A Study on Grounding of Teaching Information for Collision Avoidance by Autonomous Robot

Abstract

This paper studies grounding of humans' teaching information by a mobile robot. Human and robot have different sensors and actuators. It is difficult for the robot to understand humans' words that are indirect sensory information for the robot. This paper proposes an autonomous robot system that can ground the humans' teaching information. This system acts in an environment using the action rules given by a human, and collects data from its own sensors and actuators. Learning with the collected data using Genetic Algorithm (GA) and a Fuzzy-Neural Network (FNN) is done. The robot acquires new action rules based on its own sensors. This acquisition is called grounding of teaching information in this paper. The effects of the proposed system are shown with simulation results.