PEBFMT*: a path planning strategy for robotic arms under limited planning budgets

Abstract

Robotic arm path planning in complex environments requires planners to find feasible paths quickly and refine path quality efficiently under limited planning budgets. This paper proposes the Piecewise Ellipsoidal Bidirectional Fast Marching Tree (PEBFMT*), a sampling-based path planning strategy for robotic arms in cluttered environments. The method first obtains an initial feasible path through bidirectional fast-marching exploration and then improves the incumbent solution using a staged global-to-local refinement mechanism. Specifically, global informed refinement restricts new samples to the cost-bounded informed region, while piecewise ellipsoidal sampling reallocates part of the sampling budget to local regions along the incumbent path. Across 250 runs in five Open Motion Planning Library (OMPL) benchmark scenes, PEBFMT* achieved a pooled target-cost hit rate of 94.8%. Among the 237 runs that reached the target, the arithmetic mean and median runtime-to-threshold were 14.94 s and 7.77 s, respectively. Experiments on a 7-degree-of-freedom (7-DoF) Sawyer robotic arm further show that PEBFMT* reduces the median path cost by 8.13% compared with Bidirectional Fast Marching Tree (BFMT*) and by 31.24% compared with Batch Informed Trees (BIT*), while maintaining competitive robustness.