Robotic Systems for Automated Fruit Detection and Harvesting

Automated fruit detection and harvesting systems integrate advanced vision sensors, machine learning algorithms and robotic manipulators to address labour shortages and enhance productivity in horticulture. Central to these platforms are deep convolutional neural networks that process RGB and multispectral imagery to recognise fruit under variable lighting and occlusion. Detected targets are fed to precision end effectors—grippers or cutters—mounted on single- or multi-arm robotic arms carried by mobile bases. Control architectures coordinate perception, planning and actuation, often incorporating human–machine interfaces for supervisory intervention and on-the-fly parameter adjustment. Field trials of greenhouse and open-orchard robots have demonstrated cycle times under 30 seconds per fruit and success rates exceeding 80% in optimised conditions, signalling the potential for commercial deployment. Ongoing challenges include robust fruit localisation amid dense foliage, minimising damage during grasping, and scaling systems to diverse crop types. Emerging solutions leverage multi-view geometry, attention-based segmentation and lightweight dexterous end effectors to further improve accuracy, speed and adaptability across global production contexts.

Research from Nature Portfolio

Recent studies have demonstrated the efficacy of tailored deep-learning models for high-precision fruit detection in real time. A modified YOLOv3 framework, enhanced with dense connectivity, spatial pyramid pooling and advanced activation functions, achieved detection accuracies above 98% for tomato clusters under variable illumination and occlusion, while maintaining inference times below 50 ms. This approach supports deployment on embedded platforms and highlights the importance of architecture optimisation for robust orchard applications.

Robotic Systems for Automated Fruit Detection and Harvesting publication trend

The graph below shows the total number of articles in robotic systems for automated fruit detection and harvesting across all publications each year (not limited to Nature Index journals).

Technical terms

Instance segmentation: Pixel-precise delineation of individual objects in an image to distinguish overlapping fruits and background.

End effector: The device attached to a robot arm—such as a gripper or cutter—used to interact with or harvest the fruit.

Occlusion: The partial or complete obstruction of fruit by leaves, branches or other objects, posing challenges for detection algorithms.

Spatial pyramid pooling: A method that pools features at multiple scales from convolutional layers, enabling flexible input sizes and improved object detection.

Inference time: The elapsed time for a trained model to process an input image and output detection or segmentation results.

References

1. Field Performance of a Dual Arm Robotic System for Efficient Tomato Harvesting. Journal of Robotics Spectrum (2024).
2. Comparing YOLOv8 and Mask R-CNN for instance segmentation in complex orchard environments. Artificial Intelligence in Agriculture (2024).
3. Automatic fruit picking technology: a comprehensive review of research advances. Artificial Intelligence Review (2024).
4. Tomato detection based on modified YOLOv3 framework. Scientific Reports (2021).