Investigating heterogeneous planning spaces

Abstract

With the continuous improvement of the capabilities of robots and the increasing complexity of the environments they successfully traverse, this paper presents useful concepts and definitions about the heterogeneous nature of planning spaces within the context of motion planning. Our methodology uses the property of visibility, expansiveness and homotopy class to develop algorithms that represent the heterogeneity of the planning space. Our algorithm also include a machine learning technique that identifies sub regions and then intelligently applies necessary existing strategies to create well connected maps in that sub region. We make comparisons with two other machine learning methods in a variety of simulated robot environments ranging from simple homogeneous rooms to complicated maze environments. Our method outperforms the other two methods in terms of time to build a roadmap, the number of nodes needed and the number of connected components generated.