pengyuan-zhou / Multi-agent-RL-traffic-light-control

Decentralized reinforcement learning for city-scale traffic light control

Decentralized Reinforcement Learning Traffic Light Control

DQN branch contains the code for multi-agent DQN controlled intelligent traffic lights.

Currently DQN branch has been approved by the original Flow community and waiting to be merged.

For detailed changes compared to original Flow code, please refer to the PR

Citing

For more theoretical details such as optima proof, system design, and performance comparison, please refer and cite our paper:

@ARTICLE{9275391,
  author={P. {Zhou} and X. {Chen} and Z. {Liu} and T. {Braud} and P. {Hui} and J. {Kangasharju}},
  journal={IEEE Transactions on Intelligent Transportation Systems}, 
  title={DRLE: Decentralized Reinforcement Learning at the Edge for Traffic Light Control in the IoV}, 
  year={2020},
  doi={10.1109/TITS.2020.3035841}}

Demo:

https://youtu.be/p2sMtN_mW8s

Original instructions, please refer to Flow