From Multimodal Perception to Low-Damage Harvesting: A Review of Embodied Intelligence in Solanaceous Fruit-Picking Robots

With the rapid expansion of modern facility agriculture, harvesting robots for tomato (Solanum lycopersicum L.) and other solanaceous fruits have become key equipment for alleviating labor shortages. However, in unstructured greenhouse and open-field scenarios, branch and leaf occlusion, overlapping fruit clusters, variable illumination, and fragile fruit tissues constrain perception reliability and mechanical contact safety. Embodied intelligence, a closed harvesting loop integrating perception, spatial reasoning, motion control, and physical-contact feedback, offers a promising paradigm to address these challenges. Nevertheless, the transition from prototypes to reliable commercial deployment remains constrained by environmental variability, biological heterogeneity, and system-level integration. This review systematically examines the mechanisms and process architectures for converting multimodal perception into low-damage harvesting actions of solanaceous fruit-picking robots. First, the review analyzes deep-learning-based real-time recognition methods, covering target detection, instance segmentation, and occlusion-adaptive strategies. Second, attention turns to spatial 3D localization and pedicel feature acquisition, including depth sensing and high-precision hand–eye calibration. Furthermore, operation patterns and flexible low-damage end-effectors are examined, clarifying fruit drop collision damage and physical contact models of suction and shearing devices. Finally, technical bottlenecks and future directions are discussed, encompassing multi-modal perception–action coupling, agricultural foundation models, end-to-end closed-loop control, and commercialization requirements. Overall, it links algorithmic perception accuracy with system-level harvesting reliability, proposes a crop-specific, damage-aware evaluation framework, and provides a roadmap for robust, low-damage commercial deployment.