A Gaussian Mixture Models based Multi-RRTs method for high-dimensional path planning

Sampling based motion planning methods such as Rapidly-exploring Random Trees (RRT) are effective for high-dimensional robot motion planning problem. In these methods, how to draw samples and select trees to extend or connect has greatly influence in efficiency. In this paper, a Gaussian Mixture Models (GMM) based Multi-RRTs method (GMMM-RRTs) is proposed for robot path planning, which accelerate the planning procedure with experiences. Firstly, the GMM is adaptively learned with the experiential paths. Secondly, multiple trees are constructed at the centres of GMM components. Then, the optimal trees are selected to extend based on heuristic search algorithm, and bias sampling with the selected GMM components. GMMM-RRTs can efficiently exploit local space while maintaining the efficiency of global path planning. Simulation and experimental results show the effectiveness of the proposed GMMM-RRTs algorithm.