A Reuters review shows China accelerated military work in 2025-26 to turn civilian humanoid robots toward combat roles, while technical and operational limits have so far kept armed humanoids out of PLA units.
Robots danced and raced at the World Humanoid Robot Games; two days later the PLA Daily urged researchers to speed lab advances into military training for robotic "combatants".
A Reuters review of more than 100 Chinese procurement notices, patents, academic papers and defense-industry records shows Beijing stepped up work on humanoids in 2025 and 2026, concentrating on perception, manipulation and the massive training datasets those systems need. Bank of America Global Research estimates Chinese firms supplied about 95% of global humanoid shipments in 2025, giving the People’s Liberation Army ready access to a large domestic industry. A 2025 National University of Defense Technology study sketched a six-robot urban assault and projected availability in five to 10 years, but Reuters found no evidence an armed humanoid serves with a PLA unit and experts point to energy demands and reliability limits.
1. What Reuters review found
Records examined
The assembled documents include procurement notices, patents, academic studies, government filings, and defense company materials. They repeatedly focus on perception and manipulation. They also highlight the data pipelines and test rigs needed to generate the large training datasets humanoids require for unstructured environments. Several papers and technical solicitations examine how robots identify objects, grasp and carry loads, and coordinate movements with human personnel.
The mix of records is practical, not just promotional. Contract notices sought hardware and software for training, while academic work modeled robot interactions with troops and urban terrain. Patents and company materials described iterative improvements in control software and sensors rather than one-off showcase tricks.
Momentum in 2025-2026
Work on those topics accelerated sharply in 2025 and into 2026, with a cluster of new studies, procurement requests and industry releases across that period. The spike included calls inside the military establishment to shorten the lag between lab development and training-ground trials. Two days after a high-profile humanoid competition, an official PLA publication urged researchers to press technologies toward field testing for robotic "combatants."
Taken together, the files map a program moving from spectacle toward engineering requirements and military rehearsal.
2. PLA call to speed lab-to-field transfer
PLA Daily message
PLA Daily turned a spectacle into a directive. Its piece argued that robotics advances should not stop at demonstrations and competitions but must move quickly into military training and testing, and it used language that placed the machines in combat roles.
The paper urged a speeded transfer of cutting-edge innovations from laboratories to military training grounds for robotic "combatants." That phrase, in quotes, appeared as an operational objective rather than a rhetorical flourish.
The editorial asked researchers and units to shorten development cycles and to prioritize technologies that can be tested in realistic training scenarios. It stressed coordinated trials that pair new hardware, sensor suites and control software with troop units on mock urban terrain, signaling a shift from showcase events to rehearsal for use alongside soldiers.
Institutionally, the public call gives formal cover to procurement offices and research institutes seeking funds and approvals for joint lab, unit experiments. It also raises the bar for industry partners, who now face explicit demand from a military outlet to produce systems ready for field trials, not just stage performances.
Watch the PLA's training calendars and procurement notices for the first concrete follow-through: joint trials and equipment requests that turn the editorial’s imperative into scheduled exercises.
3. China's industrial edge and supply dominance
Chinese commercial robotics firms have quietly shifted from spectacle to systems. Engineers who once built dancing and racing humanoids are now shipping sensors, control stacks and software suites that map directly onto military testing needs. That move shows up in company materials and patents that describe iterative improvements in control software and sensor packages rather than one-off stage tricks.
Commercial manufacturers and access
• Sensors and control software. Manufacturers are pushing higher-fidelity vision and proprioception modules, and packaging those with control software that handles balance, grasping and bipedal locomotion. Technical papers and patent filings emphasize steady upgrades to those subsystems.
• Training environments and data pipelines. Firms now offer simulation platforms, test rigs and data-generation tools designed to create the large labelled datasets researchers need to teach robots to find, pick up and move objects in cluttered, urban interiors.
• Production scale and productization. Company documents show a shift toward repeatable hardware lines and software releases, moving capabilities out of labs and into packaged products that units and institutes can buy and test.
