Game-Theoretic Stochastic Model Predictive Decision-Making Considering Surrounding Vehicle Intentions for Intelligent Vehicles

When a Robotaxi encounters congestion ahead in its lane, it often faces significant delays in completing a lane change due to its limited ability to acquire space for lane-changing from adjacent lane obstacles. To optimize lane-changing opportunities while avoiding collisions, this paper proposes a Game-Theoretic Stochastic Model Predictive Decision-Making (GSMPD) method. This method accounts for the stochastic trajectories of obstacles and their yielding strategies. First, a lateral and longitudinal decision-making model was established, covering planning, control, and dynamics, where lane competition behaviors are considered as integer decision variables. Second, a collision risk assessment method was developed that considers the stochastic future trajectories of obstacles. Third, an obstacle-dominant and host-vehicle-following game optimization problem was established, which considers the obstacle's yielding intention and collision risk; the optimal game decision was obtained by solving the optimization problem using a genetic algorithm. Finally, the effectiveness of this method was validated through Hardware-in-the-Loop (HiL) experiments using a Nuvo-8108GC industrial computer, Simulink, and SCANeR. The results show that the vehicle demonstrates its lane-changing intentions by laterally deviating within the lane and safely completes overtaking maneuvers.