Multi-agent reinforcement learning for cooperative lane changing of connected and autonomous vehicles in mixed traffic

Autonomous driving has attracted significant research interests in the past two decades as it offers many potential benefits, including releasing drivers from exhausting and mitigating traffic congestion, among others. Despite promising progress, lane-changing remains a great challenge for autonomous vehicles (AV), especially mixed dynamic scenarios. Recently, reinforcement learning (RL), powerful data-driven control method, been widely explored decision makings AVs with encouraging results demonstrated. However, majority of those studies are focused on single-vehicle setting, context multiple coexisting human-driven (HDVs) have received scarce attention. In this paper, we formulate making mixed-traffic highway environment multi-agent (MARL) problem, where each AV makes decisions based motions both neighboring HDVs. Specifically, advantage actor-critic network (MA2C) is developed novel local reward design parameter sharing scheme. particular, multi-objective function proposed to incorporate fuel efficiency, comfort, safety driving. Comprehensive experimental results, conducted under three different densities various levels human driver aggressiveness, show that our MARL framework consistently outperforms several state-of-the-art benchmarks terms comfort.