Learning task-relevant features from robot data

Abstract
Feature extraction from robot sensor data is a standardway to deal with the high dimensionality andredundancy of such data. An automatic, commonlyused way to learn such features from a set of robotobservations is Principal Component Analysis (PCA).However, as we argued in previous work, PCA canyield features with little discriminatory power betweenrobot positions, leading to suboptimal localizationperformance of the robot. In order to get optimaltask-relevant features, PCA must be...

This publication has 12 references indexed in Scilit: