89 targeting unitree-go2-tier-1 (bounty branch).