Chinese scientists have unveiled a new artificial intelligence system that could revolutionize the way drone swarms locate and eliminate enemy targets on the battlefield. Traditional drone operations face significant challenges, as each unit in a swarm can monitor only a limited area, requiring constant communication to form a complete operational picture. This reliance becomes a vulnerability when enemy jamming disrupts drone networks, causing targets to intermittently vanish from operators’ view.
The newly developed algorithm, called Heterogeneous Graph Spatio-Temporal Reasoning (HG-STR), allows drones to make informed decisions even with incomplete information, while maintaining coordinated action across the swarm. According to a study published in Acta Aeronautica et Astronautica Sinica, the system enables a swarm to search vast areas and neutralize every target, even under communication blackout conditions.
HG-STR uses a “heterogeneous graph” approach, mapping every object in the operational area onto a web in which each connection carries specific, meaningful data. Each drone acts as a node, sharing its position, speed, ammunition status, and past tasks. Enemy targets are similarly represented as nodes, with details on location and remaining damage needed to destroy them. Environmental data, such as unsearched terrain, also forms part of this interconnected information network.
To further enhance operational efficiency, researchers developed a compressed memory system that retains each drone’s past observations. This allows drones to continue making intelligent decisions even when communications fail, rather than starting from scratch. Tests indicate that HG-STR could achieve a 100 percent target elimination rate, according to scientists at Northwestern Polytechnical University in Xian.
In a simulated exercise, a swarm of ten drones successfully searched and destroyed all targets across a 100 km × 100 km area, optimizing flight paths and reducing travel distance. The system can generate decisions in as little as 6.6 milliseconds, a speed critical for high-stakes battlefield operations. Unlike existing autonomous algorithms that treat all information equally—friend, foe, and terrain—the HG-STR framework distinguishes between different types of data, minimizing confusion and enhancing operational precision.
Researchers believe this breakthrough could pave the way for fully autonomous drone fleets capable of executing complex combat missions in environments cut off from human command, following a single directive to seek and destroy all enemy targets.