Lane-Changing Strategy for Autonomous Vehicle with Adaptive Adjustment of Decision-Making Preference based on Game Theory

Lane-changing is a critical task in autonomous driving. Rational lane-changing behavior can enhance both travel efficiency, driving safety. However, existing studies suffer from issues such as ambiguous causal relationships in decision-making logic and insufficient generalization capabilities in complex scenarios. To address these challenges, this paper proposes an interactive modeling approach that combines game theory with deep reinforcement learning (DRL). This method integrates the scene generalization ability of DRL and the clear decision-making causal relationship of game theory into a unified model. Furthermore, to tackle the challenge of determining optimal lane-changing timing, this study employs a long short-term memory (LSTM) neural network to enable game theory to incorporate temporal considerations. In this model, the agent intervenes in the calculation of the game theory-based payoff function and adjusts decision-making preferences based on indicators such as efficiency, safety, and comfort to align with environmental requirements, thereby adapting to diverse environments. Tests demonstrate that the model proposed in this paper is capable of achieving safe and efficient decision-making by adaptively adjusting decision-making preferences in complex and extreme scenarios. Specifically, it achieves a success rate of 99.67% in an environment with surrounding vehicles exhibiting diverse driving styles. Overall, the model presented in this paper not only ensures efficient and safe decision-making through preference adjustments but also preserves the clear causal relationships inherent in game-theoretic decision-making.