The Robot That Can't Walk Straight Yet Still Worries Defence Analysts

The Advantage Isn't the Robot That Works

• It's the Factory That Can Build It Here's what's actually happening: China doesn't have a proven combat-ready humanoid robot. Reuters found no evidence one has ever been deployed with an operational PLA unit and researchers on both sides agree the machines remain too energy-hungry and unreliable outside controlled demonstrations to fight in real battlefield conditions. So why is this story worth taking seriously as a defence development rather than dismissing it as sci-fi hype? Because the actual capability China has already locked in isn't a working combat robot - it's commercial manufacturing dominance, at roughly 95% of global humanoid shipments, that can be redirected toward military applications the moment the underlying technology matures.

Civil-Military Fusion, Explained Without the Jargon

"Civil-military fusion" sounds abstract until you see the mechanism plainly: China's massive civilian humanoid manufacturing base - the same factories and supply chains producing robots for entertainment, logistics and consumer markets - gives the PLA a scaled, cost-efficient production pathway that a country without that commercial base simply doesn't have, even if that country's military research is equally advanced.

The U.S. Army has its own humanoid competition running, with real money behind it, but its domestic robotics industry trails China specifically in two- and four-legged machines - meaning even a well-funded American military program has to build the industrial base China already possesses commercially.

That gap in commercial manufacturing capacity, not any gap in current battlefield readiness, is the actual strategic asymmetry this story is documenting.

What the Procurement Trail Actually Shows

The specific evidence Reuters compiled matters more than the PLA Daily's rhetoric about robots evolving "from supporting combat toward primary combat." A May 2025 PLA tender for a humanoid robot. A June 2026 budget line of roughly $300,000 specifically for collecting and labelling camera, radar and motion data - the unglamorous training-data infrastructure that any machine-learning system, robotic or otherwise, actually needs before it can function reliably.

An August 2025 NUDT paper modelling urban assault with six combat robots on a five-to-ten-year timeline, not an imminent one. None of this describes a military that's about to deploy killer robots. It describes a military systematically building the perception, training-data and procurement infrastructure that would let it capitalise quickly once the underlying humanoid technology - largely being matured commercially, outside military labs - actually becomes battlefield-reliable.

For the exam, the transferable insight is that a country's military technology edge in an emerging capability area is often best measured not by its current demonstrated battlefield readiness, but by whether it controls the civilian industrial base that capability will eventually be built on - because that base determines who can scale fastest once the technology matures, regardless of who's ahead in lab research today.