End-to-End Reinforcement Learning for Traffic Signal Control: Real-Time Video to Signal Decisions
Keywords: Traffic signal control, Reinforcement learning
TL;DR: MD3DQN is a novel reinforcement learning model for real-time traffic signal control using surveillance video, optimizing phase decisions with an entropy-enhanced deep dueling Q-network for improved traffic flow.
Abstract: Efficient traffic management at urban intersections is vital for reducing congestion
and improving safety. This paper presents MD3DQN, the first End-to-End novel
reinforcement learning model using surveillance video for real-time traffic signal
control. The model features two main components: an image reception module,
capturing traffic data from cameras positioned on signal poles, and a multi-agent
decision module, where each agent manages a traffic phase. These components
are connected via a bridge module for seamless integration.
Our novel Entropy Attention Mechanism enhances the multi-decision turn-based
traffic signal control by leveraging uncertainty and signal phase delays, leading
to more optimized decisions. Results show MD3DQN improved cumulative reward
by an average of 85.2% over Fixed-time 40 and 54.4% over DQN-VTP.
The entropy mechanism contributed to a 41.8% improvement upon ablation study,
demonstrating its impact on faster convergence and better performance.
Primary Area: reinforcement learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 11081
Loading