DJI Names 15 Winners In Its Drone Onboard AI Challenge 2026

DJI announced the winners of the DJI Enterprise Drone Onboard AI Challenge 2026 on 22 August 2026, recognising 15 organisations building AI models that run on the aircraft itself rather than in the cloud. The entries cover crop counting, traffic management, pollution tracing, coastal litter identification, bridge crack detection and search and rescue. Here is what won, what the hardware behind it looks like, and why onboard processing is becoming the interesting part of enterprise drone work.

What The Drone Onboard AI Challenge Set Out To Do

The competition asked developers to build models that execute during flight, using computing carried on the aircraft. That constraint is the whole point. A model that has to send imagery to a server before it can answer a question is limited by connectivity, latency and data volume, and none of those are reliable on a rural inspection or a coastline survey.

Onboard inference flips the sequence. The aircraft analyses what it sees while it is still flying, so the operator gets a result rather than a folder of images to process later.

The Two Award Categories

DJI split the 15 winners across two categories. Five organisations took the Best Onboard AI Model Award for models optimised specifically for DJI enterprise platforms. Ten teams took the Industry Application Excellence Award for solutions addressing practical operational problems.

The rewards are ecosystem access rather than large cash prizes. Winners gain exposure through DJI Enterprise channels, inclusion in DJI's Onboard AI Solutions Catalog, fast tracked access to the enterprise ecosystem, beta testing opportunities on future products and direct technical support from DJI engineers.

Counting Crops On A Colombian Banana Plantation

The clearest example of the value on offer comes from AgroCount AI and its onboard crop counting system, developed by Daniel Tovar. Tested across a 50 hectare banana plantation in Colombia, the system uses computer vision and geospatial analysis to count plants automatically.

The comparison DJI cites is the one that lands with any operations manager. The manual version of that job takes four to six workers across four days. The drone version runs on a DJI Matrice 4E carrying a Manifold 3 onboard computer, and produces the count as part of the flight.

A Nine In One Model For Traffic Management

Hangzhou New Modal Technology took a different route with a nine in one onboard fusion algorithm paired with an edge to cloud collaborative smart transportation system. A single model handles traffic enforcement, road maintenance, facility management and traffic flow monitoring.

The system runs on the Manifold 3 platform and reports into DJI FlightHub 2, which is a useful illustration of how onboard analysis and fleet management fit together. The aircraft decides what matters, and the platform is where the finding lands.

Where Else Onboard AI Showed Up

The wider group of winners gives a good picture of where this technology is heading. Entries covered air pollution tracing, coastal litter identification, bridge crack detection, search and rescue assistance and general infrastructure inspection.

Those are all jobs where a human currently reviews thousands of images after the fact. Moving the detection step onto the aircraft does not remove the human, but it changes what the human is asked to look at, which is a shortlist of flagged findings rather than the whole dataset.

Which Platforms Run Onboard AI

The compatible hardware list is short and specific. Onboard AI models in this programme run on the Matrice 4 Series, the Matrice 4D Series of docked aircraft, the Matrice 400 and DJI Dock 3, with the Manifold 3 providing the onboard computing.

For teams already operating a Matrice 4 Series aircraft, that means the route into onboard AI is a payload and software decision rather than a fleet replacement. The DJI Onboard AI Solutions Catalog is where the winning models become available to other operators.

Why Onboard Processing Matters For UK Operators

Connectivity is the practical driver. UK inspection work regularly happens in places with patchy mobile coverage, from rural rail corridors to offshore approaches and upland reservoirs, and any workflow that depends on uploading full resolution imagery mid flight will struggle there.

Data volume is the second driver. A single mapping flight can produce thousands of high resolution frames, and the cost of processing and storing all of them adds up quickly. Filtering on the aircraft cuts what needs to travel and what needs to be kept.

FAQs

What is onboard AI on a drone?

Onboard AI means the drone runs the analysis itself during flight using computing carried on the aircraft, instead of streaming imagery to a server for processing. It allows real time detection, counting and alerting without a reliable network connection. The trade off is that models have to be optimised to run within the aircraft's power and processing limits.

What is the DJI Manifold 3?

The Manifold 3 is DJI's onboard computing platform for enterprise aircraft. It gives developers a way to run custom AI models and applications directly on the drone, and it is the hardware behind most of the winning entries in the 2026 challenge.

Which DJI drones support onboard AI models?

DJI lists the Matrice 4 Series, the Matrice 4D Series docked aircraft, the Matrice 400 and DJI Dock 3 as compatible platforms for onboard AI models in this programme. The Manifold 3 supplies the computing on board.

Do drones need internet for AI processing?

Not when the model runs on the aircraft. That is the central advantage of onboard AI, because detection continues in areas with no mobile coverage. A network connection is still useful for pushing results into a platform such as DJI FlightHub 2, but it is no longer required for the analysis itself.

How do I get access to the winning AI models?

DJI adds winning entries to its Onboard AI Solutions Catalog, which is how other enterprise operators discover and adopt them. Access depends on the developer's own licensing terms alongside DJI's ecosystem programme.

Final Thoughts

The 2026 challenge is a useful signal about where enterprise drone value is moving. The aircraft and the sensors are mature, and the differentiator now is what happens to the data in the seconds after it is captured. A banana count that replaces four days of manual work, or a single model that covers four separate transport functions, both make the same point, which is that the flight is no longer the deliverable.

The DJI Matrice 4 Enterprise (M4E) is the platform behind the winning crop counting entry and is in stock now at the Coptrz official online store.