Agricultural Robot Selective Harvesting Vision 2026
Agricultural Robot Selective Harvesting Vision
Computer vision and deep learning are driving selective harvesting robotics from laboratory prototypes toward commercial deployment. This dataset spans 1998–2026, mapping core technology clusters and leading patent assignees.
Selective Harvesting Robotics: From Lab to Field
Selective harvesting computer vision integrates imaging hardware, machine learning inference, 3D localization, and robotic arm control to identify, assess, and physically retrieve individual ripe crop items while leaving immature ones in place. This selectivity based on machine-assessed maturity is the defining distinction from bulk mechanical harvesting across the retrieved dataset.
The core technical stack in this dataset involves RGB, stereo, depth, and multispectral cameras mounted on robotic platforms; CNN-based inference using YOLO variants, SSD, and MobileNet architectures for fruit detection and maturity assessment; 3D spatial localization for picking-coordinate determination; and visual servoing for robotic arm execution during physical harvest.
The 2021–2026 filing cluster in this dataset represents the highest commercial specificity, with active patents from Automated Harvesting Solutions LLC (broccoli), Bonsai Robotics Inc. (orchard autonomous harvesting), and Deere & Co. (edge-computing vision). This cluster signals the technology is transitioning from R&D toward near-commercial maturity across multiple crop categories.
In this dataset, the United States is the dominant jurisdiction by filing count, with active patents from Deere & Co., Automated Harvesting Solutions LLC, Bonsai Robotics Inc., FFRobotics Ltd, University of Pennsylvania, and Carnegie Mellon University. WO (PCT) filings appear across multiple assignees in retrieved records, indicating commercial intent to protect internationally.
Technology Cluster Distribution and Filing Activity Over Time
The retrieved dataset reveals four major technology clusters — deep learning detection, 3D vision localization, edge computing inference, and multi-robot architectures — with filing activity accelerating sharply from 2021 onward across commercial-grade platforms.
Patent Filings by Technology Cluster (Dataset Snapshot)
In this dataset, deep learning-based crop detection accounts for the largest cluster, followed by 3D vision localization, edge computing inference, and multi-robot architectures — reflecting the commercial prioritization of AI-driven maturity assessment as of 2021–2026.
↗ Click bars to exploreFiling Activity by Era — Agricultural Selective Harvesting (Dataset Snapshot)
In this dataset, filing activity is concentrated in three eras: foundational (1998–2011, ~4 filings), mid-stage development (2012–2020, ~12 filings), and recent commercial cluster (2021–2026, ~26 filings), confirming an accelerating trajectory toward commercial deployment.
↗ Click bars to exploreKey Crop Domains in Selective Harvesting Computer Vision
The retrieved dataset spans five major crop application domains — from high-attention orchard fruits to commercial-specific brassica harvesting — each representing distinct technical and commercial challenges addressed by named assignees and reviewed literature.
Orchard Fruit Harvesting
Orchard fruit is the most heavily represented domain in the dataset, with Bonsai Robotics Inc. filing apple orchard autonomous harvesting patents (US 2025, US/WO 2026) using monocular vision and virtual ray projection for tree shake-point localization. FFRobotics and FFMH-Tech patents target generic tree fruit via multi-robot sector-based platforms filed in 2016. Literature reviews confirm strawberry, apple, citrus, and grape as the most studied globally.
Orchard RoboticsGreenhouse Vegetable Harvesting
Greenhouse vegetables represent the second major domain, encompassing tomato, pepper, cucumber, and watermelon harvesting. The 2023 hybrid visual servo control study for cherry tomato harvesting used an RGB-D eye-in-hand camera with a cutting-and-clipping end effector. A 2022 watermelon harvesting robot prototype using YOLOv5s-CBAM detection achieved 89.8% detection accuracy and 93.3% harvesting success rate in simulation.
Greenhouse RoboticsBroccoli Selective Harvesting
Automated Harvesting Solutions LLC filed a multi-jurisdiction patent family (US 2021, WO 2021, CA 2021, WO 2022, US 2023) exclusively targeting broccoli selective harvesting — representing notable commercial specificity in this dataset. The system uses imaging-based maturity detection to identify edible crowns and robotic arms with cutting end effectors to harvest only mature broccoli across multiple rows while the machine continuously traverses. The US 2023 filing is active.
Field Vegetable HarvestingPrecision Scouting and Yield Estimation
Deere & Co. and Precision Planting LLC have filed patents using vision robots as pre-harvest intelligence platforms. Precision Planting LLC’s CA 2023 patent applies CNN-based 3D anchor point projection for plant detection and classification across field scans. Their AR 2024 filing uses augmented reality to project 2D reference points onto a 3D ground plane for high-confidence ML-based plant inference, representing a novel sensor-fusion modality.
Pre-Harvest AIKey Patent Assignees in Selective Harvesting Computer Vision (Retrieved Records)
In this dataset, Deere & Co. holds the highest filing volume with 5+ active US patents across edge-computing vision and task auditing, while Automated Harvesting Solutions LLC holds the broadest jurisdiction footprint for a single crop-specific selective harvesting system across US, WO, and CA in retrieved records.
Top Assignees by Filing Count — Selective Harvesting Vision (Dataset Snapshot)
↗ Click bars to exploreDeere & Co.
Deere & Co. holds the highest filing volume in this dataset with 7 retrieved US patents spanning 2021–2024, covering edge-based crop yield prediction (2021, 2022, 2023) and auditing task performance via pre- and post-task vision image comparison (2023, 2024). Their filings consistently apply on-robot ML inference without cloud dependency, with the most recent active patent filed in 2024. All retrieved filings are US-jurisdiction and reflect a focused edge-computing vision strategy.
