DLASIUT – Development of Learning Agent-based Systems for Improved Urban Traffic Control

DLASIUT addresses the development of traffic control systems based on machine learning for improved urban traffic management. Because large numbers of observed traffic parameters create an exponentially growing state and action space, the project proposes a framework for a learning-agent-based traffic control system capable of deriving optimal control policies from microscopic simulations of realistic urban road networks, including support for connected and autonomous vehicles. A structured simulation environment is used for systematic testing, learning, and identification of undesirable behaviours, with the aim of supporting real-world implementation by traffic management centres and system developers.