⚡ Training & Playing

A quick guide to training and evaluating universal motion trackers for humanoid robots using InternHumanoid.

For command-conditioned humanoid control and upper-body manipulation, see the reference application Flomo (Foundation Locomotion).

📋 Table of Contents

Train the universal motion tracker for Unitree G1-29 DoF with the following command:

python legged_gym/scripts/train.py +algo=ppo +robot=g1/g1_29dof +task=imitation/g1_29dof

To run on CPU, add: +sim_device=cpu +rl_device=cpu (sim on CPU and RL on GPU is possible).
To run headless (no rendering), add: +headless.
Tip: To improve performance, once training starts, press v to stop rendering. You can enable it later to check progress.
The trained policy is saved in issacgym_anymal/logs/<experiment_name>/<date_time>_<run_name>/model_<iteration>.pt. <experiment_name> and <run_name> are defined in the train config.
Command line arguments override config file values:
- +resume: Resume training from a checkpoint
- +experiment_name EXPERIMENT_NAME: Name of the experiment to run or load
- +run_name RUN_NAME: Name of the run
- +load_run LOAD_RUN: Name of the run to load when resume=True. If -1, loads the last run
- +checkpoint CHECKPOINT: Saved model checkpoint number. If -1, loads the last checkpoint
- +num_envs NUM_ENVS: Number of environments to create
- +seed SEED: Random seed
- +max_iterations MAX_ITERATIONS: Maximum number of training iterations

After training, play the saved checkpoint with:

python legged_gym/scripts/motion_imitation/play.py +algo=ppo +robot=g1/g1_29dof +task=imitation/g1_29dof

By default, the last model of the last run in the experiment folder is loaded.
Other runs/model iterations can be selected by setting load_run and checkpoint in the train config.