United StatesAutomated Harvesting Solutions LLC
Automated Harvesting Solutions LLC holds the broadest jurisdiction footprint for a single crop-specific selective harvesting system in this dataset, with 5 retrieved filings across US (2021, 2023), WO (2021, 2022), and CA (2021) — all targeting broccoli selective harvesting. Their system uses an imaging-based machine learning model to assess crown maturity and direct robotic cutting arms across continuous row traversal. The US 2023 filing is active.
United StatesSignals from the 2024–2026 Filing Frontier
The most recent filings in this dataset — spanning 2024–2026 — reveal five converging directions that indicate where commercial and IP competition will intensify: monocular vision, augmented reality sensor fusion, real-time ripeness classification, adaptive IoT-AI integration, and proprietary image dataset IP.
Monocular Vision Replacing Stereo for Orchard Navigation
Bonsai Robotics Inc.’s 2025–2026 patent family demonstrates monocular-only machine vision triangulating tree structure features by virtual ray projection, combined with satellite imagery for coarse positioning. This architecture eliminates the cost and calibration complexity of stereo rigs. A US 2026 pending patent in this family is among the most recent filings in the dataset.
Augmented Reality as a CNN Inference Data Layer
Precision Planting LLC’s AR 2024 patent uses augmented reality-projected 3D anchor grids fed directly into CNN inference pipelines, representing a novel sensor-fusion modality for selective detection in dense crop canopies. This approach projects 2D reference points onto a 3D ground plane to enable high-confidence ML-based plant inference during field exploration.
Deep Learning Detection vs. 3D Vision Localization: Core Cluster Comparison
Click any row to explore further.
| Dimension | Deep Learning Detection (Cluster 1) | 3D Vision Localization (Cluster 2) |
|---|---|---|
| Primary Function | Detect individual fruits/crops, assess ripeness, trigger selective harvest decision | Determine precise 3D picking coordinates of detected crop items for robotic arm guidance |
| Key Architectures | YOLO variants, SSD, MobileNet, CNN-based classifiers | Stereo cameras, RGB-D depth cameras, structured light, LiDAR-camera fusion |
| Representative Assignees (Dataset) | Automated Harvesting Solutions LLC, Rajalakshmi Engineering College, Muthayammal Engineering College | ARO Volcani Institute, Bonsai Robotics Inc., University of Pennsylvania |
| Filing Era in Dataset | Dominant in 2021–2026 commercial cluster; YOLO experiments from mid-stage 2012–2020 | Literature foundations from 2014; active patents 2017–2026 |
| Key Challenge | Maintaining detection accuracy under lighting variability, occlusion, and plant shape diversity in field conditions | 3D localization accuracy under natural lighting variability and occlusion from stems and leaves |
| Benchmark Performance | YOLOv5s-CBAM: 89.8% detection accuracy, 93.3% harvesting success in simulation (2022 watermelon study) | Angular cutting-point error quantified in ARO Volcani Institute laser-vision system validation (US 2022) |
| Deployment Stage in Dataset | Transitioning to commercial-grade: Automated Harvesting Solutions US 2023 active; Ceres Innovation US 2024 | Active patents from ARO (US 2022, 2024) and Bonsai Robotics (US 2025, WO 2025); monocular approach emerging as lower-cost alternative |
Frequently Asked Questions: Agricultural Robot Selective Harvesting Vision
Selective harvesting refers to identifying, assessing, and physically retrieving individual ripe or mature crop items while leaving immature ones in place, based on machine-assessed ripeness or maturity. This selectivity is the defining feature separating this technology cluster from bulk mechanical harvesting, according to the retrieved dataset.
The dataset identifies YOLO variants, SSD, and MobileNet as the dominant CNN architectures used for fruit and crop detection and maturity assessment. A 2022 watermelon harvesting study used YOLOv5s-CBAM, achieving 89.8% detection accuracy. A 2023 literature review documents the progression from hand-crafted features to deep learning superiority in complex orchard environments.
In this dataset, Deere & Co. holds the highest filing volume with 7 retrieved US patents (2021–2024) covering edge-based crop yield prediction and task auditing. Automated Harvesting Solutions LLC holds the broadest jurisdiction footprint for crop-specific selective harvesting, with 5 filings across US, WO, and CA targeting broccoli harvesting. Bonsai Robotics Inc. holds 4 filings focused on orchard autonomous navigation (US 2025, US/WO 2026).
The 2021–2026 cluster in this dataset represents the highest commercial specificity, with active patents from Automated Harvesting Solutions LLC, Bonsai Robotics Inc., and Deere & Co. This period shows a decisive shift toward commercial-grade selective harvesting platforms, compared to the mid-stage R&D focus of 2012–2020 and foundational work of 1998–2011.
Edge computing enables ML inference to occur directly on the robot platform, eliminating cloud round-trips for low-latency harvesting decisions. Deere & Co.’s US patents from 2021–2024 repeatedly apply on-robot ML inference for crop yield prediction and post-task performance auditing using pre- and post-task vision image comparison. The dataset describes edge computing as the architecture of record for 2024–2026 filings.
Monocular vision uses a single camera — rather than a stereo pair — to triangulate spatial positions, in Bonsai Robotics Inc.’s case by projecting virtual rays from the identification system pose to locate tree shake points. This eliminates the cost and calibration complexity of stereo rigs. Bonsai Robotics’ 2025–2026 patent family makes this approach significant as a potential cost-disruption vector for the stereo-camera-dominated prior decade.
Data and insights on this page are based on a limited patent and literature dataset and are for reference only. Figures may not represent the complete technology landscape.