tristandijkstra / halfcheetahbioinspired

AE4350 assignment - Half Cheetah with PPO RL

Half Cheetah - RL for legged robots

In this repository, I control the Half Cheetah Reinforcement Learning benchmark with various PPO models.

This project was done as an assignment for the AE4350 Bio-Inspired Learning course at TU Delft.

The final model has a reward score of 3387.

Installing dependencies

The dependencies can be installed using conda /mamba using the following command.

conda env create -f environment.yml

Important files

Models from Scratch

• Basic model: ppo.py
• Improved model: ppoImproved.py

Run files

• Train basic model: run_cheetah.py
• Train improved model: run_cheetah_improved.py
• Train stable baselines3 model: run_cheetah_sb3.py

Final Model

The final Stable Baselines model can be loading using: SB3/final.zip

Other files

Plotting files are prefaced with "plot". Videos are made using record files.

Videos

Doing Nothing

donothing.mp4

Intentional flip and moving on back

flip.mp4

Walking on face

walkonface.mp4

Okay Gait

okay.mp4

Better Gait --- Final model

better.mp4