Core Concepts
MARVEL proposes a novel multi-agent reinforcement learning framework for large-scale variable speed limit control, improving traffic safety and mobility.
Abstract
MARVEL introduces a novel framework for large-scale Variable Speed Limit (VSL) control using Multi-Agent Reinforcement Learning. The framework focuses on adaptability to traffic conditions, safety, and mobility by training policies in a microscopic traffic simulation environment. It scales to cover corridors with many agents and improves traffic safety by 63.4% compared to no control scenario. MARVEL enhances traffic mobility by 58.6% compared to the state-of-the-practice algorithm deployed on I-24. The proposed method is tested on a network with 34 VSL agents spanning 17 miles near Nashville, TN, USA. An explainability analysis is conducted to examine the decision-making process of the agents under different traffic conditions.
Stats
MARVEL-based method improves traffic safety by 63.4%
MARVEL-based method enhances traffic mobility by 58.6%