This document describes a proposed traffic signal control system using deep reinforcement learning with an emphasis on reinforcing successful experiences. The system aims to reduce traffic congestion in a multi-agent environment without direct communication between traffic signals. It does this by having each signal agent observe changes in congestion within its intersection to predict neighboring signals. Experimental results using a traffic simulator show the proposed system reduces average waiting times at signals by 33% compared to conventional deep reinforcement learning methods. The system aims to apply this research to real-world traffic control environments in the future.

Traffic Signal Control System Using Deep Reinforcement Learning With Emphasis On Reinforcing Successful Experiences PDF

The list of books you might like