The Center for Data Innovation recently spoke with Rossen Kolev, CEO of Smart Farm Robotix, a Bulgaria-based company developing AI-powered agricultural robots that identify and remove weeds without herbicides. Kolev explained how the company combines computer vision, an AI-driven identification system, and autonomous robotics to help farmers reduce manual labor and improve weed control.
Kertai: What inspired the creation of Smart Farm Robotix?
Rossen Kolev: Across organic agriculture, weed control remains one of the most difficult and labor-intensive challenges farmers face. Although organic farming can command higher market prices due to growing consumer demand for chemical-free produce, it also comes with strict standards that restrict chemical herbicides, which limits weed-control options. On farms growing lavender and other aromatic plants, weeds often become a major obstacle to expansion. Farmers rely on manual labor to remove weeds by hand, but this work is slow, expensive, and difficult to scale.
Existing agricultural robots also provide limited help because many systems are either too expensive or designed for the flat farmland common in Northern Europe. As a result, these machines struggle in the hilly, dry, and uneven terrain typical across Southern Europe. For growers trying to expand while protecting delicate young plants, these limitations create a significant barrier.
Smart Farm Robotix addresses this problem by building an affordable autonomous weeding robot for small and medium-sized farms. The compact field robot uses cameras, an AI system, and computer vision to navigate between crop rows, identify weeds, and remove them with high precision in real farming environments. While working with local farmers on sales and marketing, we saw firsthand that automation can fill the labor gap while still removing weeds carefully around sensitive crops.
Kertai: How does your robotic system use cameras and AI to distinguish weeds from crops in the field?
Kolev: The robots use cameras and an AI system trained to distinguish weeds from crops in real time. Computer vision enables the system to identify plants directly in the field while it navigates autonomously between crop rows. We train the models in-house using thousands of images of weeds and crops, then load them onto compact onboard computers inside the robot. Because processing happens locally, the robots operate reliably even without constant Internet access.
The same camera system also guides navigation through the field. Instead of relying on expensive high-precision positioning systems, the robot uses standard GPS mainly to define field boundaries and then switches to image recognition to move between rows and avoid obstacles. This approach proves more reliable and cost-effective in rural and mountainous environments.
Kertai: How do your robots remove weeds without damaging nearby crops?
Kolev: The robots remove weeds using a focused beam of light that damages the weed internally, causing it to dry out and die over the following days without disturbing the soil or harming nearby crops. The beam targets an area of roughly 3 to 4 square centimeters, allowing the robot to remove weeds with high precision while supporting regenerative and no-till farming practices that minimize soil disruption.
Early versions of the robot used a single weeding unit, but we are now testing systems with up to 12 units operating simultaneously. This setup allows the robot to treat more weeds at once while keeping costs lower than many laser-based competitors.
Kertai: What challenges have you faced while developing autonomous agricultural robots?
Kolev: One of the main challenges comes from integrating commercially available components into a unified and reliable robotic system. Suppliers sometimes change parts or specifications without notice, which means a component that works in one prototype may behave differently in the next batch. These inconsistencies create ongoing engineering and testing challenges.
Kertai: Could you share a real-world example of your technology in use?
Kolev: We already work with several farms, particularly those producing lavender and other essential-oil crops common in Bulgaria. The robots deliver the most value during the first and second years of plant growth, when weeds pose the greatest threat and crops have not yet reached full production.
We are also collaborating with vegetable farmers, including both small family farms and larger greenhouse producers. One early adopter near Sofia, Bulgaria, well-known locally for supplying organic produce to farmers’ markets, allowed us to test the robots early under real farming conditions. Using his fields over multiple growing seasons helped us refine the system based on daily agricultural operations.
