A Robot Motion Planning Approach Based on Adaptive Multi-Tree Sampling

Abstract

Abstract Efficient motion planning methods are essential to ensure industrial robots to work properly and safely. Although sampling-based planning algorithms are viable ones, they often struggle to adapt to highly constrained and complex environments. This paper introduces a new robot motion planning approach for such environments, utilizing a multi rapidly-exploring random trees exploration structure. The approach combines the fast exploration property of RRT-based methods with the global exploration property of multi-tree structures. In the subtree generation, an information gain-based method is used to analyze the sampled information from multiple trees to compute the potential information gain at various subtree generation locations. By selecting the locations with higher information gain, our method can effectively improve the exploration quality of the environment. Furthermore, an adaptive local subtree planning method is developed, which relies on local structure information and dynamically updates the sampling distribution to maximize the possibility of forming feasible trajectories in narrow passages. The effectiveness of the proposed approach is tested in 2D, 4D, and 6D environments, along with a complex material picking scenario. These experiments demonstrate that the proposed approach surpasses the performance of other algorithms, particularly in those highly constrained and complex environments. Our study contributes to the development of advanced and highly-adaptive motion planning methods for robots in complex environments.