Large Language model assisted reinforcement learning for robotic scanning measurement path planning of Large-Scale component

With the iterative updates of large-scale component from aerospace industry, the requirements for their manufacturing processes are becoming increasingly demanding. To meet the new technological needs, there is an urgent need to develop measurement approaches featuring higher accuracy, greater reliability, higher automation levels, and improved efficiency. To address this issue, this paper conducts research on robot automated scanning measurement for large-scale component. Firstly, a composite robot scanning measurement system was established and a solution was proposed. In this process, two main research aspects were carried out, namely viewpoint planning and path planning. In viewpoint planning, a random ray interference method is proposed to remove the interference information of the triangular surface model. And Spatial constraint conditions are established to generate and adjust viewpoints with position and attitude information. In path planning, taking the viewpoints as waypoints, a Large Language Model (LLM) assisted reinforcement learning method for path planning is proposed. The specific implementation of how LLM guides reinforcement learning to explore the environment in the path planning task is studied. The results indicate that, under identical conditions, the paths generated by the LLM assisted approach consistently outperform those obtained using reinforcement learning alone and classical TSP solver. Finally, experimental validation is carried out to verify the effectiveness of the proposed approach. The planned path successfully acquires point cloud data that meet the measurement requirements and reduces the measurement time by approximately 22.3% compared with manual planning.