Expectancy-based robot navigation through context evaluation

Abstract

Agents that operate in a real-world environmenthave to process an abundance of information, which maybe ambiguous or noisy. We present a method inspired bycognitive research that keeps track of sensory information,and interprets it with knowledge of the context. We test thismodel on visual information from the real-world environmentof a mobile robot in order to improve its self-localization.We use a topological map to represent the environment,which is an abstract representation of distinct places andthe connections between them. Expectancies of the placeof the robot on the map are combined with evidence fromobservations to reach the best prediction of the next place ofthe robot. These expectancies make a place prediction morerobust to ambiguous and noisy observations. Results of themodel operating on data gathered by a mobile robot confirmthat context evaluation improves localization compared to adata-driven model.