Capable of deciding autonomously, without human decision or intervention, in a few milliseconds and acting despite jamming, Chinese drones piloted by AI could transform the battlefield.
Researchers at Northwestern Polytechnic University in Xi’an have developed an AI algorithm, HG‒STR, capable of allowing swarms of drones to track and neutralize targets almost completely autonomously, even under jamming and degraded visibility, according to a study published on May 19, 2026 in the journal Acta Aeronautica and Astronautica Sinica and relayed by the South China Morning Post. The system, which bases its reasoning on a “heterogeneous graph” of the battlefield, claims a decision time of 6.6 milliseconds and a “destruction rate” of 100% in simulation on fixed-wing drone scenarios.
Where traditional systems process data in a relatively uniform manner, HG‒STR finely distinguishes allies, enemies and terrain elements, each entity becoming a specialized node within a dynamic network. This structuring allows drones to prioritize in real time threats, search areas and cooperation opportunities, without permanently depending on a human operator. Each device also carries a compressed memory of its position, its speed, its ammunition stock and the mission history, which allows coherent reasoning of the swarm even in the event of a communication breakdown.
Also read:
War in Ukraine: Ukrainian drones strike Saint Petersburg, 1,100 km from Kyiv, on the first day of the International Economic Forum, the “Russian Davos”, where Putin is expected
“Heterogeneous map” of the battlefield
The central contribution of HG‒STR lies in this “heterogeneous map” of the battlefield, where each object is encoded according to its operational significance and its tactical priority. When an enemy target is identified, it is immediately propagated as a top threat in the graph, while a friendly drone is treated as a coordination or support opportunity. Several analyzes underline that such a model would open the way to swarms capable of operating in high-risk areas and under intense jamming, with the only initial instruction being a fully automated attack mission.
Also read:
A revolution in the face of intrusions into NATO airspace? The A1-Falke, this interceptor drone designed to capture other drones in nets, already adopted by Germany
This breakthrough is part of Beijing’s broader strategy to massively integrate autonomous systems and robotics at the heart of its military apparatus. According to the World Robotics 2025 report from the International Federation of Robotics (IFR), China had approximately 2.027 million industrial robots operational in 2024, the largest fleet in the world and a major industrial base for this race for autonomous weapons systems.
Also read:
War in Ukraine: to meet its operational needs, the Ukrainian army is testing “legged” drones capable of setting up ambushes
Despite the scale of the promises displayed, many experts are calling for these performances, obtained for the moment only in simulation, to be put into perspective. Virtual scenarios do not fully replicate the complex jamming, weather conditions, friction of real combat or the adaptability of a human adversary, limiting the reach of the advertised “100%”.
