Fruit and plant vision
RGB, near-infrared, thermal, depth, end-effector cameras, and controlled illumination.
Engineering architecture
Technology built around the field
Multisensor perception, local compute, plant-level memory, modular control, independent safety, and serviceable agricultural hardware.
05
Perception suite
Different sensors answer different questions.
The prototype architecture reserves positions, power, timing, and compute for the full sensor suite while allowing early testing to begin with the smallest validated subset.
Navigation and protection
LiDAR, RTK GNSS, IMU, wheel odometry, proximity sensing, and obstacle classification.
Crop and soil intelligence
Multispectral and hyperspectral provisions, environmental sensing, and a physical soil probe.
Onboard and local compute
Fast decisions on the robot. Persistent intelligence in HIVE.
A Raspberry Pi-class supervisory computer with a Hailo accelerator is a plausible prototype path. Dedicated real-time controllers handle motors, force limits, scissors, safety outputs, and deterministic machine functions.
The AI perception stack must never be the only path to a safe stop. Safety-related controls remain independently designed, verified, and validated.
Perception
Images, depth, force, position, and context.
Edge AI
Detection, grading, segmentation, and confidence.
Real-time control
Motion, grip, snip, stop, and machine state.
Safety layer
Independent limits, guarded scissors, E-stop, and protective response.
HIVE
Mission context, records, models, service, and orchestration.
Plant digital twin
The system remembers the plant, not just the image.
Each plant can retain a longitudinal record of imagery, flowering, fruit development, interventions, soil observations, harvests, defects, and yield. “Good fruit” comparison should combine validated reference examples with the plant’s own history and local crop context.
PlantPersistent identity and location
FruitDetection, maturity, grade, and outcome
PackageWeight, occupancy, case, and pallet association
CycleLongitudinal field and distribution timing
Evidence and standards
Development must be measured against real safety and food-handling obligations.
The prototype is not certified. These sources define important parts of the verification program and public claims discipline.
ISO 18497-1:2024Machine design principles and vocabulary for partially automated, semi-autonomous, and autonomous agricultural machinery.
ISO 18497-2:2024Design principles for obstacle protection systems.
FDA FSMA Produce Safety RuleScience-based standards for growing, harvesting, packing, and holding covered produce.
NIOSH ergonomics and musculoskeletal healthWork design and risk factors involving repetitive motion, force, and awkward posture.