- requirements.txt includes all the dependecies required to run this project.
- You can watch a few of our agents in this video
-
Clone the repo and cd into it:
git clone https://github.com/aagrawal20/DeepParkour.git cd DeepParkour -
If have access to a CUDA-compatible gpu then install tensorflow gpu.
pip install tensorflow-gpu
Refer to TensorFlow installation guide for more details.
-
Install DeepParkour package.
pip install -e .
- You can train an agent using the train_agent.py file.
- You can add specific flags to the argument parser.
python src/main/train_agent.py
- You can render an agent using the render_agent.py file.
- You can add specific flags to the argument parser.
Note: PyBullet only supports CPU rendering. Turn off render flag or manually turn off gpu.
python src/main/render_agent.py
- You can visualize different stastics for eg: loss vs timesteps or reward vs timesteps.
jupyter notebook
- After loading the local host navigate to the visualization.ipynb in src/util.