Chinese scientists develop “kill-them-all” algorithm for drone warfare

A research group based in northwestern China has unveiled a new algorithm that could significantly reshape how coordinated drone swarms locate and strike targets in combat environments, South China Morning Post reports.

The system, called HG-STR (Heterogeneous Graph Spatio-Temporal Reasoning), is designed to enable a formation of fixed-wing unmanned aircraft to independently scan large operational zones and neutralise every identified adversary, even in conditions where communications are disrupted and sensor visibility is degraded.

According to a peer-reviewed study published on May 19 in China’s leading aerospace journal Acta Aeronautica et Astronautica Sinica, it is reportedly the first algorithm capable of sustaining a theoretical 100 per cent mission success rate while still operating at speeds compatible with contemporary battlefield dynamics.

At present, most drone missions rely heavily on remote human operators, a Beijing-based defence specialist not involved in the research noted.

“This technology suggests a future where swarms of drones could be sent into a high-risk, jammed environment, cut off from human command with a single final order: find and kill them all,” he said, requesting anonymity because he is not authorised to speak to media.

Conventional systems typically process battlefield information without distinguishing between categories such as allies, enemies, or environmental elements.

The researchers argue that this leads to analytical overlap and operational uncertainty. The study team, headed by Zhang Dong, an associate professor at the School of Astronautics at Northwestern Polytechnical University in Xian, addresses this issue.

Their approach introduces a “heterogeneous graph”, effectively a structured intelligence network in which every element is assigned a specific role or identity. For instance, a friendly drone is treated as one type of node, a surveillance zone as another, and an enemy unit as a separate category entirely.

The system is designed to dynamically prioritise relationships within this network. When a drone detects an adversary, that signal is elevated as a critical threat input. When allied units are nearby, the same framework interprets the interaction as cooperative coordination.

This architecture allows the swarm to rapidly determine both cooperative links and hostile objectives, enabling it to distinguish between allies and targets in real time, according to Zhang and his team.

By Jeyhun Aghazada