That commercial trend matters because it shortens the gap between a lab prototype and a field trial. The industry isn't selling novelty acts anymore. It's selling components, datasets and test systems that match the technical priorities military researchers are asking for, and those offerings are already being folded into military research plans.
4. Technical priorities: perception, manipulation and training data
Chinese labs and military institutes have narrowed the engineering agenda to three hard problems that will decide whether humanoids move from spectacle to service. The list is simple: reliable sensing, repeatable manipulation, and the massive labelled data pipelines those systems require.
At the front is robot perception, a stack that must spot threats, obstacles and friendly forces in cluttered, smoke filled interiors while the platform stumbles and recovers. Researchers are racing to improve sensor suites and software that fuse multiple inputs, cut false positives and keep situational awareness when cameras blur or occlusion hides a doorway. robot perception sits behind nearly every other capability.
Manipulation is the second bottleneck. Engineers are working on control loops and gripper designs that can repeat basic physical tasks under unstable balance and unpredictable contact forces. That work is mechanical, electrical and software heavy, and it needs to be proven in rigs that mimic stairwells, narrow doorways and crowded rooms.
The third priority is scale: creating, curating and labelling the data these systems need. Teams are building testbeds, simulation runs and controlled exercises to generate the edge cases that wreck demo robots. The effort ties industry simulation tools to military training grounds so algorithms encounter realistic noise, human movement and furniture layouts before a real trial.
Get past these three, and demonstrations become rehearsals. Watch for announced joint trials that pair lab stacks with unit exercises.
5. Urban warfare concept: the NUDT six-robot assault
How robots would operate with infantry
A 2025 National University of Defense Technology scenario laid out a coordinating package of six "combat robots", a mix of humanoids, robot dogs and unmanned platforms, tasked with clearing buildings floor by floor. The machines are cast as point elements that move before soldiers, probe rooms and map interiors, and perform the highest-risk physical tasks so troops can follow and secure space. That choreography treats robots as scouts, breachers and payload carriers working in concert with human teams rather than as stand-alone fighters. The study framed those roles around data links and shared situational feeds so commanders and nearby units can make rapid decisions from robot sensor inputs. The concept leans on the idea of sending machines into dangerous places to reduce human casualties, while keeping final control and decisive clearing to people.
Estimated availability window
The paper presented its notional timetable alongside a list of technical milestones, tying any fielding to advances in sensing, grasping and the large training datasets those systems require. Availability therefore depends on both engineering breakthroughs and a steady production pipeline. Watch for two concrete signals that would move the clock: scheduled joint trials that pair lab stacks with troop exercises, and procurement notices that buy integrated robot kits for unit testing. Those steps would mark the shift from demonstration to rehearsal as China pushes military research accelerates.
6. Operational limits, deployment status and expert views
Energy use and reliability
Major technical limits have so far kept humanoid robots out of frontline PLA units. Reuters found no evidence that China has deployed an armed humanoid with an operational PLA formation, and reporting notes current models remain energy intensive and unreliable outside controlled demonstrations. Demonstration robots frequently require high power for short bursts of motion and struggle to sustain balance and sensing in cluttered, unstructured interiors, making long patrols or sustained combat tasks impractical.
Testing environments mask failure modes. In competitions and staged trials robots fall, misidentify objects and lose tracking when cameras blur or smoke appears. Those same failure patterns appear in military solicitations and academic papers, which repeatedly flag power consumption, battery life and mechanical wear as gating problems for field use.
Expert commentaries
Researchers and commentators quoted in reporting say the main appeal is clear: send machines into hazardous spaces to reduce human casualties. At the same time, other experts warn that the chaotic variables of a real battlefield, moving civilians, debris, variable lighting, electronic noise, present hard obstacles for perception and manipulation systems.
So deployment will depend on more than industry scale. Joint trials, sustained unit testing and reliable logistics for power and repairs must come first. The single concrete benchmark to watch remains the National University of Defense Technology study from 2025, which projected such equipment could be available within five to 10 years.
Originally reported by Reuters.
