HimLoco

• forked from: https://github.com/InternRobotics/HIMLoco
• him paper: https://arxiv.org/abs/2404.14405
• hinf paper: https://arxiv.org/abs/2304.08485 (code to be released)
• amp integrated from: https://github.com/Alescontrela/AMP_for_hardware.git
• rewards integrated from:

Installation

1. Create an environment and install PyTorch:

2. Install Isaac Gym:

• Download and install Isaac Gym Preview 4 from https://developer.nvidia.com/isaac-gym
• cd isaacgym/python && pip install -e .

3. Clone this repository.

• cd HIMLoco

4. Install HIMLoco.

• cd rsl_rl && pip install -e .
• cd ../legged_gym && pip install -e .

5. Install LidarSensor

• cd LidarSensor && pip install -e .

Usage

1. Train a policy:

• flat terrain

  • python legged_gym/legged_gym/scripts/train.py --task aliengo --headless
  • python legged_gym/legged_gym/scripts/train.py --task aliengo_recover --headless
  • for lidar:
    • if consider robot sel-occlusion, should combine the robots' meshes first: python legged_gym/resources/robots/aliengo/process_body_mesh.py, then change the consider_self_occlusion=True in env configs (暂时自遮挡后的光线追踪有点问题)
    • python legged_gym/legged_gym/scripts/train.py --task aliengo_lidar --headless

• stairs terrain

  • change the resume flat terrain log path in legged_gym/legged_gym/envs/aliengo/aliengo_stairs_config.py lines 192 load_run = ... and change resume = True

  • python legged_gym/legged_gym/scripts/train.py --task aliengo_stairs --headless

    or

  • python legged_gym/legged_gym/scripts/train --task aliengo_stairs --resume --load_run Jul29_14-35-18_ --headless

• use amp

  • recommand direct 1-stage training (see aliengo_stairs_amp_config.py):
  • python legged_gym/legged_gym/scripts/train.py --task aliengo_stairs_amp --headless

2. Play and export the latest policy:
   • python legged_gym/legged_gym/scripts/play.py --task aliengo --load_run <run_name> --load_cfg
   • python legged_gym/legged_gym/scripts/play.py --task aliengo_stairs --load_run <run_name> --load_cfg
   • train aliengo_stairs_amp and play with random vel_x from -2.0 to 2.0, yaw from -1.0 to 1.0:
   • 
   • some pretrained weights link