Usage Instructions

March 3, 2025 · View on GitHub

Training && Playing Policy

First, please go to the scripts folder

cd legged_gym/legged_gym/scripts

1. Stage I Discovery Policy Training

1.1 Getting Up Policy

  • Training:
bash run.sh g1waist [your_exp_desc] [device]
# e.g. bash run.sh g1waist stage1_get_up cuda:0
  • Evaluation:
bash eval.sh g1waist [your_exp_desc] [checkpoint]
# e.g. bash eval.sh g1waist stage1_get_up -1

1.2 Rolling Over Policy

  • Training:
bash run.sh g1waistroll [your_exp_desc] [device]
# bash run.sh g1waistroll stage1_roll_over cuda:0
  • Evaluation:
bash eval.sh g1waistroll [your_exp_desc] [checkpoint]
# e.g. bash eval.sh g1waist stage1_roll_over -1

For the main training args:

  • --debug disables wandb and sets the number of environments to 64, which is useful for debugging;
  • --fix_action_std fixes the action std, this is useful for stablizing training;
  • --resume indicates whether to resume from the previous experiment;
  • --resumeid specifies the exptid to resume from (if resume is set true);

For the main evaluation args:

  • --record_video allows you to record video headlessly, this is useful for sever users;
  • --checkpoint [int] specifies the checkpoint to load, this is default set as -1, which is the latest one;
  • --use_jit use jit model to play;
  • --teleop_mode allows the user to control the robot with the keyboard;

2. Stage II Deployable Policy Training

2.1 Log the Stage I policy trajectory

sh log.sh g1waistroll [your_exp_desc] [checkpoint]  # getting up policy
sh log.sh g1waist [your_exp_desc] [checkpoint]  # rolling over policy

Then, please put all trajectories under the simulation/legged_gym/logs/env_logs, the structure looks like:

.
└── env_logs
    ├── getup_traj
    │   ├── dof_pos_all.pkl
    │   └── head_height_all.pkl
    └── rollover_traj
        ├── dof_pos_all.pkl
        ├── head_height_all.pkl
        └── projected_gravity_all.pkl

To help further development over our HumanUP, we provide our discovered trajectories on Google Drive. Feel free to download it to directly train Stage II policy.

2.2 Getting Up Tracking

  • Training:
bash run_track.sh g1waist [your_exp_desc] [device] [traj_name]
# e.g. bash run_track.sh g1waist stage2_get_up cuda:0 getup_traj
  • Evaluation:
bash eval_track.sh g1waist [your_exp_desc] [checkpoint] [traj_name]
# e.g. bash eval_track.sh g1waist stage2_get_up -1 getup_traj

2.3 Rolling Over Tracking

  • Training:
bash run_track.sh g1waistroll [your_exp_desc] [device] [traj_name]
# bash run_track.sh g1waistroll stage2_roll_over cuda:0 rollover_traj
  • Evaluation:
bash eval_track.sh g1waistroll [your_exp_desc] [checkpoint] [traj_name]
# e.g. bash eval_track.sh g1waist stage2_roll_over -1 rollover_traj

3. Save jit model

# bash to_jit.sh g1waist [your_exp_desc] # e.g. bash to_jit.sh g1waist pretrained_exp
python save_jit.py --proj_name g1waist_track --exptid [your_exp_desc] --checkpoint [checkpoint] --robot g1
# e.g. python save_jit.py --proj_name g1waist_track --exptid stage2_get_up --checkpoint -1 --robot g1

You can specify which checkpoint exactly to save by adding --checkpoint [int] to the command, this is default set as -1, which is the latest one.

You can display the jit policy by adding --use_jit in the eval script.

Notes

There are some useful notes:

Simulation Frequency

The simulation frequency has a huge impact on the performance of the policy. Most existing codebases for humanoid robots or quadruped robots use a sim frequency of 200Hz. This is enough for locomotion tasks like walking. For the getting up policy learning, we use a higher frequency of 1k Hz (dt=0.001). Although you can train a reasonable policy in simulation under 200Hz, but it will not work in the real world.

Collision Mesh

For the G1 humanoid robots, we have customized the original G1's collision mesh to simplified and modified collision mesh so that we can accelerate training and improve Sim2Real performance.