Answers are grounded in this paper and cite their sources.
Reliable robotic strawberry harvesting depends on accurate mature-fruit recognition and the estimation of picking poses that satisfy the geometric constraints imposed by the detachment mechanism and surrounding canopy. This review critically examines the relationship between visual recognition and three-dimensional (3D) picking-point localization in strawberry-harvesting robots. Using a narrative technical synthesis, it compares methods for maturity and pickability assessment, fruit-instance segmentation, anatomical landmark representation, depth acquisition, geometric reconstruction, coordinate transformation, and perception-guided manipulation. Cross-crop evidence is considered without assuming equivalent performance in strawberries. The analysis highlights distinctions among image-level recognition, metric localization, tool-pose estimation, and physical harvesting success. Occluded peduncles, invalid or mixed depth, color–depth registration errors, hand–eye calibration uncertainty, and sensing-to-actuation delays may compromise the accuracy and executability of a predicted picking pose. Furthermore, heterogeneous cultivars, cultivation systems, target materials, end effectors, and success criteria limit direct comparison of reported results. By relating visual representations to geometric observability, error propagation, tool-specific accessibility, and execution feedback, this review establishes an action-oriented framework for evaluating the evidence. It further identifies priorities for uncertainty-aware perception, explicit anatomical landmark definitions, independently validated 3D localization, and real-fruit experiments that jointly report positioning error, detachment success, fruit damage, and cycle time.
No takes yet. Share an insight, caveat, or question.
Jiang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: