UoE RL Reading Group | 30 March 2023

Speaker: James Ault (Texas A&M University)

Title: MARL for multi-intersection signal control

Abstract: A recent stream of publications proposed to model traffic signal control as a Markov decision process and optimize it with standard or adjusted reinforcement learning (RL) algorithms. While presenting compelling results for optimizing such controllers in stand-alone intersections, limited success was shown for cooperative control over multiple intersections. Still, current control schemes such as green-wave propagation show that coordinating and optimizing signal controllers over adjacent intersections has the potential to greatly reduce congestion and, as a result, emissions and travel times. In this talk we discuss the applicability of state-of-the-art multiagent RL (MARL) algorithms to the multi-intersection signal control domain. Empirical results indicate that such algorithms suffer from limitations in this domain which prevent them from converging to an optimal coordinated policy.

Links:
Cooperative Multi-agent Reinforcement Learning Applied to Multi-intersection Traffic Signal Control: https://people.engr.tamu.edu/guni/pis...

Reinforcement Learning Benchmarks for Traffic Signal Control: https://datasets-benchmarks-proceedin...

Bio: James Ault is a Ph.D. student in the department of Computer Science and Engineering at Texas A&M University advised by Dr. Guni Sharon. He has a broad research interest in reinforcement learning and multi-agent systems, with a particular focus on their applications in traffic control and transportation systems.