Integrating Reinforcement Learning and Multi-Agent Systems for Autonomous Traffic Management in Smart Cities: A Review

The recent development in information technology including artificial intelligence, and internet of things is affecting every aspect of human existence. This development has given birth to multi-agent deep learning systems. This article takes a broad look at reinforcement learning in multi-agent systems for autonomous traffic management. Reinforcement learning has found application in intelligent traffic management in areas such as traffic flow prediction, traffic management using internet of things and the operation of autonomous vehicles. Spatio-Temporal Graph Convolutional Networks (STGCN) and Spatio-Temporal Graph Neural Network (STGNN) are deep learning models that have researched for traffic prediction, and Vehicles Ad-hoc Networks (VANETs) for traffic management. Also Convolution Neural Network (CNN) has been researched in the operation of autonomous vehicles. However, there are challenges associated with reinforcement learning and multi-agent systems in autonomous traffic management. These include but not limited to quality of data, assumption usually made during simulations which are not valid in real-life situation and lack of definite safety and intelligibility affordances in current algorithm.