Information-Theoretic Approach to Efficient Adaptive Path Planning for Mobile Robotic Environmental Sensing

摘要

Recent research in robot exploration and mapping has focused on samplingenvironmental hotspot fields. This exploration task is formalized by Low,Dolan, and Khosla (2008) in a sequential decision-theoretic planning underuncertainty framework called MASP. The time complexity of solving MASPapproximately depends on the map resolution, which limits its use inlarge-scale, high-resolution exploration and mapping. To alleviate thiscomputational difficulty, this paper presents an information-theoretic approachto MASP (iMASP) for efficient adaptive path planning; by reformulating thecost-minimizing iMASP as a reward-maximizing problem, its time complexitybecomes independent of map resolution and is less sensitive to increasing robotteam size as demonstrated both theoretically and empirically. Using thereward-maximizing dual, we derive a novel adaptive variant of maximum entropysampling, thus improving the induced exploration policy performance. It alsoallows us to establish theoretical bounds quantifying the performance advantageof optimal adaptive over non-adaptive policies and the performance quality ofapproximately optimal vs. optimal adaptive policies. We show analytically andempirically the superior performance of iMASP-based policies for sampling thelog-Gaussian process to that of policies for the widely-used Gaussian processin mapping the hotspot field. Lastly, we provide sufficient conditions that,when met, guarantee adaptivity has no benefit under an assumed environmentmodel.