Engineered exclusively for delicate fruit harvesting under tight greenhouse constraints.
RoboCilek is a purpose-built autonomous strawberry harvesting system, not a generic agricultural machine. We combine advanced camera-based perception, real-time ripeness logic, precision robotic motion, and a unique stem-cut end-effector to identify harvest-ready fruit and collect it without ever touching the berry.
Engineered exclusively for delicate fruit harvesting under tight greenhouse constraints.
The custom end-effector is designed to strictly avoid direct pressure on the berry.
Smart perception and planning run seamlessly above deterministic, low-level physical actuation.
Total accessibility for non-technical users isn’t a feature; it is a fundamental design requirement.
Every feature of RoboCilek is a direct response to real farmer needs. Each capability exists solely because it decisively improves harvest quality, operational clarity, or practical fit within real cultivation conditions.
Advanced camera-based perception identifies candidate fruit and isolates the exact harvest target before a single motor moves.
The robot is intelligently designed to harvest only commercially ready fruit, rather than blindly picking anything visible in the frame.
A multi-degree robotic arm ensures perfectly controlled motion inside a defined workspace, completely eliminating unsafe reach attempts.
The surgical end-effector is built to approach the stem directly, drastically minimizing any contact with the delicate fruit surface.
The system operates independently on battery power, removing the need for constant, expensive external infrastructure during operations.
Intuitive start-stop control and clear on-device status feedback ensure the system is immediately understandable for non-specialist operators.
RoboCilek does not rely on vague promises of autonomy. We execute an explicit, bulletproof chain of perception, filtering, coordinate conversion, safety checks, and highly controlled actuation.
The RealSense D435i captures aligned RGB and depth data, allowing the robot to simultaneously assess both visual appearance and physical distance.
YOLOv8n-seg generates precise instance masks, while our custom HSV logic strictly verifies that the unripe signal ratio is low enough for a premium harvest.
Depth pixels inside the target mask are processed with robust IQR filtering and reduced to a stable median distance, effectively neutralizing sensor noise.
Complex deprojection and extrinsic calibration instantly translate camera-space visual information directly into the robot’s physical base frame.
Strict workspace checks, inverse kinematics, joint limits, and calculated stem offset logic absolutely prevent any unsafe or fruit-damaging movements.
The embedded FSM flawlessly executes the approach, stabilize, cut, and home states in sequence, ensuring every single harvest is deterministic and perfectly repeatable.
True value in harvesting comes from absolute fruit safety. That is why our end-effector strategy is uncompromising: RoboCilek targets the stem with a calculated offset and clean cutting action, intentionally keeping the delicate fruit body completely out of the process.
Applying direct pressure to strawberries causes bruising, severely shortens shelf life, and immediately strips the fruit of its premium export value.
Our motion objective is surgical: locate the berry, confidently infer the stem approach, and execute the cut while keeping the fruit untouched and safe.
Our stationary validation prototype deliberately masters precise execution within a tightly controlled workspace, ensuring reliability before scaling up complexity.
We strike the perfect engineering balance: proven off-the-shelf components accelerate our physical prototyping, while the core harvesting intelligence and logic remain 100% proprietary to our team.
Intel RealSense D435i provides flawless RGB-depth sensing, backed by YOLOv8n-seg and our StrawDI dataset for rapid instance-level detection.
Our proprietary HSV-based unripe signal ratio logic acts as the brain, definitively deciding whether a piece of fruit is truly ready for market.
Camera-space coordinates are mathematically converted into precise robot-space targets, strictly validated before a single motor is allowed to move.
High-level strategic planning runs powerfully on a Raspberry Pi 5, while fast, deterministic low-level actuation is securely handled by an Arduino Nano.
Precision servo-driven actuation, custom-engineered geometry, and our specialized stem-cut tool effortlessly translate AI perception into physical action.
Standardized electronics guarantee rapid iteration, while the critical arm geometry, mounting logic, and harvesting mechanism are entirely team-designed.
We don’t oversell our technology. We deliver strict controlled-environment validation, stable selective harvesting logic, and uncompromising safety behaviors—while remaining totally transparent about our roadmap.
Extensive controlled tests prove our vision system can detect, segment, and classify strawberries with absolute consistency within the prototype envelope.
Thousands of repeated cycles demonstrate rock-solid positioning behavior, fully supported by conservative motion sequencing and hard joint limits.
Rigorous workspace checks, unreachable-target rejection, soft limits, and communication watchdogs virtually eliminate the chance of undefined behavior.
Our cutting sequence is explicitly designed around preserving fruit quality, refusing to maximize raw, dangerous speed at the expense of your crop.
We believe in building in public. Here is exactly what RoboCilek flawlessly executes today, and the ambitious milestones we are targeting next.
You’ve seen our coherent perception-to-harvest logic in action. Now, dive deeper into the technical architecture that powers it, or explore how our business model is designed for rapid deployment.