Importance Sampling based on Adaptive Principal Component Analysis

Abstract

Sampling-based approaches are currently the most efficient ones to solve path planning problems, being their performance dependant on the ability to generate samples in those areas of the configuration space relevant to the problem. This paper introduces a novel importance sampling method that uses Principal Component Analysis to focalize the region where to sample in order to increase the probability of finding collision-free configurations. The proposal is illustrated with a 2D configuration space with a narrow passage and compared to the uniform random sampling method.