Task-Oriented Path Planning for Campus Waste-Sorting Mobile Manipulators Using an Improved BiRRT Method

Campus waste-sorting mobile manipulators operate in cluttered environments. This paper presents a task-oriented framework decoupling mobile-base navigation from manipulator motion. Its contribution is the safety-verified integration of goal-biased sampling, adaptive step-size adjustment, artificial potential field (APF)-guided directional correction, collision-recovery direction selection, and collision-checked bidirectional tree connection within Improved-BiRRT. After obtaining a feasible path, visibility-based pruning and piecewise cubic Hermite interpolating polynomial (PCHIP) smoothing are applied, followed by segment-wise collision verification with fallback to the verified pruned path. MATLAB R2024b simulations cover simple campus, complex campus, and narrow-passage environments. RRT, GoalBias-RRT, BiRRT, and Improved-BiRRT are evaluated under identical settings and post-processing. In the narrow-passage environment, Improved-BiRRT achieves a mean raw feasible-path length of 32.155±1.935m, a mean final collision-verified path length of 27.041±0.327m, and 39.25±12.64 iterations. Holm-adjusted Wilcoxon rank-sum tests show significantly lower final path lengths and iteration counts than baselines in the complex campus and narrow-passage environments (padj<0.05). A five-target validation yields a cumulative collision-verified path length of 136.171m and a cumulative MATLAB online planning time of 0.1305s. The results demonstrate feasibility for static two-dimensional campus waste-sorting tasks with predefined target ordering.