Offline Motion Tracking

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

Sim2Sim

Start the MuJoCo execution process:

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 \
  --robot g1 \
  --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, then press 9 in the MuJoCo viewer to disable the virtual gantry.

Sim2Real

Replace the MuJoCo execution process with the real bridge:

uv run scripts/real_bridge.py

In a second terminal, run the same tracking policy:

uv run sim2real/rl_policy/tracking.py \
  --robot g1 \
  --policy_config checkpoints/lafan-aa/policy-ec592bb4_lafan_100style_student-5000.yaml

Process roles:

• scripts/real_bridge.py bridges Unitree DDS low_state / low_cmd to the shared ZMQ runtime.
• sim2real/rl_policy/tracking.py stays unchanged between sim2sim and sim2real.

Next Steps

• Pico Teleoperation