Knowledge, Innovation and Technology in Scientific Research Vol. 4

Robotic harvesting has progressed from proof-of-concept machines that separately addressed fruit detection, manipulator motion, or detachment to increasingly integrated systems that combine deep learning, three-dimensional sensing, compliant end-effectors, closed-loop control, and autonomous task planning. Yet commercial horticulture exposes a persistent gap between component-level performance and reliable harvest at crop scale. This critical narrative review evaluates that gap across fruit and vegetable harvesting systems, with emphasis on the coupled perception–action–crop problem rather than on isolated algorithms. Literature published from January 2000 through 18 July 2026 was examined across agricultural, engineering, robotics, and multidisciplinary scholarly sources, supplemented by citation searching and verification of bibliographic metadata. The evidence indicates that modern vision systems have substantially improved fruit recognition, instance segmentation, pose estimation, and picking-point localisation, but field performance remains constrained by occlusion, illumination change, depth uncertainty, fruit motion, clutter, and domain shift. Manipulation has similarly advanced from rigid grippers and simple cutters towards suction, soft and hybrid devices, yet compliance, hygiene, durability, detachment mechanics, fruit damage, and sensing remain tightly crop-dependent. Integrated field studies show that canopy architecture, crop preparation, target eligibility, retry logic, and the definition of success can influence reported performance as strongly as the selected perception or manipulation method. Consequently, cross-study comparisons based only on detection accuracy, harvest success, or per-fruit cycle time are often misleading. The strongest pathway towards commercial deployment is a system-level strategy that combines action-centric perception, uncertainty-aware manipulation, failure recovery, crop–robot co-design, and standardised field evaluation. Future progress should be judged by throughput, fruit quality, uptime, intervention frequency, maintainability, seasonal robustness, and economics, not solely by success on favourable targets.