GINT Attaches “AI Kits” to Existing Farm Machinery to Prevent Farm Work Accidents

According to the Disaster Yearbook of the Ministry of the Interior and Safety, the number of agricultural machinery accidents that occurred over the four years from 2020 to 2023 was 4,708. By year, there were 1,269 cases in 2020, 1,076 cases in 2021, 1,384 cases in 2022, and 979 cases in 2023, with around 1,000 accidents recurring each year. 

 Accidents are concentrated among elderly operators. According to a study by the National Institute of Agricultural Sciences on the status of agricultural machinery traffic accidents, 37% of such accidents occurred among operators aged 70 or older, followed by those in their 60s at 27.6%, and those in their 50s at 17.6%. Most causes were human factors such as failure to watch the road ahead, errors in judgment, operational mistakes, and drinking. The analysis is that slower judgment and response with age lead to accidents. 

The aging of rural communities is becoming increasingly severe. In the Statistics Korea(now the National Data Agency) Agriculture, Forestry and Fisheries Survey, among 973,707 farming households nationwide in 2024, farms whose operators were aged 70 or older accounted for 50.8%, exceeding half. Including those in their 60s(33.0%) brings the figure to 83.8%. As the farm population declines and the amount of work handled by each person increases, this is why some point out that it is difficult to reduce accidents through safety management that relies only on human attention. 

Agricultural autonomous driving company GINT(GINT) has stepped forward to solve this problem with AI. GINT has formed a consortium with Agmo and Dverse and is carrying out the task, as part of the Ministry of Science and ICT’s Public AX project program ‘Zero Farm Work Accidents! AI Agricultural Machinery Safe Operation Service Development·Demonstration’ . This task runs from October 2025(Year 1) to December 2026(Year 2) . The ultimate goal is to turn agricultural machinery into ‘AX-enabled’ equipment by attaching an ‘AI integrated solution kit’ to tractors, sprayers and other machines that farms already own, without requiring them to buy new machinery. 
 

 Warnings for distraction, detouring or stopping when encountering obstacles 

The kit focuses on two types of risk. One is driver carelessness, and the other is unexpected situations encountered during work. The driver-seat camera detects drowsiness, mobile phone use, smoking, and whether the seat belt is fastened. External cameras determine sudden obstacles or terrain that agricultural machinery cannot enter, warn the driver, and inform them how to respond. GINT Principal Engineer Jung Hee-jong explained, “It is a model that prevents accidents by making users as aware as possible of accidents caused by carelessness or insensitivity to safety.” 

GINT explained that obstacle response has already been implemented to a level that does not require human intervention. For example, if an obstacle appears ahead when the ridge on the left has already been completed, the kit makes its own judgment and detours to the right. If it cannot detour, it stops the machine. 

The learning model was initially built with public data from AI Hub, and its performance was improved through field demonstrations. According to GINT, it has surpassed all first-year targets: obstacle recognition accuracy of 75%, route regeneration within 10 seconds, and a driver risky-behavior detection rate of 85%. Principal Engineer Jung said, “It differs by item, but we have secured reliability of more than 90%.”The second-generation kit currently on sale does not have this recognition function. AI models will be included starting with the new product to be released next year. Even after launch, the company plans to update the learning model in the cloud and deliver it to terminals in real time. 
 

 Alarm at a 30-degree tilt, forced stop if it moves for 5 more seconds 

Principal Engineer Jung explained that when rollover accidents occur, many people lose their lives because they are working alone and are found too late. The kit sounds an alarm when the terminal tilts by about 30 degrees. If the machine continues moving for 5 seconds after the alarm sounds, it is forcibly stopped. If the angle then changes rapidly, such as tilting by more than 10 degrees within 1 second, an emergency notification is sent to predesignated contacts. Jung said, “Going forward, we are also reviewing ways to link it with emergency reporting services such as 119 or 112 through policy proposals.” 
 

