High Intensity Illumination Systems and Methods of Use Thereof Powerful, Uniform Lighting for Vision-Based Robotic Weeding

This patent covers high intensity illumination arrays and the camera and processing pipeline they feed, designed to make vision-based detection of plants in a field accurate enough for autonomous weed-targeting machinery. By controlling the lighting that hits each region of interest, the system reduces the variability that normally hobbles outdoor computer vision and the volume of training data needed to keep object detection models reliable.

The Problem

Automated object detection in outdoor images is heavily dependent on consistent imaging conditions. In agriculture, ambient sunlight changes by the minute, shadows fall across crop rows, and dust, dew and reflective leaf surfaces all add noise. Training a detector that performs across every combination of these conditions takes very large datasets, and any model still degrades at edge cases such as dawn, dusk or overcast skies. For a precision agriculture machine that has to identify individual weeds and direct an implement or laser at them, missed or misclassified plants translate directly into crop damage, missed weeds or wasted energy. A way to make the imaging environment more predictable, rather than asking the machine learning to compensate for it, would meaningfully lift performance.

What This Invention Does

The patent describes a lighting array consisting of multiple lights arranged to evenly illuminate a defined region of interest in a field. Coupled to that array is a detection system with a camera that captures images of the region under the controlled lighting, processors that determine the location of each target plant from those images, and an implement that physically acts on the plant at the determined location.

By controlling the spectrum, intensity and uniformity of the light reaching the inspection area, the system gives the camera and the downstream classifier the kind of stable visual input usually only achievable in laboratory or warehouse settings. The detection pipeline then maps each identified plant to a target coordinate and dispatches the implement, which in the applicant’s commercial system is a laser weeding head.

Key Features

• Even illumination of the region of interest. A multi-light array engineered to deliver uniform brightness across the area the camera sees, reducing variability that would otherwise need to be absorbed by the detector.
• Camera plus processor stack. A vision sensor and onboard compute that turn the controlled-illumination images into target coordinates in real time at field speeds.
• Coupled implement. A device, in the applicant’s product line a laser, that acts on each detected plant at its determined location without disturbing surrounding crop.
• Independence from ambient conditions. Reduced reliance on natural lighting widens the operating window for autonomous field machinery to dawn, dusk and overcast conditions.
• Reduced training data burden. Because the imaging environment is held more constant, the machine learning models behind detection need fewer corner-case samples to remain accurate.

Who Is Behind It

The applicant is Carbon Autonomous Robotic Systems Inc., a Seattle-based agricultural robotics company best known for its LaserWeeder platform, which uses computer vision and high-power lasers to remove weeds in vegetable crops without herbicides. The named inventors are Alexander Sergeev and Paul Mikesell, both engineering leaders at Carbon Robotics with backgrounds in distributed systems and computer vision. The Australian application is a divisional of AU 2022380472. The Australian patent attorney of record is Spruson and Ferguson in Sydney.

Why It Matters

Herbicide-free weeding is one of the highest-value targets in precision agriculture, with implications for organic production, herbicide resistance management and chemical exposure for farm workers. Australia has a large broadacre and horticultural sector with significant interest in autonomous machinery, and protecting the underlying imaging stack here positions Carbon Robotics for direct sales and licensing into the local market. The detail being protected, controlled illumination plus detection plus targeted implement, is the sort of system-level patent that tends to be broadly enforceable across competitor designs in this space.

Related Concepts

• Computer vision in agriculture – the broader application area this invention sits in.
• Object detection – the underlying ML problem the controlled lighting is intended to make easier.
• Laser weed control – the actuator side of Carbon Robotics’ commercial product.
• Convolutional neural network – the typical model class used for the plant classifier behind such systems.
• Integrated pest management – the agronomic framework that herbicide-free weeders contribute to.

---

AU 2026201867 was published in the Australian Official Journal of Patents on 2 April 2026 and is open for public inspection. Patent applications represent inventions that are sought to be protected and do not necessarily reflect commercially available products.