Generating and Applying Contingency Handling Procedures in Human-Robot Teams in Manufacturing Applications

In manufacturing, minimizing operational delays is crucial for efficiency and resilience. This paper introduces a novel approach to recover from contingencies by using Large Language Models. The core of our approach leverages LLMs to enable quick and autonomous response to unforeseen contingencies. The results demonstrate advancements in:(1) successful recovery from potential assembly operation setbacks, through tailored procedures; (2) the creation of adaptive, reactive strategies to overcome contingency, utilizing generative models; and (3) a significant reduction in human effort and makespan.

IEEE International Conference on Robotics and Automation (ICRA), 2024