The complex surface topography and uneven fluorescence signal distribution of stone carvings render traditional fixed-path scanning inefficient and prone to damage. This paper proposes an adaptive path planning method integrating multi-source information from fluorescence, geometry, and mechanics. First, a multi-modal environmental model is constructed by combining 3D point clouds with fluorescence intensity maps to establish a cost map coupling curvature, fluorescence intensity, and safety constraints. Subsequently, an enhanced RRT* algorithm is designed by incorporating fluorescence-weighted biased sampling and probe orientation constraints, enabling global path planning that simultaneously covers critical low-fluorescence areas while maintaining probe verticality. Finally, an online fine-tuning strategy based on reinforcement learning (RL) optimizes contact force and pose in real time, ensuring artifact safety and signal quality. Simulations and physical experiments demonstrate that this method achieves 96.7% coverage of critical regions on Buddhist statue and stele models, reduces force-limit duration to 1.2%, and improves signal-to-noise ratio by 58%. It significantly outperforms grid-based methods and standard RRT*, providing an efficient and safe robotic solution for the non-destructive digital preservation of stone carvings.
Kai Chen (Sun,) studied this question.