[Farm Machinery of the Week] Wipes Out 4,000 Weeds in Just 1 Second… ‘Pixel-Level AI Weeding Robot’ Appears

An intelligent AI weeding robot has emerged to solve the difficulties of open-field weeding work. Professor Kim Hyeong-seok’s team at the Intelligent Robotics Research Institute of Jeonbuk National University has gone beyond the limits of existing global technologies, developing a prototype precision weeding robot optimized for Korean farm environments and completing a demo demonstration.

Until now, drone spraying and autonomous driving have been commercialized, but weeding technology that distinguishes crops from weeds in real time remained uncharted territory. The research team found a solution that overcomes the limits of human vision by converging electronic engineering with artificial intelligence and robotics technologies.

Combining a 64-bit GPU with inkjet principles… identifies weeds even when work conditions and shapes change
This solution identifies weeds even as crops grow and change shape from immediately after sowing up to 37 days later. It also accurately distinguishes weeds even in pitch-black shadows created by strong sunlight during the summer rainy season, and at night it can turn on a searchlight and work around the clock. Furthermore, it is precise enough to classify weeds into three types (grass forms) and prescribe dedicated herbicides suited to each. It can also identify accurately even when surrounding conditions change, such as variations in sunlight or at sunset.

Professor Kim Hyeong-seok said, “The human brain has only about 32 levels of discrimination (5 to 6 bits) for a single pixel, but with just 10 bits, it can express an almost infinite range of numbers. Furthermore, the AI system has almost no error thanks to the computing power of HBM memory and the latest 64-bit GPU, and it calculates and identifies areas accurately at the pixel level.” He added, “Experimental results confirmed that it is better than humans at distinguishing weeds.”

The ‘non-contact spraying’ method used by foreign global companies had the problem of herbicide drift in windy open fields, causing damage that could kill the main crop as well. The global ‘laser weeder’ introduced as an alternative is precise and eco-friendly, but the machine costs 2 billion won. In addition, because its moving part is a robotic arm, there is a speed limitation. In fact, its weed removal speed is only 27.8 weeds per second, making it difficult to apply in fields overgrown with weeds.

In response, the research team boldly introduced the ‘inkjet printer principle.’ As the robot moves, AI calculates where the herbicide should fall and precisely sprays it only onto the center of the weed. In addition, through 150 nozzles arranged at 8 mm intervals, it removes as many as 4,000 weeds per second. This is a level at which perfect weeding is possible at normal walking speed. It is currently being demonstrated for field soybeans, and the plan is to expand its application range to fruit trees and other crops.

A weeding robot that does the work of six brush cutters
The research team’s weeding robot can operate without human supervision, and its weeding speed is 0.09h per hour (about 272 pyeong), making it possible to weed 0.72h (2,178 pyeong) over 8 hours. Compared with the work area covered by one person using a brush cutter, it can handle 2.5 to 3 times more per hour and 4 to 6 times more per day (based on 8 hours). Recently, the team has also been preparing an eco-friendly weeder equipped with a brush cutter instead of spraying herbicide from the robot platform’s nozzles.

However, despite the technical success, challenges still remain before commercialization.
Professor Kim Hyeong-seok said, “The ‘mechanical durability’ needed to withstand the rough handling by farmers is still at a stage that requires improvement.” He continued, “For commercialization, it must be connected with companies,” adding, “There needs to be a process of improving the industrial structure through open innovation in which the government promotes joint commercialization with university research institutes and startups that possess innovative technologies.”