Cooperative Autonomous Driving Control among Vehicles of Different Sizes Using Deep Reinforcement Learning

This study explores acquiring cooperative control in an environment where only multiple autonomous vehicles with different sizes navigate using deep reinforcement learning (DRL). Previous works proposed a cooperative learning approach utilizing information obtained through vehicle-to-vehicle communication. However, the previous research assumed all vehicles to be the same size. Indeed, in the real environment, vehicles of various sizes, such as compact cars and large trucks, coexist on the roads and must cooperatively drive. This study aims to achieve autonomous driving control among vehicles of different sizes using DRL and analyze the impact of vehicle size on cooperative behavior. We conducted simulation experiments using vehicles of sizes corresponding to regular cars and medium-sized trucks. Experimental results revealed that cooperative autonomous driving control was successfully acquired even in environments where vehicles of different sizes coexisted. Furthermore, the model trained exclusively with medium-sized trucks demonstrated the ability to mitigate collisions when operating even in environments with a mixture of vehicle sizes. In contrast, the model trained with a mixture of vehicle sizes demonstrated more collisions in the medium truck-only environment. This result indicates that a model trained in a challenging environment (where only large-sized vehicles exist) performs better in simpler environments (environments with only small-sized vehicles or a mix of small and large-sized vehicles) in terms of collision avoidance.