Run a Motion Controller in Simulation

Use this guide to watch a trained controller reproduce a motion in MuJoCo. This is the quickest way to answer two basic questions before adding VR or a real robot:

• Does the policy load and keep the G1 stable?
• Does the retargeted motion look like the source motion?

Before You Start

Complete Installation with the base profile and the robots gmr ckpt bvh asset bundle.

1. Run the Sample Motion

python scripts/run/run_sim.py \
    controller.policy_path=ckpt/track_g1.onnx \
    input.bvh_file=data/sample_bvh/aiming1_subject1.bvh \
    playback.keyboard.enabled=true

The sim2sim window is the result that matters: it shows the G1 produced by physics and the policy, not just a kinematic target.

Key	Action
Space or P	Pause or resume
R	Replay from the first frame
Q	Stop

The run is healthy when the robot remains stable and follows the overall timing and pose of the clip. Small tracking error is normal; falling, frozen joints or a clearly wrong facing direction is not.

2. Compare the Three Views

Open all views when you need to find where a bad result starts:

python scripts/run/run_sim.py \
    controller.policy_path=ckpt/track_g1.onnx \
    input.bvh_file=data/sample_bvh/aiming1_subject1.bvh \
    viewers=all

View	What you are looking at
mocap	The human skeleton read from the BVH file
retarget	The kinematic G1 pose produced by GMR
sim2sim	The G1 after policy inference and MuJoCo physics

If mocap is wrong, check the BVH format. If mocap looks right but retarget does not, inspect the retargeting setup. If only sim2sim is wrong, check the policy and observation configuration.

You can also select views explicitly:

# Only the physics result
python scripts/run/run_sim.py \
    controller.policy_path=ckpt/track_g1.onnx \
    input.bvh_file=data/sample_bvh/aiming1_subject1.bvh \
    viewers=sim2sim

# No windows; useful for a server or timing test
python scripts/run/run_sim.py \
    controller.policy_path=ckpt/track_g1.onnx \
    input.bvh_file=data/sample_bvh/aiming1_subject1.bvh \
    viewers=none

Closing every active viewer ends the simulation.

3. Try Your Own BVH

For a LAFAN1-style file:

python scripts/run/run_sim.py \
    controller.policy_path=ckpt/track_g1.onnx \
    input.bvh_file=/path/to/motion.bvh \
    input.bvh_format=lafan1

For an hc_mocap file:

python scripts/run/run_sim.py \
    controller.policy_path=ckpt/track_g1.onnx \
    input.bvh_file=/path/to/motion.bvh \
    input.bvh_format=hc_mocap

Teleopit does not guess an unknown skeleton layout. A file can be valid BVH and still need an adapter before it matches a supported format.

4. Save a Video

Use the renderer when you want repeatable output instead of interactive windows:

MUJOCO_GL=egl python scripts/render/render_sim.py \
    --bvh data/sample_bvh/aiming1_subject1.bvh \
    --policy ckpt/track_g1.onnx

Add --format hc_mocap for that input format. The renderer writes synchronized mocap, retarget and sim2sim videos.

Useful Playback Options

# Hold the final pose instead of exiting
playback.pause_on_end=true

# Stop after 300 simulation steps; 0 means no step limit
num_steps=300

# Keep wall-clock timing even with no viewer
realtime=true

For every available field, see Configuration.