Applications of Large Language Models in Mobile Robotics

Abstract

As large language models have advanced technologically, the use of them has increased in multiple fields including robotics. Robots have also come closer to the daily lives of people and in the future the collaboration and assistance from robots is expected to increase. This thesis aims to present ways large language models (LLMs) can be applied in mobile robotics through a literature review of current research. The thesis focuses on three areas of mobile robotics: human-robot interaction (HRI), task planning, and navigation because of the necessity for these areas in future robots working with humans in their daily environments in addition to the utilization of the strengths of LLMs in these areas.

In HRI, LLMs can be applied to generate and understand natural language to improve verbal communication. In addition to verbal communication, nonverbal communication can be improved with the reasoning skills of LLMs. LLMs can additionally be utilized for reasoning about human behavior. The problem solving and reasoning skills of LLMs can be utilized for task planning in generating task plans from natural language instructions. LLMs are additionally used to translate task planning specific programming language to natural language making it easier to understand. In navigation LLMs are applied to enhance navigation capabilities and to improve navigation from natural language instructions. In some applications LLMs achieve results comparable to or better than traditional methods on their own. However, in other applications LLMs could be utilized with traditional methods for the best possible performance.