Dynamic Parameter Optimization Framework for Frenet Path Planning: From Static to Adaptive Algorithms

The Frenet coordinate system is widely adopted for autonomous driving path planning due to its computational efficiency. However, its reliance on static parameters becomes a critical bottleneck in dynamic and complex scenarios such as campus logistics and disaster rescue, often resulting in planning failures near goals and sensitivity to uneven obstacle distributions. To overcome these limitations, this paper proposes a dynamic parameter optimization framework that enhances the adaptability of Frenet-based planning. Our approach introduces a state-aware mechanism, which utilizes real-time inputs including obstacle density, minimum vehicle-obstacle distance and goal distance to dynamically adjust lateral sampling intervals, time spans and constraint thresholds. The framework integrates three core innovations: a hierarchical sampling strategy that increases resolution in dense environments, a goal-sensitive retry mechanism that relaxes parameters when approaching targets, and a demand-driven obstacle generation model that ensures balanced scenario coverage for robust evaluation. Experimental results demonstrate that the proposed framework significantly improves performance and efficiency. It achieves a 67.98% reduction in computation time while maintaining an overall success rate of 82.0%. Notably, success rates on challenging W-shaped paths are increased by 16.0%, and endpoint planning failures are completely eliminated. This work provides a practical and efficient solution for reliable path planning in dynamic autonomous driving applications.