 New-vehicle option costs 50 million won, kit costs 15 million–20 million won 

The reason for choosing the method of attaching a kit to existing agricultural machinery is cost. Principal Engineer Jung said, “By minimally modifying a tractor, it can deliver autonomous driving performance comparable to expensive tractors,” adding, “For new agricultural machinery, the autonomous driving option costs about 50 million won, whereas the kit can be installed for 15 million–20 million won, saving 30 million–35 million won in initial adoption costs.”It can also be installed on older models without electronic control units. The consortium has set a target price at 25% of the price of high-end autonomous agricultural machinery on the market. Based on its experience installing devices on more than 300 types of agricultural machinery in Korea and abroad, it believes the kit can be applied to more than 90% of agricultural machinery in Korea. 

While carrying out this task, GINT also signed agreements with agricultural machinery manufacturers including LS Mtron Co., Ltd., Hada Co., Ltd., and Asia Technology Co., Ltd. . The agreements state that manufacturers will also adopt this technology if the kit proves performance above a certain level. 
 

From resistance to acceptance… demonstration at 8 rental centers 

The demonstration has been underway since last September at 8 agricultural machinery rental centers, with one unit at each center, consisting of 7 tractors and 1 sprayer(SS machine). The reason for choosing rental centers is data. Principal Engineer Jung explained, “An individual farm operates a tractor for less than 600 hours a year, but rental centers operate them 3–4 times more than that,” adding, “We chose rental centers to collect as much data as possible in the early stage.”

The response in the field was not positive from the beginning. Jung said that the more experienced farmers were with agricultural machinery, the more they tended to have a habit of driving while distracted, so they felt significant inconvenience whenever warnings appeared, saying, “In the early days of the project, some even told us to take it off.” There was also strong resistance to stopping the machine regardless of the user’s will, so for some farms, the stop function was removed and only the detour function was included. 

Such complaints have now subsided. Jung said, “There was some discussion during the familiarization process right after installation, but currently no farms are filing claims.” However, cases of misrecognition continue to be reported, such as sunlight reflected off plastic being mistaken for a puddle, and these are being supplemented through retraining. Citing Tesla’s Full Self-Driving(FSD) as an example, he said, “There may be people who have never used it, but no one uses it only once,” adding, “Once users begin to feel the convenience of having to lift even one finger less, trust in the product will develop.” 
 

Data preservation and responsibility remain challenges 

The task has also left challenges. Camera footage generates about 20GB based on 8 hours of operation per day. The kit terminal does not store data and sends only necessary data directly to the server. Due to privacy issues, in consultation with the farms participating in the demonstration, data used for learning is deleted once learning is complete. Jung said, “AX projects leave behind video data, but it is regrettable that we cannot retain it, making it difficult to use when developing other products.” 

AI has not yet determined who would be responsible if it misses a danger and an accident occurs. Jung pointed out, “Clear regulations are needed on whether the manufacturer or the user will be responsible.” GINT explains that once a system is in place to determine the responsible party, functions can be applied more actively in line with the control capabilities already secured. As the budget for the agricultural machinery rental program has been transferred to local governments starting this year, conditions for distribution will also vary depending on local government finances. GINT is consulting with local governments while also making proposals to the Ministry of Agriculture, Food and Rural Affairs so that, after certification, the kit can lead to subsidized projects. 

The expected effect is limited to accidents inside farmland. Because the kit’s autonomous response function is designed on the premise of work in farmland, accidents during road driving are excluded. Jung said, “If an accident is predicted, the final decision the machine makes is to stop,” adding, “If it detects 9 out of 10 cases, it stops in all 9 unless the user takes separate action, so if the recognition rate is 90%, the prevention rate can also be seen as 90%.”He added, “As data accumulates and updates continue, it will rise to 91%, 92%,” and, “If performance is guaranteed to the extent that users trust the kit, we believe 100% can also be achieved.” 

However, this prevention rate applies only to situations in which the kit can recognize the danger. Machines with the stop function removed do not stop even if they detect danger, and accidents that are not captured by cameras, such as a body part being caught in a rotating part, are not included in the calculation. 

Regarding future plans, GINT Senior Manager Roh Dong-won said, “The task ends in December this year, so we will sequentially end the pilot service from late November, and we plan to transfer ownership of the outputs developed through this project to the ministries and other bodies that carried out the project.” He added, “We plan to advance the developed AI functions so that farmers can use them without resistance and commercialize them in the first half of next year.”