Skip to content

Latest commit

 

History

History
64 lines (49 loc) · 2.84 KB

quadruped_training_isaac_lab.md

File metadata and controls

64 lines (49 loc) · 2.84 KB

Author: Shiyuan Yang
Email: [email protected]

This tutorial is about how to train a quadruped RL policy by Isaac Lab, and how to export .pt file which can be used in Isaac Sim.

Basic Usage

Steps to Use Isaac Lab to train quadruped

  1. Activate the Isaac Lab environment:

    conda activate isaaclab

  2. Navigate to the Isaac Lab directory:

    cd <path_to_isaac_lab>

  3. Run the desired task command:

    ./isaaclab.sh <your task command>

  4. Example command to train a model:

    ./isaaclab.sh -p source/standalone/workflows/rsl_rl/train.py --task Isaac-Velocity-Flat-Unitree-Go2-v0 --num_envs 4096 --headless --video --enable_cameras

  5. After training, run the following command to play the trained model:

    ./isaaclab.sh -p source/standalone/workflows/rsl_rl/play.py --task Isaac-Velocity-Flat-Unitree-Go2-v0 --num_envs 4096

    This will play the trained model, allowing you to see its performance. It will also export the checkpoint file to a model file.

    Note: The checkpoint file does not contain constant.pkl, which is used to temporarily store training status. For more details, see this blog.

  6. After playing and exporting, navigate to:

    /home/ubuntu-user/robot_repo/shiyuan_ws/IsaacLab/logs/rsl_rl/unitree_go2_flat/2024-12-30_13-55-59/exported

    There should be .pt and .onnx files as the final model files.

Issue

When running the following command:

./isaaclab.sh -p source/standalone/workflows/rsl_rl/train.py --task Isaac-Velocity-Flat-Spot-v0 --num_envs 4096 --headless --video --enable_cameras

You might encounter the following error:

omegaconf.errors.UnsupportedValueType: Value 'Tensor' is not a supported primitive type
    full_key: env.scene.robot.actuators.spot_knee.joint_parameter_lookup
    object_type=dict

Re-installing Isaac Lab v1.3.0 (GitHub - isaac-sim/IsaacLab at v1.3.0) instead of the master branch resolved the issue, indicating a potential issue between versions 1.3.0 and the master branch.

References