An Agent of Massage Robot for Task Planning Based on Large Language Models

Traditional robot language systems have limited task planning capabilities. With the rapid development of large language model (LLM) technology, integrating its powerful language analysis capabilities with robots has become a current research hotspot. This paper focuses on the task planning domain of massage robots, and optimizes the prompt strategy using the Chain of Thought (CoT) and ReAct methods. By employing Langchain to combine LLMs with massage control, an intelligent agent for massage robots with automatic task planning and environmental interaction is designed. The optimized prompt strategy is validated on the ALFWorld platform, with a task planning success rate of 72%. In addition, the success rate of using ChatGPT-4 for massage task planning in the real world is observed to be 95%. Finally, a comparison is made between using an external LLM interface and training an LLM locally for task planning. It is found that the method of calling an external interface generally has a higher success rate than the local training method.