Educational Drone with Onboard AI Trained to Detect Unregistered Vessels

In Nizhny Novgorod, the educational drone Geoscan Pioneer Mini 2 was used for the first time over a water surface in optical flow mode. Neural networks were trained to find mock-ups of small vessels, distinguish objects with and without registration markings, and perform photo capture. The drone had to autonomously fly around a model water area and find vessels on it. The neural network determined the color of each vessel, the presence of registration markings, and checked whether the vessel was in a permitted zone. Thus, the system could identify unregistered vessels and those that went beyond set boundaries. The drone photographed found objects, counted and recorded their total number, then returned to the start and landed on its own. Working over water complicates optical navigation: glare, reflections, and lack of contrast landmarks hinder the quadcopter in determining its movement. Participants had to select a trajectory and flight parameters that maintained stability and provided suitable images for recognition. Experts evaluated detection and classification accuracy, photo capture, absence of false positives, return to start, and task completion time. Daniil Zolotnik, Deputy General Director for Education at Geoscan Moscow, said the country needs engineers who can not only use ready-made generative services but also make physical AI algorithms work on real equipment. At this competition, participants master such skills on an accessible task: starting with block programming, moving to text-based, and eventually writing their own neural network algorithm for drone control. Geoscan Pioneer Mini 2 is the first serial Russian educational quadcopter with onboard artificial intelligence. It is equipped with a 2K camera on a rotating servo drive and an 8-core processor with a neural accelerator. Processing data directly onboard allows the copter to recognize objects, perform autonomous flights, and operate in a swarm without a server connection. These capabilities enable Pioneer Mini 2 to be used in engineering shifts, school and university projects, where students not only learn drone control but also develop their own AI algorithms for applied tasks. The quadcopter was developed by Geoscan in 2025.