This study aims to address decentralized, collaborative decision-making for autonomous vehicles (AVs) in cyber-physical-social systems (CPSS), in which physical states, cyber-decision logic and social behaviors are tightly coupled in the absence of a central controller.
The authors propose a causal-evaluation-based evolutionary game theory (CEGT) framework. The core mechanism is a causal evaluation module that estimates the historical causal influence of each AV on the collective reward and uses these data to adaptively modulate the imitation–mutation rate of the underlying evolutionary game, thereby enabling AVs to learn cooperative behaviors from experience without centralized coordination.
Simulation results indicate that CEGT provides enhanced system-wide resilience within the tested scenarios. Quantitatively, CEGT reduces the average collision count to approximately 0.4 in the complex lane-changing scenario, which is lower than that observed for the Nash (approximately 3.0) and Stackelberg (approximately 0.7) games. In addition, CEGT achieves an approximately threefold higher cumulative system reward and maintains platoon speeds within the 10–15 m/s range even under a strong collision penalty, suggesting a favorable trade-off between safety and efficiency rather than a claim of universal superiority.
Traditional approaches often struggle to provide the resilience required in highly dynamic CPSS environments. This study introduces a novel integration of a causal evaluation module with evolutionary game theory, thereby offering a decentralized framework that explicitly quantifies the influence of historical interactions to support collaborative behavior and improve system-level effectiveness in intelligent transportation.
