An universal GMMs-based sample strategy and collision checker for robot motion planning

Abstract

The sampling-based motion planning methods acquire obstacle information by performing collision detection for a large number of samples, which can effectively solve autonomous obstacle avoidance problem in the high-dimensional configuration space. In order to improve the efficiency of planning, this paper proposes a new adaptive sampling strategy and collision checker based on Gaussian Mixture Models(GMMs): For the "narrow corridor" problem, the sample set obtained by Gaussian sampling strategy of adjustable standard deviation is used to train the GMMs. The models fitting the target area in the manipulator configuration space can guide the adaptive sampling; using the GMMs fitting the obstacle region to predict the collision probability of the samples rapidly, and using Greedy K-means initializes the EM algorithm to improve the fitting accuracy of the model. Finally, the sampling strategy and collision detection method are combined with various sampling-based motion planning algorithms, and multiple simulation experiments are carried out to verify the results. The results show that the proposed method has significantly improved planning efficiency compared with the traditional method.