People’s Liberation Army Expands Humanoid Tests; No Armed Humanoids Serve Operational Units

China’s military institutions increased their acquisition, testing and study of humanoid robots during 2025 and 2026, but the activity has not produced a known operational armed deployment. A Reuters review of more than 100 procurement notices, studies, patents, official publications, government records and defense-company materials found growing work on perception, manipulation and training data along with substantial uncertainty about what hardware has actually been delivered.

The industrial backdrop is striking. Chinese manufacturers supplied an estimated 95% of global humanoid shipments in 2025, according to BofA Global Research. Yet this dominant share describes an immature market rather than a mature robotic workforce. Around 7,000 humanoids were sold globally that year for industrial and professional service applications, according to International Federation of Robotics figures reported by Reuters, and many went to research organizations or companies gathering artificial-intelligence training data.

Procurement points to experiments, not fielded units

A May 2025 People’s Liberation Army tender sought a humanoid robot together with installation and technical training. A later notice using a contact address associated with the National University of Defense Technology requested a system for humanoid perception and dexterous operation. The university has also presented research and competitions intended to assess the machines against military requirements.

In June 2026, the PLA budgeted about $300,000 for a system to collect and label camera, radar and motion data, including information related to terrain traversal. That is an important development step because a robot cannot reliably control its actuators from raw sensor inputs alone. It needs extensive, correctly labeled data to identify surfaces, estimate its position, plan movement and recover when reality differs from its model.

However, it could not be determined whether the cited tenders were filled, which suppliers might have won them or which models were delivered. PLA Daily has called for advanced robotic technologies to move more quickly from laboratories to military training grounds, but that appeal itself highlights the transition still under way. Laboratory performance and operational availability are separate engineering thresholds.

China’s state-owned Norinco has described its Fuxi humanoid as suitable for patrol, reconnaissance, sentry and hazardous assignments. The company also said the machine could use teleoperation, allowing a person to control it remotely. These are company-stated applications, not proof that Fuxi has entered an operational PLA unit.

The human shape buys access at a high engineering cost

A humanoid’s potential advantage is compatibility with spaces and equipment designed around people. Two arms and two legs could, in theory, combine climbing, mobility and manipulation in buildings, tunnels or ship interiors. That may justify the complex form where a machine must reach controls, handle objects or negotiate human-scale obstacles.

For simpler reconnaissance or logistics work, however, quadruped, tracked and wheeled robots would generally be cheaper and easier to deploy at scale, according to researcher Juo-Min Chou. Those configurations usually avoid the continuous balance problem imposed by walking on two legs and can devote more of their energy and payload capacity to the mission rather than simply remaining upright.

Energy is only one constraint. Unpredictable terrain increases the sensing and control burden: cameras and radar must interpret the environment, software must select stable footholds, actuators must respond quickly enough to disturbances, and the complete machine must keep working despite dust, weather, impacts and imperfect communications. A polished demonstration on a prepared surface does not establish endurance, recovery performance or reliability away from technicians.

Human control remains another unresolved boundary. China’s defense ministry said in March that artificial-intelligence-powered weapons should remain under human control. Teleoperation offers a direct control path, but it can also make performance dependent on communications and operator workload. Greater autonomy can reduce that burden while raising harder questions about supervision and accountability, especially where force is involved.

The United States is measuring the same gaps

China is not alone in exploring the technology. In September 2026, the U.S. Army named five winners of its humanoid-robot competition after receiving 103 initial submissions and observing nine finalist demonstrations. The selected technologies covered power, mobile energy generation, durable actuation, tactile sensing, command software and combinations of artificial intelligence with human control.

That component list is revealing: military organizations are still establishing technical baselines and addressing the systems beneath the humanoid shell. China’s manufacturing scale can supply machines, components and training platforms quickly, giving its defense researchers a broad experimental base. It cannot by itself solve endurance, balance, perception, field reliability or the requirement for accountable human control. Until those constraints are demonstrated outside controlled settings, China’s humanoid advantage remains principally an advantage in experimentation not evidence of an operational robotic force.

By Jonathan Barrett — Editor for AMI’s future mobility and autonomous systems section, with two decades covering robotics, e-mobility, drone-vehicle convergence, and transport mechanical systems.