ATOM: Leveraging Large Language Models for Adaptive Task Object Motion Strategies in Object Rearrangement for Service Robotics

In the field of service robotics, multi-object re-arrangement is an indispensable skill, extensively employed in various tasks such as desk clearing, shelf organizing, or furniture arrangement. Traditionally, achieving multi-object rearrangement involves complex steps including precise object recognition, spatial planning, and fine-grained motion control. These methods are not only time-consuming but also struggle to adapt to dynamic environments. Recently, Large Language Models (LLMs) have been gaining increasing attention in the field of artificial intelligence, and their integration with robotic technologies has opened new possibilities for multi-object rearrangement. Our proposed approach leverages the advanced features of LLMs to acquire commonsense knowledge about semantically effective object configurations related to multi-object rearrangement. We then employ LLMs for task planning, resorting to traditional methods only in the final stage for actualizing specific arrangement actions. By combining LLMs with traditional techniques, our method significantly simplifies the process of multi-object rearrange-ment tasks for robots. Furthermore, our approach demonstrates adaptability to dynamic environments, thereby expanding the potential applications of service robots in real-world settings.