China's leading drone-swarm developer counts air-ground-sea coordination among its gaps

The document reads like a vendor's own capability matrix with the gaps left in. The institute grades the program in five areas: survivability, situational awareness, planning and decision-making, mission execution, and autonomous learning and evolution ^[4]. Each strength arrives with a limit named beside it. Under survivability, keeping a formation resilient once different types of uncrewed systems work together across multiple domains is still a challenge ^[6]. Under situational awareness, data from several drones is fused to screen targets, identify camouflage and fix target locations, while small concealed objects with weak signatures become much harder to find in heavy background clutter and intense electronic interference ^[7][8].

The same shape holds for the software. CETC says its swarms already change formation, plan routes, avoid collisions and reallocate tasks on their own, and that inventing new tactics in unfamiliar environments against a capable adversary is still under development ^[9]. On learning, the institute says swarm AI has moved past fixed programming into adaptive optimisation and experience accumulation through reinforcement learning and adversarial iteration, with fully autonomous real-time evolution listed as a future goal ^[13].

Count the gaps and the picture gets more specific. Two of the five sit on domain mixing: survivability across different platform types, and mission execution, where air-ground-sea coordination and joint-service adaptability are the named shortfalls ^[1]. The other three would constrain a formation of a thousand identical aircraft just as much: weak-signature detection in clutter, tactical novelty against a capable opponent, and real-time autonomous evolution.

What makes mixing hard is written out. Different platforms have different communications requirements, sensors, movement characteristics and operating environments, and the systems also have to work with military units from different services ^[11]. The first four are engineering; the fifth is a command arrangement, decided by whichever service owns the mission.

For the next demonstration, two questions sort most of the claims. Was the behaviour performed by a single platform type or by several, and was it performed with the communications link intact or degraded. Same-type and degraded is where this document puts what China has; mixed-type across domains is where it puts what it wants ^[14]. The institute's own summary is that several areas are mature and that significant challenges remain ^[5].