Autonomous vehicles (AVs) stand as the forefront of future societies, yet their coexistence with human-driven vehicles (HVs) in the near-to-mid-term necessitates a comprehensive understanding of the mixed autonomy traffic. Evidence has shown that 1) the real-world interactions between AVs and HVs are not well understood, 2) undesirable AV behaviors can have adverse effects on other road users, and 3) the benefits of reducing greenhouse gas (GHG) emissions are not significant, particularly at low AV penetration rates. This research includes a game theory-guided, AI-driven framework to address these issues. The goal is to understand vehicle interactions in mixed traffic and leverage the understanding to shape AV behaviors, simultaneously benefiting all road users (i.e., ensuring equity) and achieving emission reduction, even with a limited proportion of AVs. Centered on the concept of collective cooperation proposed in the researchers’ previous work (Li et al., 2022), they empirically verify its existence in real-world mixed traffic. Leveraging this property, the researchers will design AV behaviors using deep reinforcement learning (DRL), steering the system towards Pareto-efficiency. The encouragement of spatial separation in collective cooperation also suggests the potential for emission reduction. Preliminary results show higher speeds for both AVs and HVs, along with spatial separations and platooning behaviors. These findings suggest the promise of the researchers’ proposed framework in fulfilling the equity and emission objectives, which is worth further investigation.
