Awesome
MACAD-Agents
Multi-Agent algorithms for Multi-Agent Connected Autonomous Driving using MACAD-Gym
How to train/test MACAD-Agents?
git clone https://github.com/praveen-palanisamy/macad-agents
If you want to avoid building and running the Docker container, you can follow the instructions in theRunning MACAD-Agents witout Docker section instead and skip the next 2 steps.
-
Build the MACAD-Agents Docker container:
docker build --rm -f macad-agents/Dockerfile -t macad-agents:latest . -
Run the MACAD-Agents training container:
bash run.shYou can pick from one of the available multi-agent training options:
-
To train multiple agents using PPO where the agents communicate/share learned weights, modify the last line in
run.shto look like this:macad-agents:latest python -m macad_agents.rllib.ppo_multiagent_shared_weights.py -
To train multiple agents using IMPALA where the agents communicate/share learned weights, modify the last line in
run.shto look like this:macad-agents:latest python -m macad_agents.rllib.impala_multiagent_shared_weights.py
-
Running MACAD-Agents without Docker
If you have all the necessary dependencies installed an configured on your host machine, you can run the agent script like shown below:
cd macad-agents/src && python -m macad_agents.rllib.ppo_multiagent_shared_weights
A brief gist of what you need to setup on your host machine is listed below:
Where -e is equivalent to export using the bash terminal.
Citing
If you find this work or MACAD-Gym useful in your research, please cite:
@misc{palanisamy2019multiagent,
title={Multi-Agent Connected Autonomous Driving using Deep Reinforcement Learning},
author={Praveen Palanisamy},
year={2019},
eprint={1911.04175},
archivePrefix={arXiv},
primaryClass={cs.LG}
}