Work

Execution-Aware NBV Planning via joint-space clustering for apple perception in orchard-like environments

Year: 2026

Type: article

Abstract: In orchard-like apple perception scenarios, occlusion caused by leaves and branches often degrades apple localization, making Next-Best-View (NBV) planning important for reliable perception with eye-in-hand manipulators. However, due to joint limits, kinematic singularities, and joint-space discontinuities, an information-optimal viewpoint in task space does not necessarily correspond to a stable and efficient executable motion, revealing a mismatch between perceptual optimality and execution robustness. To bridge this gap, we propose a dual-layer NBV framework for apple perception under orchard-like occlusion conditions. Layer-1 generates task-oriented candidate viewpoints under geometric and visibility constraints in task space (i.e., SE(3)). Layer-2 maps these candidates into joint space and uses clustering as a practical grouping heuristic to organize sampled IK-feasible configurations. The grouped candidates are then evaluated using joint-space motion cost and conservative kinematic margins, with Predicted Surface Coverage (PSC) retained as a lightweight perception-side ranking signal in the downstream execution-aware evaluation. Experimental results show that, compared with an IK-filtered viewpoint-level baseline, the proposed L1+L2 pipeline improves planning success rate from 73.3% to 93.3%, reduces joint-space motion cost by approximately 40%, and increases joint-limit safety margins by about 28%, while maintaining comparable planning time. These results highlight the practical value of introducing a joint-space organizational layer for execution-aware NBV selection under controlled, static, single-target mock-orchard conditions.

Source: Smart Agricultural Technology

DOI: 10.1016/j.atech.2026.102220

Authors: Jinxing Niu, Chang Liu, Jie Yang, Yuhang Wang, Dingyi Ma, Tao Zhang

Institution: North China University of Water Resources and Electric Power

Language: English

Topic: Smart Agriculture and AI

Subfield: Plant Science

Field: Agricultural and Biological Sciences

Domain: Life Sciences

Open Access status: gold

Estimated APC paid: $1,770

Funders: Science and Technology Department, Henan Province, Henan Provincial Science and Technology Research Project