ONNX Export¶
ONNX export is tied to playback in the training scripts. The PPO and HIM-PPO
scripts set EXPORT_POLICY=True when run as scripts, then export during
training.play_only=true playback. APPO, off-policy, and MLX playback paths
also export policy.onnx and verify it with ONNX Runtime in their script code.
Examples¶
uv run eval --algo ppo --task go2_joystick_flat --sim mujoco --load-run -1
uv run eval --algo sac --task g1_walk_flat --sim mujoco --load-run -1
Use the same --algo, --task, and --sim values that produced the
checkpoint. For deployment context, see
ONNX Runtime.