final-project-done_finally-1 created by GitHub Classroom Yi Chang 260619034 Wenting Wang 260367035 Yun Chen 260772822

This project explores applying Reinfrocement Learning approaches to traffic signal control

State, action definitions are implemented based on the understanding of the paper 'Adaptive Traffic Signal Control : Exploring Reward Definition For Adaptive Traffic Signal Control : Exploring Reward Definition For Reinforcement Learning' written by Saad Touhbia, Mohamed Ait Babrama, Tri Nguyen-Huub, Nicolas Marilleaub, Saad Touhbi , Mohamed Ait Babram , Tri Nguyen-Huu , Nicolas Marilleau, Moulay L. Hbida, Christophe Cambierb, Serge Stinckwichb (https://ac.els-cdn.com/S1877050917309912/1-s2.0-S1877050917309912-main.pdf?_tid=dd25f67a-c877-4538-a81e-8f044412a2da&acdnat=1524859146_4fbe7f74db4fd49aa0e9d9d7b4b23dab)

A OpenAI Gym (https://gym.openai.com/envs/#classic_control) environment is defined to simulate the environment using the following packages

• Simulation of Urban Mobility, SUMO (http://sumo.dlr.de/index.html) for microscopic traffic simulation
• TraCi (http://sumo.dlr.de/wiki/TraCI) for communication between agents and SUMO in Python

Follow instruction to install SUMO and add environment variable SUMO_HOME (http://sumo.dlr.de/wiki/Installing)

four_intersects.py
definition of RL agents, control of traffic lights using fixed control Q learning and dyna_Q

graph_results.py
graph experiment result based on the output of four_intersects.py

gym_env

• four_intersects_env.py
  definition of Gym environment, different reward defintion is implemented here (please fix directory as you need)

• register.py
  register the new Gym environment. 10 version of the environment is registered for 10 random sets of traffic scenario (please fix directory as you need)

• import.py
  import the new Gym environment

sumo_env

• includes all files for traffic simulation