Multi-Agent Reinforcement Learning With Cross-Layered Adaptive Wireless Video Streaming for Road Traffic Monitoring

This paper presents the design and implementation of a multi-agent reinforcement learning framework for adaptive wireless image sequence streaming in road traffic monitoring systems. This work extends previous research that utilizes Apache Kafka for real-time wireless image transmission. To promote cooperation and fairness among agents, a multi-agent architecture with independent learners employing a social welfare function as a joint reward is implemented. The learning agents are trained and evaluated under various scenarios, and their performance is compared to a baseline without learning agents. Experimental results show that, after sufficient training, the proposed approach outperforms the baseline by 3.98% to 31.55% in joint reward. An emulated software-defined wireless mesh network is built with Mininet-WiFi to test the scalability and convergence time. This study highlights the potential of multi-agent reinforcement learning for improving adaptive wireless image streaming in road traffic monitoring, with significant implications for future research and real-world applications.