Cooperative Evaluation of Actions in Swarm Robots Using Large Language Models

Abstract

Evaluation of the success or failure of actions is essential for robot systems. Evaluations should be autonomous and challenges remain like how to tackle sensor troubles. In such situations, cooperative evaluation of actions will be effective via autonomous information sharing among robots. We design an autonomous swarm robotic system that allows for information sharing between robots using natural language, based on foundation models. Then, we verify that the robots autonomously evaluate the action that a single robot cannot judge by the individual through information sharing.