UNCLE-Grasp: Uncertainty-Aware Grasping of Leaf-Occluded Strawberries

Abstract

Robotic strawberry harvesting remains challenging under partial occlusion, where leaf interference introduces significant geometric uncertainty and renders grasp decisions based on a single deterministic shape estimate unreliable. From a single partial observation, multiple incompatible 3D shape completions may be plausible, such that grasps deemed feasible on one completion can fail on another. This paper presents an uncertainty-aware grasping pipeline for partially occluded strawberries that explicitly models geometric uncertainty arising from both occlusion and learned shape completion. The proposed approach employs point cloud completion with Monte Carlo dropout to sample multiple shape hypotheses, generates candidate grasps for each completion, and evaluates grasp feasibility using physically grounded force-closure metrics. Rather than selecting a grasp from a single shape estimate, feasibility is aggregated across completions and a conservative lower confidence bound (LCB) criterion is used to decide whether grasping a strawberry should be attempted or safely abstained. The method is evaluated in simulation and on a physical robot under increasing levels of synthetic and real leaf occlusion. Experimental results demonstrate that uncertainty-aware decision making enables reliable abstention from high-risk grasp attempts under severe occlusion while maintaining robust grasp execution when geometric confidence is sufficient, outperforming deterministic baselines in both simulated and physical robot experiments.