We use RL to train ANYmal to achieve two tasks, "walk stably" and "navigation" in Raisim.
This is the environment for the task "walk stably", we simply want the robot to walk stably with minimum torque.
This is the environment for the task "navigation", we want our robot to navigate to a certain position. The reward is based on distance error between current position and the target position.
Different from navigation_ori_anymal
only in reward, which is based on distance error between current position and the target position.
Please first check out the installation guide on RaiSim website to install Raisim.
Clone this repo and place these three folders in PATH_TO_raisimLib/raisimGymTorch/raisimGymTorch/env/envs
.
cd PATH_TO_raisimLib/raisimGymTorch
mkdir data
cd data
The trained model can be found in this link, please download the folder and place it in the directory data
.
In the following please run commands in the directory PATH_TO_raisimLib/raisimGymTorch
.
- Compile raisimgym:
python setup.py develop
- Run runner.py of the task (for rsg_anymal example):
python raisimGymTorch/env/envs/rsg_anymal/runner.py
- Compile raisimgym:
python setup.py develop
- Run tester.py of the task with policy (for rsg_anymal example):
python raisimGymTorch/env/envs/rsg_anymal/tester.py --weight data/rsg_anymal/FOLDER_NAME/full_XXX.pt
- Compile raisimgym:
python setup.py develop
- Run runner.py of the task with policy (for rsg_anymal example):
python raisimGymTorch/env/envs/rsg_anymal/runner.py --mode retrain --weight data/rsg_anymal/FOLDER_NAME/full_XXX.pt
Please check out this link for more details about RaisimGymTorch.