ClimbingMachine/deep-reinforcement-learning-pedestrian-signal-design

This study is to investigate the optimal control strategies at crosswalks using traffic signal controllers. A multi-agent reinforcement learning framework will be proposed as the “smart” control strategy, and several experiments will be conducted using microsimulation. The proposed multi-agent reinforcement learning framework is aimed to (1) find the optimal control policy that minimizes the number of conflicts (safety) while reducing traffic delay (efficiency), (2) account for different scheduling scenarios with various combinations of pedestrian flow rates and vehicle flow rates, and (3) make comparisons with baseline traditional traffic signal controllers and semi-controlled strategy.

Jupyter Notebook