Add robot profile: agibot/x2/agibot.x2.mujoco-drake-push.v1

[hossein6191]

AgiBot X2 · Tier 1 — simulator-only. No physical robot.

A payer names where a puck is and where it should end up. Payment is verified
before anything moves, and a failed action never settles.

• registry/vendors/agibot/x2/agibot.x2.mujoco-drake-push.v1/ — profile package
• bridge/agibot/x2/ — Zenoh bridge, x402 gate, IK planner, both engines

Not a replayed animation

Both ends of the motion are parameters of the paid action. Every waypoint is
solved at run time by a constrained IK solver against the puck's measured
pose, and the push direction comes from the live puck-to-goal vector. Stage
transitions fire on sensed conditions; timers exist only as failure guards.

Measured results

Reproducible with python -m sim_bridge.tools.collect_evidence --sim2sim-cases 10 --json.
Targets are sampled from the advertised envelope with a fixed seed, so the
numbers are not hand-picked.

Targets delivered	10 / 10
Sim-to-sim verdicts matching	10 / 10
Worst inter-engine disagreement	51 mm (tolerance 100 mm)
Payment gate cases	13 / 13 as specified
Settlement on failure	never
Tests	48

Method, full numbers and the raw evidence: docs/validation-report.md and
docs/evidence/.

Runs against the tunnel's real wire format

This is worth flagging because it was wrong first. The tunnel in this
repository does not publish a flat action envelope: POST /action sits behind
the x402 middleware and handlers.go publishes

{payload, transaction_details, timestamp}

to robot/tunnel/action — the client's body wrapped verbatim, with the
resolved x402 payload beside it. Two defects followed, and neither could have
been caught by tests that spoke the same invented dialect as the bridge:

1. Only the flat envelope parsed, so every message the real tunnel sends
   would have been refused as malformed — an integration that passes its own
   tests and works with nothing.
2. A transaction hash was required before the robot would act. x402
   verifies, runs the resource handler, then settles — skipping settlement
   when the handler fails. No hash exists when the robot is asked to move.
   Demanding one rejected every real message, and inverted the exact lifecycle
   that gives no-settle-on-failure its meaning.

Both shapes now go through one parser. On arrival the bridge requires a
verified payment plus an authorizationRef — a digest of the x402
authorisation the tunnel verified — rather than a settlement reference that
cannot exist yet.

Checked against the tunnel itself, not just its source.
tunnel/cmd/tunnelprobe drives the real handlers.PostAction through the real
zenoh.Session publisher, with the two context values the x402 gin middleware
sets on a verified payment:

probe	bridge verdict
default (payment verified)	act_probe_307824656000 SUCCESS, settle=true
-unpaid	act_probe_323940412000 PAYMENT_REQUIRED, settle=false

The bytes that crossed the wire are committed as
docs/evidence/tunnel-wire-capture.json. Two things in them are why the fix
was needed: the action fields sit wrapped under payload, and there is no
tx_hash anywhere in the message.

This also closes a hole the wrapper opens. The tunnel forwards the request body
verbatim, so a caller reaching the topic can put
payment: {verified: true, txHash: ...} inside it. Verification is now read
only from transaction_details. Measured: a body claiming a verified payment
of 999999 with a forged hash is refused with PAYMENT_REQUIRED.

Model defects found and fixed

Three things had to be corrected before the two engines could be compared
honestly, each of which otherwise produced something that looked like a physics
result and was not. Chief among them: the tool point was the wrist frame, not
the hand. The hand is a slab ~200 mm deep hanging below that frame, so
planning to it aimed 100 mm under every waypoint — MuJoCo still caught the puck
with the top edge of its larger collision mesh and appeared to work, while
Drake's smaller box passed underneath and never touched it. The sim-to-sim
disagreement was not physics; it was one engine's geometry accidentally
covering a planning bug.

The arm is also not uniform: shoulder and elbow carry 24–36 Nm while the two
wrist trim joints carry 2.2 Nm and stall against any contact, so the planner is
not allowed to spend them.

What is not claimed

Two links in the chain remain unexercised, and neither can be closed from this
repository alone:

• No payment was actually settled. The probe supplies a verified-payment
  context rather than driving the x402 facilitator, because genuine
  verification needs a funded Base Sepolia key — which belongs to the operator,
  not to a repository.
• The tunnel's WebSocket link to the Fabric proxy was not used. The tunnel
  serves its router through internal.NewClient(cfg.ProxyWSURL, ...) rather
  than binding a local port, and the proxy is not part of this repository.

What is established is everything between the tunnel's HTTP handler and the
robot's verdict: the real handler's bytes, the bridge's parse, the execution,
and the settlement decision.

There is no ROS2 here; the profile states ros2: null rather than listing a
topic nothing publishes.