A codebase for running the MPPI algorithm on CarRacing and OpenAI gym style environments

This repo requires Python3 (3.7.7 for best compatibility) If you have a NVIDIA GPU, install CUDA 10.2 to allow the repo to access it.

• Python Install Python>=3.6 from https://www.python.org/downloads/
• CUDA (Optional) Install CUDA 10.2 from https://developer.nvidia.com/cuda-10.2-download-archive

To install the python dependencies after cloning the repo, run

pip3 install -r requirements.txt

Watch an MPPI agent control the CartPole-v0 or Pendulum-v0 environment in in Python with:

python3 train_agent.py [CartPole-v0|Pendulum-v0] mppi --trial --render

To see a MPPI agent control a car along various race tracks, run:

python3 train_agent.py CarRacing-[curve|curve1|curve2|curve3|curve4|curve5]-v1 mppi --trial --render

To see the visualization of the samples trajectories evaluated for the curve car racing track, run:

python3 test.py

The MPPI parameters for the control optimization are in src/agents/init.py where the agent_configs parameters are for the OpenAI gym style environments and the env_agent_configs are for the CarRacing environments.