MULTI-AGENT REINFORCEMENT LEARNING FOR INTEGRATED AND NETWORKED ADAPTIVE TRAFFIC SIGNAL CONTROL

摘要

A system and method of multi-agent reinforcement learning for integrated andnetworkedadaptive traffic controllers (MARLIN-ATC). Agents linked to traffic signalsgenerate controlactions for an optimal control policy based on traffic conditions at theintersection and one ormore other intersections. The agent provides a control action considering thecontrol policy forthe intersection and one or more neighbouring intersections. Due to thecascading effect of thesystem, each agent implicitly considers the whole traffic environment, whichresults in an overalloptimized control policy.