https://github.com/perrin-isir/gym-cassie-run

gym RL environment in which a mujoco simulation of Agility Robotics' Cassie robot is rewarded for walking/running forward as fast as possible
https://github.com/perrin-isir/gym-cassie-run

Last synced: 2 months ago
JSON representation

gym RL environment in which a mujoco simulation of Agility Robotics' Cassie robot is rewarded for walking/running forward as fast as possible

• Host: GitHub
• URL: https://github.com/perrin-isir/gym-cassie-run
• Owner: perrin-isir
• License: other
• Created: 2022-09-29T14:12:28.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2023-11-17T11:44:56.000Z (11 months ago)
• Last Synced: 2024-07-02T11:28:35.384Z (3 months ago)
• Language: Python
• Homepage:
• Size: 33.9 MB
• Stars: 27
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Contributing: CONTRIBUTING.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-legged-locomotion-learning - [**gym-cassie-run** - isir/gym-cassie-run.svg?logo=github) (Code / Quadrupeds)

README

        
# gym-cassie-run

RL environment (with OpenAI Gym interface) in which a mujoco simulation of Agility Robotics' Cassie robot is rewarded for walking/running forward as fast as possible.

The Cassie model is the one available in DeepMind's [mujoco_menagerie](https://github.com/deepmind/mujoco_menagerie), and the gym environment is inspired from 

[HalfCheetah-v4](https://github.com/openai/gym/blob/master/gym/envs/mujoco/half_cheetah_v4.py). In addition to the basic reward for forward motion, hand-designed reward signals 

encourage walking rather than jumping, keeping feet horizontal and not too far apart, and avoiding lateral movements.

The following episode has been obtained after training a [TQC](https://arxiv.org/abs/2005.04269) agent for 17M timesteps, using the [xpag](https://github.com/perrin-isir/xpag) RL platform (at real speed, 100 timesteps = 1 second, cf. top left corner of the gif):

![](episode.gif)

The episode can also be viewed with the [view_episode.ipynb](https://github.com/perrin-isir/gym-cassie-run/blob/main/demo_trajectory/view_episode.ipynb) notebook in the [demo_trajectory/](https://github.com/perrin-isir/gym-cassie-run/blob/main/demo_trajectory/) folder. The trajectory is saved in the [demo_trajectory/episode/](https://github.com/perrin-isir/gym-cassie-run/blob/main/demo_trajectory/episode/) folder.

## Installation

Option 1: pip



    pip install git+https://github.com/perrin-isir/gym-cassie-run



Option 2: conda



    git clone https://github.com/perrin-isir/gym-cassie-run.git

    cd gym-cassie-run

Choose a conda environmnent name, for instance `cassierunenv`.  

The following command creates the `cassierunenv` environment with the requirements listed in [environment.yaml](environment.yaml):

    conda env create --name cassierunenv --file environment.yaml

If you prefer to update an existing environment (`existing_env`):

    conda env update --name existing_env --file environment.yml

To activate the `cassierunenv` environment:

    conda activate cassierunenv

Finally, to install the *gym-cassie-run* library in the activated virtual environment:

    pip install -e .



Once the installation is complete, you can import the environment in python with:  

```import gym_cassie_run```  

This directly registers the environment *CassieRun-v0* in gym.