An Adaptive Multimodal Enhanced RRT* Path Planning Algorithm Based on Node Collision Feature Calculation Model
Menghao Li, Liang Shan, Zhichao Wu, Weixi Wang · 2024
This paper proposes an improved RRT*(Rapidly-exploring Random Tree) algorithm, named CBERRT*(Collision-Based Exploration Rapidly-exploring Random Tree Star.), to address issues in conventional RRT* algorithms for mobile robotic arm path planning, such as low node information utilization, poor extension effectiveness, and frequent invalid collision detections. CBERRT* introduces a node collision feature calculation model to dynamically update node and environmental state information in the RRT* tree, enabling adaptive node adjustments. CBERRT* adaptively selects different expansion and sampling strategies based on updated environmental information. When a node is near an obstacle, an obstacle perception-guided strategy expands along the edge of the obstacle. In narrow passages, a fixed direction perception escape strategy uses a three-wheel fixed sampling test to find the optimal escape direction, reducing invalid collision detections. Finally, experiments in a dock container environment demonstrate CBERRT*'s superior performance in mobile robotic arm inspection tasks, reducing invalid collision detections by 20% and improving narrow passage exploration efficiency by 80%. CBERRT* also shows excellent performance in both 2D mobile robot and 3D robotic arm path planning tasks.