Xiaomi has moved its humanoid robots from the lab to the production floor of its electric vehicle factory, where the machines completed a three-hour autonomous run at a die-casting workstation. The test achieved a 90.2% success rate for dual-side simultaneous nut installation and met the line’s fastest cycle time of 76 seconds, according to reports from Sina Finance and CnEVPost. The trial is an early example of a humanoid robot being tested for sustained, real-world assembly work on an active automotive production line, and it highlights how Chinese tech firms are pushing to merge AI-driven robotics with EV manufacturing.
Three Hours on the Line
The robot operated continuously for three hours at a self-tapping nut installation workstation inside the die-casting workshop of Xiaomi’s EV factory. Self-tapping nut installation is a repetitive but precision-dependent task: the fastener must be driven at a consistent angle and torque into cast metal components, and errors can compromise structural integrity. The fact that Xiaomi chose this station for its first factory trial suggests the company is targeting tasks where consistency matters more than dexterity, a pragmatic starting point for a machine still proving itself outside controlled environments.
The 90.2% dual-side simultaneous installation success rate is a meaningful but imperfect number. In a typical automotive plant, human operators or dedicated robotic arms aim for near-perfect accuracy on fastening tasks. A roughly one-in-ten failure rate would not yet be acceptable for full-speed production without human oversight. But the metric carries weight because the robot also met the line’s fastest production cycle time requirement of 76 seconds, meaning it kept pace with the factory’s actual throughput demands rather than running at a slower demonstration speed designed to flatter the technology.
In materials shared alongside the announcement, Xiaomi also showed the robots attempting tasks beyond nut installation, including screwing and material handling. That range hints at a broader ambition: rather than building single-purpose machines, Xiaomi appears to be developing general-purpose humanoid platforms capable of rotating between workstations much like a human worker. If the same unit can be reassigned from fastening to basic logistics or inspection, the return on investment for each robot looks very different from that of a fixed, single-function arm.
While Xiaomi has not disclosed detailed failure modes, the current performance implies a hybrid operating model in the near term: humanoids handling the bulk of repetitive work, with human technicians stepping in to correct misinstalls, clear alerts, or take over when edge cases arise. That kind of human-in-the-loop setup is common in early automation deployments and provides a path to gradually tightening tolerance thresholds as software and hardware improve.
The AI Model Behind the Robot
The technical backbone of this effort is a vision-language-action model called Xiaomi-Robotics-0, described in a technical report on arXiv (ID 2602.12684). The paper, titled “Xiaomi-Robotics-0: An Open-Sourced Vision-Language-Action Model with Real-Time Execution,” outlines a system designed to process visual input, interpret task instructions, and execute physical actions in real time. The model was evaluated on simulation benchmarks and real-robot bimanual manipulation tasks, meaning it was tested on coordinated two-handed operations that mirror the kind of work required on an assembly line.
Xiaomi open-sourced the code and model checkpoints, a decision that separates this project from the proprietary approach taken by most industrial robotics companies. Open-sourcing an AI model of this type invites external scrutiny and collaboration, but it also accelerates adoption. If third-party developers and academic labs can build on Xiaomi-Robotics-0, the model could improve faster than it would inside a single company’s R&D pipeline. For the broader robotics field, this is a notable data point: a major consumer electronics and EV manufacturer is treating its core robotics AI as a shared resource rather than a closely guarded trade secret.
The real-time execution capability is what makes factory deployment feasible. Industrial environments demand low-latency response times; a robot that pauses to process visual data or recalculate a grasp trajectory can disrupt production flow. The arXiv paper’s emphasis on low-latency perception and control suggests Xiaomi engineered the model specifically with factory constraints in mind, rather than retrofitting a general AI system onto a physical platform after the fact.
Lei Jun Calls Them “Interns”
Xiaomi CEO Lei Jun framed the deployment in deliberately modest terms, describing the robots as “interns” at the automotive factory in a social media post. The language is strategic. By setting expectations low, Lei Jun gives the company room to iterate without being held to the standard of a finished product. It also softens the narrative around automation displacing workers, a sensitive topic in China’s manufacturing sector where millions of jobs depend on assembly-line work.
The “intern” framing, though, should not obscure the scale of Xiaomi’s ambition. Reporting that carried Lei Jun’s comments noted that the company plans mass deployment of humanoid robots across its manufacturing operations. No specific timeline or unit count has been disclosed through official filings or detailed roadmaps, so the scope of that plan remains vague. Without concrete production targets or capital expenditure figures tied to robotics, the mass deployment claim sits closer to aspiration than commitment, but the early factory trial shows that the concept has progressed beyond slideware.
Intern branding also serves another purpose: it signals to employees and regulators that humans remain central to the process, at least for now. By positioning robots as trainees learning from experienced staff, Xiaomi can experiment with new workflows while maintaining a narrative that emphasizes augmentation rather than replacement.
Why a Humanoid Shape Matters
Most automotive factories already use robots extensively, but they are typically fixed robotic arms bolted to the floor, each programmed for a single repetitive motion. A humanoid form factor offers a different value proposition: it can navigate spaces designed for human workers, use tools built for human hands, and switch between tasks without requiring the factory to be redesigned around the machine. That flexibility comes at a cost in precision and speed compared to purpose-built industrial arms, which is why the 90.2% success rate, while promising, still trails what a dedicated fastening robot would deliver.
The trade-off makes sense only if Xiaomi can push accuracy higher while maintaining the robot’s ability to generalize across tasks. A humanoid that matches a fixed arm’s precision on one task but can also walk to the next station and perform a different operation would fundamentally change the economics of factory automation. Instead of buying dozens of specialized machines, a manufacturer could deploy a smaller fleet of adaptable robots and reassign them as production needs shift. That is the bet Xiaomi is making, and the early factory test results reported by Sina Finance and CnEVPost represent an initial data point on whether it can pay off.
Humanoid robots also offer a hedge against uncertainty in product design. EV architectures, battery pack layouts, and manufacturing techniques are evolving quickly. A plant filled with hard-to-reconfigure machinery can become obsolete if a company changes its vehicle platform. In theory, humanoids can be retrained for new assembly sequences without major capital upgrades, turning software updates into the primary lever for adapting to new models.
China’s EV-Robotics Feedback Loop
Xiaomi is not pursuing this strategy in isolation. China’s broader EV ecosystem, chronicled by outlets such as CnEVPost’s news feed, has become a proving ground for aggressive manufacturing innovation. Intense competition on price and features pushes automakers and tech firms to squeeze more efficiency out of their factories, and humanoid robots are emerging as one of the more visible tools in that race.
For now, Xiaomi’s three-hour trial is best understood as an advanced pilot rather than a production milestone. The robots still fall short of human or traditional industrial robot reliability on critical fastening tasks, and Xiaomi has not indicated when, or if, it plans to run them for full shifts or across multiple stations without human backup. But by putting humanoids on a live line and tying their performance to real cycle times, the company has moved the conversation from glossy demos to measurable factory metrics.
If subsequent trials show steady gains in accuracy and robustness, Xiaomi’s approach could influence how other EV makers think about automation, especially in markets where labor costs are rising and product cycles are shortening. If the performance plateaus, the experiment will still have generated valuable open-source tools and datasets for the wider robotics community. Either way, the die-casting workstation in Xiaomi’s EV factory has become an early test case for whether general-purpose humanoid robots can earn a permanent place on the modern automotive line.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
