Probabilistic Approach for Visual Homing of a Mobile Robot in the Presence of Dynamic Obstacles

An autonomous mobile robot returns back to a reference (home) position using just the visual information in the visual homing task. Challenges in this autonomous task involve efficient information extraction from images and executing motion based on this information. This paper aims to develop visual homing technique by extracting coarse location information with respect to the reference position using a novel bit encoding approach. Coarse information is used in probabilistic framework to build stronger beliefs on the robot's location with successive movement of the robot. The changes in visual information due to the presence of dynamic obstacles are accounted in likelihood information using bit encoding. Novel information encoding and proposed probability distribution develop sensor model. The action model aims for building stronger beliefs with successive movement, when limited number of commands are used. Experiments are performed using Fire Bird XI robot equipped with a multiple camera system for capturing panoramic view at the robot's location. Multiple experiments show that the robot reaches the reference position without the knowledge of initial location and changing environment conditions. This paper also reports experimental results in the presence of
dynamic obstacles.