Researchers at Northwestern Polytechnical University (西北工业大学) in Xi’an published a new AI targeting algorithm called HG-STR (Heterogeneous Graph Spatio-Temporal Reasoning) in China’s leading aviation journal Acta Aeronautica et Astronautica Sinica on May 19, claiming it is the first system capable of achieving a 100% kill rate while operating fast enough for modern combat. According to SCMP, the algorithm allows fixed-wing drone swarms to autonomously locate and eliminate all targets even when communications are jammed and sensors are blocked. A Beijing-based defense analyst, speaking anonymously, described the premise as deploying a swarm into a contested environment with a single final instruction: “find and kill them all.”
The key technical advance is how HG-STR handles battlefield data. Conventional targeting AI treats all inputs — drones, buildings, terrain, enemy vehicles — as equivalent objects in a dataset. HG-STR instead constructs a dynamic relational graph where entities are classified by type and linked through relationships: a radar installation becomes a threat node tied to jamming sources and airspace; a forest becomes a concealment asset; allied drones become information-sharing nodes. This lets a swarm infer enemy positions and adapt after losing comms or direct visual confirmation, addressing a vulnerability prominently exposed by electronic warfare in Ukraine. Analysts note the 100% kill rate is a simulation figure, and real-world variables — weather, damaged sensors, deception, civilian presence — remain untested. The US, NATO nations, and Russia are all pursuing comparable autonomous swarm programs.