SUMO and Reinforcement Learning

Simulation of Urban MObility is an open source, portable, microscopic and continuous multi-modal traffic simulation package designed to handle large networks. (https://sumo.dlr.de/docs/index.html)

This repo contains my main work while developing Single Agent and Multi Agent Reinforcement Learning Traffic Light Controller Agent in SUMO environment.

Further details is as follows:

• Project 1:
  • Implementation of non-RL MaxPressure Agent in SUMO. Work focused on using queue lenght and vehicle waiting time to control a Traffic Light Controller (TLC)
• Project 2:
  • Generating dynamic graph of traffic road netowork in SUMO
• Project 3
  • Implementation of Max Pressure Agent using OOP
• Project 4
  • Implementation of a naive deep Q network in SUMO
• Project 5
  • Implementation of a Deep Q Network in SUMO (ref paper link)
• Project 6
  • Implementation of a Double Deep Q Network in SUMO (ref paper link)

Future Work Direction

• Traffic generation plan
• State space representation
• MARL

Special Thanks

Special Thanks to Dr Phil Tabor for his material on implementing RL research papers.