China’s Military Humanoid Robot Push: What the Evidence Actually Shows

China’s defense establishment is building a pathway from commercial humanoid robotics to military experimentation. Public records indicate work on procurement, training data, perception and operational concepts, including urban combat and missions behind enemy lines. But the available evidence supports testing and planning—not the fielding of armed humanoid infantry. 33

The evidence: a military research and procurement pipeline

A Reuters review of more than 100 Chinese military procurement notices, academic studies, patents, official publications, government records and defense-industry materials found that Chinese military institutions are testing humanoids against battlefield requirements and examining how they could work alongside troops. The activity reportedly accelerated in 2025 and 2026. 33

That distinction matters. A procurement notice for a robot, a technical study or a public demonstration can show institutional interest and capability development; it does not prove that the People’s Liberation Army has created deployable robot combat units.

The public record points to several connected strands:

• Military concepts and research: PLA-linked researchers have modeled uses in urban combat and behind enemy lines, while the PLA Daily called for transferring robotic “combatant” technologies from laboratories to military training grounds. 33
• Procurement and training infrastructure: reporting identifies military interest in acquiring humanoid robots and technologies for perception, dexterous operation and training data. Those are the building blocks needed to make a robot useful outside a scripted demonstration. 33
• Commercial-to-defense links: Unitree is a prominent commercial robot maker. A U.S. congressional assessment says the company has documented links to PLA-affiliated institutions and Chinese Communist Party entities, participation in military-civil fusion programs, state funding and defense-research involvement. 1
• Defense-industry activity: state-owned defense conglomerate Norinco has publicized its Fuxi humanoid for functions including sentry duty, reconnaissance and patrol, with teleoperation presented as an option. 6
• Public competitions: China’s robot competitions demonstrate improving balance, mobility and recovery from falls. They are useful technology showcases, but they should not be treated as proof of combat readiness. Unitree’s H1 won a 1,500-meter humanoid race in Beijing in 2025. 4

What China may use humanoids for first

Humanoid form has one practical appeal: it is designed to operate in infrastructure built for people. A machine with arms, hands and legs could potentially use stairs, doors, ladders, tools and vehicles without requiring every environment to be rebuilt for a wheeled or tracked platform.

That makes several support roles more credible than a robot rifleman.

Urban reconnaissance and assault support

Military research has explored humanoids operating with forces in urban combat. Likely tasks include entering buildings ahead of troops, mapping rooms, carrying sensors, opening doors, inspecting interiors and relaying video or other sensor data. 33

These jobs could reduce exposure for soldiers, but urban warfare is also where perception is hardest: smoke, damaged buildings, civilians, poor lighting, debris and electronic interference make reliable machine judgment exceptionally difficult.

Infiltration and operations behind enemy lines

Researchers have modeled missions behind enemy lines. In principle, a humanoid could place sensors, inspect facilities or move through dense built environments where humanlike mobility and manipulation are useful. 33

Public information does not establish that such concepts have progressed into operational PLA missions.

Guard, patrol and sentry duty

Sentry and patrol functions are more plausible early applications because they can be geographically constrained and operate under clearer rules. Norinco’s descriptions of Fuxi include sentry duty, all-weather reconnaissance and intelligent patrol. 6

Logistics, recovery and hazardous work

Humanoids could eventually carry supplies, retrieve equipment, inspect dangerous sites or assist with casualty recovery. A U.S. defense-focused humanoid developer has similarly identified supply handling, reconnaissance, equipment or casualty recovery, and hazard inspection as potential applications. 43

China’s wider military interest in unmanned ground systems also extends beyond humanoids. A PLA tender sought AI-powered robot dogs for collaborative scouting and explosive-hazard clearance, suggesting that more mature ground robots may be fielded sooner for many high-risk tasks. 34

The technology China is prioritizing

The hardest problem is not making a robot walk across a stage. It is making it perceive, decide and act reliably in an uncontrolled environment.

