ROBBRIEF
DJI's Enterprise Drone Onboard AI Challenge names 15 projects that move inference off cloud servers and onto the drone body, putting decisions in flight for crops, cracks, and rescue work.
A drone flies low over a 50-hectare banana plantation in Colombia, counts the plants as it goes, and lands with a number the operator can act on that afternoon. Daniel Tovar's AgroCount AI does the counting with computer vision and geospatial analysis on imagery captured during the flight, replacing a manual tally that would otherwise run for days after the drone returns. AgroCount is one of 15 winning entries in DJI's Enterprise Drone Onboard AI Challenge 2026, and the pattern across them is the point: the AI runs on the drone body, not on a server the drone has to call after landing.
Other winning projects handle bridge inspection by flagging cracks as the drone flies back, support search-and-rescue by classifying debris in real time, read pollution off river surfaces, and map road networks. DJI invited developers and companies to build models that run on its enterprise drone platforms, with developer tools and deployment support provided.
The shift matters because time-sensitive jobs get decisions in flight instead of after the upload. Emergency response, infrastructure checks, and precision agriculture are exactly the workflows that benefit from on-drone analysis. DJI's framing is that frontline operators know their workflows best, so the contest rewards practical applications built by the people who would actually fly them.
The 15 winners are proof-of-concept demos on DJI hardware in a DJI-curated contest, not independently benchmarked field systems. Treating the showcase as a category verdict would over-read the evidence.