RT info:eu-repo/semantics/conferenceObject
T1 Reinforcement learning for Traffic Signal Control: Comparison with commercial systems
A1 Cabrejas Egea, Álvaro
A1 Zhang, Raymond
A1 Walton, Neil
K1 Tráfico
K1 Traffic
K1 Infraestructuras
K1 Infrastructures
K1 Ingeniería civil
K1 Civil engineering
K1 Transporte
K1 Transportation
K1 Tecnología
K1 Technology
AB In recent years, Intelligent Transportation Systems are leveraging the power of increasedsensory coverage and available computing power to deliver data-intensive solutionsachieving higher levels of performance than traditional systems. Within Traffic SignalControl (TSC), this has allowed the emergence of Machine Learning (ML) based systems.Among this group, Reinforcement Learning (RL) approaches have performed particularlywell. Given the lack of industry standards in ML for TSC, literature exploring RL often lackscomparison against commercially available systems and straightforward formulations ofhow the agents operate. Here we attempt to bridge that gap. We propose three differentarchitectures for RL based agents and compare them against currently used commercialsystems MOVA, SurTrac and Cyclic controllers and provide pseudo-code for them. Theagents use variations of Deep Q-Learning (Double Q Learning, Duelling Architectures andPrioritised Experience Replay) and Actor Critic agents, using states and rewards based onqueue length measurements. Their performance is compared in across different mapscenarios with variable demand, assessing them in terms of the global delay generated by allvehicles. We find that the RL-based systems can significantly and consistently achieve lowerdelays when compared with traditional and existing commercial systems.
PB Universidad de Burgos. Servicio de Publicaciones e Imagen Institucional
SN 978-84-18465-12-3
YR 2021
FD 2021-07
LK http://hdl.handle.net/10259/7002
UL http://hdl.handle.net/10259/7002
LA eng
NO Trabajo presentado en: R-Evolucionando el transporte, XIV Congreso de Ingeniería del Transporte (CIT 2021), realizado en modalidad online los días 6, 7 y 8 de julio de 2021, organizado por la Universidad de Burgos
NO This work was part funded by EPSRC Grant EP/L015374 and part funded by The Alan Turing Institute and the Toyota Mobility Foundation. The authors thank Dr. W. Chernicoff for the initial discussions and drive that made this project possible.
DS Repositorio Institucional de la Universidad de Burgos
RD 04-may-2024