Evolutionary Game Theory-Driven Multi-Mode Control Coordination for Autonomous Vehicle Trajectory Tracking

Abstract

Autonomous vehicles operating under highly dynamic conditions, such as high-speed cornering and varying road surfaces, face significant challenges in trajectory tracking owing to the highly nonlinear and time-varying nature of vehicle dynamics. This paper proposes an adaptive time-varying weighted cooperative control method based on evolutionary game theory for trajectory tracking of autonomous vehicles. The method dynamically adjusts weights between Pure Pursuit (PP) and an evolutionary game-based model predictive controller (EG-MPC) via replicator dynamics, enabling the coordinated optimization of tracking accuracy and control stability. A normalized speed-zone definition is introduced to reduce dependence on fixed absolute speed thresholds. This normalization transfers only the zone-definition rule, while vehicle-specific model parameters, actuator constraints, tire characteristics, and raw-cost payoff maps must still be regenerated before deployment on different platforms. CarSim simulations under three road adhesion conditions ( $\mu =0.85$ , 0.60 and 0.30) demonstrate that the proposed method reduces the maximum lateral displacement error by up to 73.2% under high-adhesion conditions and the maximum yaw angle error by up to 63.0% compared to standalone controllers, with negligible additional runtime relative to the PP baseline ( $\le 0.32$ %). Real-vehicle experiments on a 107 m track at speeds up to 60 km/h further validate the approach, achieving reductions of 41.7% in maximum lateral error and 44.1% in yaw angle error relative to the standalone PP controller, while maintaining smooth mode transitions. These results suggest that the proposed framework can improve transient peak-error suppression and smooth mode transition under the tested racing conditions, while further validation on broader vehicle platforms remains necessary.