Chinese robotics company UBTECH has deployed dozens of its Walker S1 humanoid robots inside an electric vehicle factory operated by ZEEKR, testing whether teams of bipedal machines can handle real production tasks like sorting parts, moving materials, and assembling components. The trial, branded “Practical Training 2.0,” represents what UBTECH calls the world’s first multi-humanoid collaborative training across multiple tasks and scenarios in a live manufacturing environment. The effort signals how aggressively Chinese firms are pushing to integrate humanoid robots into factory floors, though independent verification of the program’s scale and results has not yet surfaced.
What is verified so far
The core facts trace back to a single company-issued release. UBTECH says it deployed dozens of its Walker S1 humanoid robots across production areas at ZEEKR’s 5G Intelligent Factory, where the machines performed collaborative sorting, handling, and precision assembly. The company describes the deployment as the world’s first instance of multiple humanoid robots training together in multi-task, multi-scenario factory conditions.
The robots are coordinated through a system UBTECH developed called BrainNet, which connects the humanoids via what the company terms the “Internet of Humanoids,” or IoH. That architecture relies on cloud-device inference nodes, meaning processing is split between onboard computing in each robot and centralized cloud servers. The design aims to enable what UBTECH frames as swarm intelligence, where individual machines share learned behaviors and adapt collectively rather than operating as isolated units.
ZEEKR, the factory partner, is a premium electric vehicle brand under Geely Auto Group. Its 5G Intelligent Factory serves as the testing ground, giving the Walker S1 units access to actual automotive production workflows rather than simulated lab environments. The choice of an EV factory is notable because automotive assembly lines already rely heavily on traditional industrial robots, meaning the humanoids must prove they can add value in spaces where automation is already mature.
The deployment is structured as a training exercise rather than a permanent workforce replacement. UBTECH’s framing as “Practical Training 2.0” suggests an iterative approach, building on earlier trials of the Walker platform. The “intern” label, used widely in coverage of the event, reflects the company’s own positioning of these robots as learners rather than finished products ready for full-scale commercial deployment. That language signals that the robots are still in a developmental phase, with the factory acting as a high-stakes classroom.
All of these details come from UBTECH itself. The company distributed its announcement via a PR Newswire channel, which is designed for organizations to publish official statements directly to media and industry audiences. Within that controlled format, UBTECH sets the narrative: the robots are framed as a coordinated “swarm,” the factory as a cutting-edge testbed, and the project as a milestone for Chinese humanoid robotics.
What remains uncertain
Several significant gaps exist between what UBTECH has announced and what outside observers can confirm. The most obvious is the exact number of robots deployed. The company’s release specifies “dozens” of Walker S1 units, but broader reporting has referenced figures above 100. No independent source has verified a count exceeding what UBTECH itself stated. Readers should treat any number beyond “dozens” as unconfirmed and recognize that the difference between 24 robots and 120 robots is substantial when evaluating scale.
Performance data is entirely absent from the public record. UBTECH has not disclosed error rates, task completion speeds, downtime, or any comparison between the humanoids’ output and that of human workers or conventional industrial robots. Without those metrics, it is impossible to assess whether the Walker S1 units actually improved factory efficiency or simply demonstrated basic competence in controlled conditions. The distinction matters: a robot that can sort parts in a short demo is not the same as one that can do so reliably across an eight-hour shift at production speed, with real-world variability in parts, lighting, and workflow interruptions.
There is also no independent confirmation from ZEEKR about the trial’s scope or outcomes. The factory operator has not issued its own public statement, leaving the narrative entirely in UBTECH’s hands. This single-source dynamic is common in early-stage technology announcements but limits the ability to evaluate claims about real-world impact. Without corroboration from the automaker, outside observers cannot verify whether the robots were integrated into critical production steps or confined to low-risk, easily controlled corners of the line.
