Robustness and Adaptability of Reinforcement Learning-Based Cooperative Autonomous Driving in Mixed-Autonomy Traffic

Building autonomous vehicles (AVs) is a complex problem, but enabling them to operate in the real world where they will be surrounded by human-driven (HVs) extremely challenging. Prior works have shown possibilities of creating inter-agent cooperation between group AVs that follow social utility. Such altruistic can form alliances and affect behavior HVs achieve socially desirable outcomes. We identify two major challenges co-existence HVs. First, preferences individual traits given human driver, e.g., selflessness aggressiveness are unknown an AV, it almost impossible infer real-time during short AV-HV interaction. Second, contrary expected policy, do not necessarily stationary policy therefore hard predict. To alleviate above-mentioned challenges, we formulate mixed-autonomy problem as multi-agent reinforcement learning (MARL) propose decentralized framework reward function for training cooperative AVs. Our approach enables learn decision-making implicitly from experience, optimizes utility while prioritizing safety allowing adaptability; robustifying different behaviors constraining safe action space. Finally, investigate robustness, sensitivity various behavioral present settings which policies adaptable situations.