SUMO Based Hardware/Software Co-simulation for Two-Intersection Adaptive and Collaborative Traffic Signal Controller

Due to safety reasons, the development of Reinforcement Learning (RL) based Traffic Signal Controller (TSC) can not be perform in real-world environment. A training and hardware/software co-simulation platform is needed to verify that the RL-based TSC works well before being deployed in real-world environment. In this paper we discuss the design and development of such platform using SUMO and python for 2-agent Q-Learning Algorithm. The platform receives Q-Matrix from an external device and then process the Q-Matrix to control the environment. Overall, the design works well. The platform could train the RL Agent and could affect the environment. Improvement could be made in minimizing the execution time of the simulation.