A Survey About Learning-Based Variable Speed Limit Control Strategies: RL, DRL and MARL

In the domain of road traffic control, Variable Speed Limit strategies play a crucial role, addressing objectives like mobility enhancement, safety improvement, and environmental considerations. This survey extensively explores the landscape of Learning-based VSL control strategies, investigating the integration of Reinforcement Learning, Deep Reinforcement Learning and Multi-agent Reinforcement Learning techniques. Through a thorough examination of existing literature, we trace the evolution of these strategies, progressing from single-agent VSL approaches to intricate multi-agent VSL strategies. The survey systematically reviews studies that investigate the effectiveness of various Reinforcement Learning algorithms in VSL systems, providing insights into the challenges, advancements, and future directions of both single-agent and multi-agent VSL control strategies. In conclusion, this paper offers a critical review, highlighting key issues identified from the literature review, and suggests directions for future research that should be addressed in the next generation of Learning-based VSL control strategies.