Skip to main content

Offline Motion Tracking

This tutorial uses the root project tracking policy with an offline motion clip.

Sim2Sim​

Start the MuJoCo execution process. It prints the mjviser URL after startup:

uv run sim2real/sim_env/base_sim.py --robot g1

In a second terminal, start the tracking policy:

uv run sim2real/rl_policy/tracking.py \
  --policy-config checkpoints/lafan-aa/policy-ec592bb4_lafan_100style_student-5000.yaml

Process roles:

  • sim2real/sim_env/base_sim.py executes low_cmd in MuJoCo and publishes low_state.
  • sim2real/rl_policy/tracking.py consumes low_state, runs the exported policy, and publishes the next low_cmd.

After both processes are up, press ] in the policy terminal to start. Use the Elastic Band controls in the mjviser UI to disable or tune the virtual gantry.

Integrated Sim2Sim​

Use the integrated runner when the policy and MuJoCo should live in one process. It loads the policy immediately, sets the robot to the first frame of the motion, waits five seconds, tracks until the motion ends, and then holds the last frame. Elastic band is disabled by default for this runner, and the runner prints its mjviser URL after startup.

uv run sim2real/sim_env/integrated_sim2sim.py \
  --robot g1 \
  --policy_config checkpoints/sonic_groot_6k/policy-sonic-groot-006000.yaml \
  --motion_path ../any4hdmi/output/sonic/motions/240529/macarena_001__A545.npz

Add --headless for non-visual runs. The mjviser scene updates once per environment step when a browser client is connected. In mjviser mode, use the Restart motion button after the final-frame hold to reset the robot to the first frame and repeat the wait-track-hold sequence.

For quantitative evaluation, add --trajectory-output <path>.npz and compute motion progress, global root tracking, and local body tracking with the scripts under scripts/tracking_experiment/.

Sim2Real​

For hardware, first choose the deployment path in Robot I/O. For example, the tracking policy still starts with:

uv run sim2real/rl_policy/tracking.py \
  --policy-config checkpoints/lafan-aa/policy-ec592bb4_lafan_100style_student-5000.yaml

Add only the robot I/O flag or bridge process required by the mode you chose.

Next Steps​