Appearance-based robotics

A novel appearance-based framework for active robot localization in partially explored environments is introduced. It provides qualitative measurements of the position of the robot, thus monitoring the progress of the overall task. The chromatic and spatial attributes of the color sets extracted from snapshots of the environment are used together with a stochastic evaluator based on partially observable Markov decision process (POMPD). Also, robot localization is performed without using explicit object models. The robustness of the appearance-based framework is demonstrated by a long series of experiments in each of the three environments with different characteristics and with different percentages of knowledge acquired during the visual tour. The proposed metric, based on color sets and weighted walkthrough, easily allows several variations to the similarity of the evaluation.