Graph Attention Neural Network Using Reinforcement Learning for Mixed Traffic Control

We look at the problem of unsignalized mixed traffic control in large urban areas and propose a method for topology independence that should allow for a more generalizable approach to this problem. We will utilize a Graph Attention Network (GAT) for feature extraction and a Dueling Deep Q-Network (DQN) for autonomous vehicle control. to optimize real world traffic flow data from a benchmark dataset. SUMO (Simulation of Urban Mobility) was used to conduct the experiments.