The BrainNet system and the IoH concept lack published technical documentation beyond UBTECH’s promotional materials. No peer-reviewed research, third-party audit, or independent benchmark has been made available. The cloud-device inference architecture sounds promising in theory, but without specifics on latency, redundancy, failure modes, or security protocols, it is difficult to judge whether the system can scale beyond a controlled trial. Cloud-reliant control can introduce vulnerabilities: connectivity outages, cyberattacks, or overloaded servers could all affect robot behavior, yet none of these risks are addressed in the available materials.
Worker impact is another blind spot. The announcement does not address whether any human jobs were displaced, reduced, or restructured during the trial. It also does not mention retraining programs, union consultations, or labor policy considerations. Given that humanoid robots are explicitly designed to operate in spaces built for human workers, the absence of any discussion about workforce effects is a meaningful omission. It leaves open key questions: Were human workers reassigned to supervise and maintain the robots? Did the presence of humanoids change safety protocols on the shop floor? Were there concerns or pushback from staff?
Safety performance is similarly opaque. UBTECH does not disclose whether any incidents, near-misses, or unexpected shutdowns occurred during the trial. In a factory filled with heavy equipment, moving vehicles, and human workers, the introduction of mobile humanoid robots raises nontrivial safety questions. Without independent reporting or regulatory filings, the public has no way to assess how rigorously those concerns were tested or mitigated.
How to read the evidence
All verified claims in this story originate from a single primary source: UBTECH’s own press release distributed through a corporate newswire. That makes the evidence base unusually narrow for a story generating this level of attention. Company-issued releases are useful for establishing what a firm says it did, but they are not substitutes for independent reporting, and they carry an inherent promotional bias. They tend to emphasize firsts, breakthroughs, and strategic importance while downplaying limitations, costs, and failures.
The “world’s first” claim attached to the multi-humanoid collaborative training deserves particular scrutiny. UBTECH uses this language in its own release, but no external body has verified the claim. Other companies, including Tesla with its Optimus program and Figure AI with its partnership at BMW’s Spartanburg plant, have also tested humanoid robots in factory settings. Whether UBTECH’s trial is genuinely the first to involve multiple humanoids working collaboratively across different tasks and scenarios depends on how narrowly those terms are defined, and the company is the one defining them. Phrases like “multi-task” and “multi-scenario” can be stretched to carve out a unique niche even when similar work is underway elsewhere.
The broader context of China’s push into humanoid robotics is real and well-documented. Chinese government policy has identified humanoid robots as a strategic technology, and multiple Chinese firms are racing to deploy prototypes in manufacturing, logistics, and service roles. UBTECH’s trial fits within that pattern, showcasing a domestic manufacturer partnering with a high-profile electric vehicle brand to demonstrate technological ambition. But fitting a pattern is not the same as proving a breakthrough, and the gap between a controlled factory trial and commercially viable humanoid labor remains vast across the entire industry, not just for UBTECH.
One analytical thread worth pulling: the choice to frame these robots as “interns” rather than workers is strategically smart but also revealing. It lowers expectations, positions failures as learning opportunities, and avoids direct comparisons with human productivity. If the robots were genuinely performing at production-grade levels, the framing would more likely emphasize reliability, throughput, and cost savings. Instead, the language of training and practice signals that UBTECH is still collecting data, refining control algorithms, and testing hardware durability in the messy reality of an operating factory.
For readers, the most grounded way to interpret this trial is as a high-profile prototype deployment rather than a revolution in factory labor. UBTECH has demonstrated that dozens of humanoid robots can be coordinated in a real manufacturing environment to perform some useful tasks, at least under conditions the company controls and curates. That is a nontrivial engineering achievement. But until independent observers can verify performance, safety, economic impact, and worker outcomes, the story remains one of potential more than proof.
Humanoid robots like Walker S1 are designed to step into spaces built for people, using legs, arms, and hands to interact with existing tools and infrastructure. That vision continues to drive investment and hype across the robotics sector. UBTECH’s ZEEKR trial shows how far that vision has come in China and, just as importantly, how many questions still need answers before humanoids can be treated as standard equipment on the factory floor rather than temporary “interns” on a closely watched assignment.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