Public reporting on China’s military-relevant work points to four major priorities:

1. Perception and navigation. Robots need to interpret visual, radar and motion data, identify obstacles and navigate varied terrain. Military procurement has included systems to collect and label this type of data for humanoid training. 5
2. Dexterous manipulation. A useful humanoid must operate doors, handles, tools and other objects designed for human hands. Military interest in “intelligent perception and dexterous operation” reflects this requirement. 7
3. Locomotion and balance. Robots need to climb stairs, traverse uneven ground, carry loads and recover after a fall. Competitions can test parts of this capability, but controlled sporting events are a long way from battlefield durability. 4
4. Embodied-AI training data. Humanoid autonomy depends on huge amounts of paired data linking sensors, physical actions and outcomes. China’s procurement activity suggests an effort to build that data and evaluation pipeline. 5

Why armed battlefield deployment remains uncertain

The case for support robots does not erase the limitations of two-legged machines in war.

Power and endurance are major constraints. Dynamic walking, onboard computing and manipulation consume energy, while batteries add weight and create a resupply burden.

Reliability is another problem. Robots that work in indoor demonstrations must still contend with mud, dust, rain, heat, damaged joints, sensor obstruction, blast effects and maintenance far from a workshop.

Communications are vulnerable. Teleoperation can keep a human involved in decisions, but it relies on links that can suffer from jamming, latency or degraded video. Greater autonomy reduces dependence on the link but increases safety and accountability risks.

The cost exchange ratio may be unfavorable. A humanoid combines expensive actuators, sensors, batteries and computing hardware, yet could be disabled by mines, fragments, small drones, jamming or small-arms fire.

Most importantly, there is no reliable public evidence that an armed humanoid has been deployed with an operational PLA unit. Reuters’ review found research, acquisitions and planning for eventual use—not confirmation of robot infantry in service. 33

Autonomous targeting is the central ethical and legal issue

A robot’s humanlike shape is not the key issue. The crucial question is whether it can independently select and engage people.

In urban combat, a system would need to distinguish combatants from civilians, recognize a wounded person or someone surrendering, interpret rapidly changing behavior and apply commanders’ rules of engagement. Current public evidence does not show humanoid robots can do this reliably in real combat conditions.

International humanitarian law requires distinction, proportionality and feasible precautions in attack. These obligations cannot simply be transferred to an algorithm. Meaningful human control would require an operator or commander with sufficient information, time and authority to approve, refuse, pause or abort the use of force.

Teleoperation does not eliminate risk. A remote operator can face delay, incomplete video, disrupted communications and automation bias—the tendency to defer to a machine’s recommendation even when it is wrong.

China, the United States and the broader robotics race

China’s potential advantage is its large commercial robotics base and a system that can connect civilian firms, state-backed research and defense institutions. Unitree’s links to PLA-affiliated institutions and defense research illustrate why commercial humanoid progress can have military relevance. 1

The United States remains deeply involved in military autonomy, particularly in drones, maritime systems and AI-enabled weapons development. Reuters has reported that designs behind Unitree’s successful robot dogs drew on innovations funded by U.S. military research, underscoring how technologies can move between research ecosystems and applications. 3

Still, humanoids are not the most immediate transformation in military robotics. Cheaper, more mature systems—air and maritime drones, loitering munitions, robot dogs, automated sensors and other uncrewed vehicles—are more likely to shape near-term conflict. The United States, China and their partners are already competing to develop AI-controlled military systems across domains. 44

Bottom line

China is credibly preparing humanoid robots for possible future military use. The evidence is strongest for experimentation with urban support, reconnaissance, guard duty, logistics and hazardous operations, plus the enabling technologies of perception, manipulation, mobility and embodied-AI data. 33

The prudent reading is neither dismissal nor hype: China’s work is strategically significant, but armed autonomous humanoid soldiers remain publicly unconfirmed and technically immature. If humanoids enter operational use first, they are more likely to appear in constrained support and surveillance roles than as independently target-selecting battlefield fighters.