A Reasoning Method for Micro Assembly Sequence Based on Knowledge Graph and Large Language Model

For the assembly of multi-species laser inertial confinement fusion microtargets, micro-assembly robots cannot adaptively respond to changes in the structure of the microtargets and generate assembly sequences autonomously, and manual intervention is needed to reset the assembly process, so there are the problems of autonomous configuration capability and low rapid response capability of the robot system for the assembly of new microtargets and the problem of low response capability, which can not satisfy the needs of small batch and multi-species assembly and rapid response. For this reason, this paper introduces Large-scale Language Model (LLM) and Knowledge Graph (KG) in the assembly sequence generation of micro-assembly robots, and puts forward the method of "Thinking with Knowledge Graph" (TwKG). By integrating external knowledge such as manual assembly experience, a priori information of parts, and robot information, the method enables the micro-assembly robot to generate assembly sequences autonomously, optimize the sequences based on assembly experience, and update the sequences autonomously in case of environmental changes. Since the sequences directly generated by this method cannot be directly used for device control, a Reactive Behavioral Trees method (RBT) based on dynamic behavioral trees is designed based on the analysis of rationality relationships among assembly sequences for automatically creating and updating assembly sequences to cope with abnormal situations. Finally, the effectiveness of the proposed method in improving assembly efficiency and sequence rationality is verified by comprehensive tests in the equipment platform.