Traffic light control using deep policy‐gradient and value‐function‐based reinforcement learning. Issue 7 (20th September 2017)

• Record Type: Journal Article
• Authors: Mousavi, Seyed Sajad; Schukat, Michael; Howley, Enda
• Is Part Of: IET intelligent transport systems. Volume 11:Issue 7(2017)
• Page Start: 417
• Page End: 423
• Publication Date: 2017-09-20
• Subjects: gradient methods -- learning (artificial intelligence) -- adaptive control -- road traffic control -- traffic engineering computing -- control engineering computing -- digital simulation; traffic light control -- value‐function‐based reinforcement learning -- deep neural network architectures -- complex control problems -- high‐dimensional state space -- action spaces -- deep policy‐gradient RL algorithm -- value‐function‐based agent RL algorithms -- traffic signal -- traffic intersection -- adaptive traffic light control agents -- graphical traffic simulator -- control signals -- PG‐based agent maps -- optimal control -- urban mobility traffic simulator -- training process; Intelligent transportation systems -- Periodicals; Electronics in transportation -- Periodicals; 388.31205
• DOI: 10.1049/iet-its.2017.0153
• Languages: English
• ISSNs: 1751-956X

Abstract

Abstract : Recent advances in combining deep neural network architectures with reinforcement learning (RL) techniques have shown promising potential results in solving complex control problems with high‐dimensional state and action spaces. Inspired by these successes, in this study, the authors built two kinds of RL algorithms: deep policy‐gradient (PG) and value‐function‐based agents which can predict the best possible traffic signal for a traffic intersection. At each time step, these adaptive traffic light control agents receive a snapshot of the current state of a graphical traffic simulator and produce control signals. The PG‐based agent maps its observation directly to the control signal; however, the value‐function‐based agent first estimates values for all legal control signals. The agent then selects the optimal control action with the highest value. Their methods show promising results in a traffic network simulated in the simulation of urban mobility traffic simulator, without suffering from instability issues during the training process.