Expecting the Unexpected: Measure the Uncertainties for Mobile Robot Path Planning in Dynamic Environment

Unexpected obstacles pose significant challenges to mobile robot navigation. In this paper we investigate how, based on the assumption that unexpected obstacles really follow patterns that can be exploited, a mobile robot can learn the locations within an environment that are likely to contain obstacles, and so plan optimal paths by avoiding these locations in subsequent navigation tasks. We propose the DUNC (Dynamically Updating Navigational Confidence) method to do this. We evaluate the performance of the DUNC method by comparing it with existing methods in a large number of randomly generated simulated test environments. Our evaluations show that, by learning the likely locations of unexpected obstacles, the DUNC method can plan more efficient paths than existing approaches to this problem.