sim2real-gauntlet

Installation
SKILL.md

Sim-to-Real Gauntlet (robosim_gauntlet)

Library and CLI in src/robosim_gauntlet (put src/ on sys.path; cli.ensure_gl_backend() preserves explicit MUJOCO_GL or selects packaged egl/osmesa). Reference consumers: playground_tasks.py and color_blob_ik_controller.py.

Workflow

  1. Task (task.Task protocol): build_device_layer(mode="sim"|"real", extra_devices) returns (numpy DeviceLayer, reset, close) (Impl.CPP in sim; real homes on reset() and parks on close()). Also deployment_keys(), success(poses, poses0), and safety (SafetyLimits: 3D contact wrenches + indirect crush, 35 N grasp cap, bimanual arm_prefixes / gripper_tokens / command_to_joint, and drop_z_tol_m=None / require_release=False / goal_setter for insertions, hinged fixtures, or tool-in-hand tasks).
  2. Controller: either an out-of-process command driving proxy.RemoteDeviceLayer(os.environ["GAUNTLET_URL"]) (prompt in GAUNTLET_PROMPT, printing {"status": "done"|"abort", "fault": ...} last) or an in-process callable fn(dl) -> dict wrapped in wrappers.AllowlistWrapper (include_latency=False default). Both see only deployment_keys() + <cam>_timestamp (upstream timestamps forwarded/delayed; camera reads tracked via last_cam_read_tick).
  3. Preflight: preflight.run(task) audits scene invariants + negative/positive goal controls (safety.goal_setter for articulated fixtures).
  4. Sweep / CLI: episode.run_suite([task], cmd_or_fn, episodes=..., mcap_dir=...) or python -m robosim_gauntlet.cli --task-py=path.py[:fn] --robot=yam --controller="..." --output-json=v1.json --baseline-json=v0.json --mode=sim|real --mcap-log-dir=... (set GAUNTLET_CRN_SECRET for HMAC CRN).
  5. Debug: episode.fork(result, task, tick, cmd_or_fn) replays logged commands to tick deterministically, then hands off; McapWrapper writes .mcap logs via mujoco_playground.real.create_logger identically in sim and real.

Tiers

Installs
15
GitHub Stars
161
First Seen
3 days ago