Toward Cooperative Driving in Mixed Traffic: An Adaptive Potential Game-Based Approach with Field Test Verification

A research team from Tsinghua University has developed a novel Adaptive Potential Game (APG) framework that enables connected autonomous vehicles (CAVs) to cooperate more effectively with human-driven vehicles (HDVs) in mixed traffic environments. The system addresses a critical limitation of existing methods by accounting for individual vehicle needs and heterogeneity through a dual optimization approach that simultaneously considers both individual and system objectives. By establishing a system utility function based on individual utility relationships, the framework allows CAVs to make cooperative decisions that benefit both themselves and the overall traffic flow.

The APG framework introduces two key innovations: Shapley value calculations to quantify each vehicle's marginal contribution to system utility, and dynamic refinement of human driver preference estimation through continuous comparison of observed HDV behavior with estimated actions. This adaptive approach allows the system to learn and respond to human driving patterns in real-time. Ablation studies demonstrated that these adaptive components significantly improve cooperation success rates in mixed traffic scenarios, while comparative experiments showed the APG's advantages in both safety and efficiency metrics over other cooperative driving methods.

Perhaps most importantly, the researchers validated their approach through real-world field tests, moving beyond simulation to demonstrate practical applicability. The framework represents a significant step toward solving the complex challenge of integrating autonomous vehicles into existing traffic systems dominated by human drivers, potentially accelerating the deployment of CAV technology by making it safer and more efficient in real-world conditions.