Reasoning with robot execution failures in noisy environments

Robust execution of robotic tasks is a difficult problem. Improper execution could lead to very costly and difficult diagnosis of robot failures. In this paper a solution for the problem of dealing with noisy in robotics execution failures from a stochastic sampling (STOCHS) or probabilistic point of view is derived. Our experimental results show the STOCHS algorithm as more robust to noise compared to machine learning techniques (or classifiers) such as decision trees and naïve Bayes classifier with the performance of RIPPER as substantially inferior.