Dynamic obstacle avoidance for autonomous vehicles based on improved APF and iteratively optimized MPC

Reliable path planning algorithms and real-time tracking control strategies in dynamic obstacle avoidance scenarios constitute a critical technical challenge in ensuring the safe operation of Autonomous Vehicles (AVs). In response to the limitations of conventional Artificial Potential Field (APF) regarding isotropic potential field assumptions, susceptibility to local minima, and neglect of vehicle performance boundary constraints, this study proposes a planning algorithm driven by an Improved Artificial Potential Field (IAPF). The algorithm achieves optimal path generation consistent with vehicle motion characteristics by establishing a dynamic obstacle avoidance parametric model and integrating road environmental potential field, target position potential field, and vehicle dynamics constraints. Furthermore, to enhance the real-time responsiveness of the controller and target the sensitivity of model-driven control frameworks to the precision of vehicle dynamics characterization, an Iteratively Optimized Model Predictive Control (IOMPC) strategy is proposed. The strategy employs a real-time iterative updating mechanism incorporating nonlinear perturbation terms, which dynamically compensates for model errors and optimizes the control input sequence, thereby enhancing computational efficiency while maintaining predictive control precision. Finally, simulation studies under diverse environmental constraints validate the effectiveness and superiority of the proposed approach, with experimental results demonstrating that the yaw angles of IAPF-generated paths are stabilized within a ±15deg range, while the IOMPC controller achieves 56.4% computational time maximum reduction compared to nonlinear MPC and strictly confines sideslip angles within ±5deg safety thresholds, offering a viable solution for dynamic obstacle avoidance in AVs.