Timeline: 2023-2026
Keywords: Traffic signal control; Reinforcement learning; Artificial intelligence; Traffic congestion; Greenhouse gas emissions
Team participants: Filipe Rodrigues (PI), Carlos Lima Azevedo (Co-PI), Andrew Nguyen
Lead Organization: DTU Management
Description
Reinforcement learning (RL) constitutes a promising solution for alleviating the problem of traffic congestion and associated greenhouse gas emissions by producing adaptive traffic signal controllers that outperform conventional systems. However the existing solutions are merely reactive. This project aims at developing a new generation of proactive traffic control approaches by integrating into a RL framework a novel fully-Bayesian context-aware traffic prediction model that can forecast the future evolution of traffic and provide uncertainty estimates for its predictions while accounting for contextual information (e.g. about planned events, incidents and weather) traffic network flow theory and the traffic signal control actions. The core idea is that by accounting for the effects of external factors on future traffic conditions the RL agent can learn to be proactive and take preemptive measures to alleviate (or even mitigate) future congestion scenarios and reduce emissions.
Partners
- Technion – Israel Institute of Technology