Key feature-based approach for efficient exploration of structured environments

Abstract

This paper presents an exploration approach for robots to determine sensing actions that facilitate the building of surface maps of structured partially-known environments. This approach uses prior knowledge about key environmental features to rapidly generate an estimate of the rest of the environment. Specifically, in order to quickly detect key features, partial surface patches are used in combination with pose optimisation to select a pose from a set of nearest neighbourhood candidates, from which to make an observation of the surroundings. This paper enables the robot to greedily search through a sequence of nearest neighbour poses in configuration space, then converge upon poses from which key features can best be observed. The approach is experimentally evaluated and found to result in significantly fewer exploration steps compared to alternative approaches.