Xiaomi has placed humanoid robots on the assembly line of one of its electric vehicle factories, framing the deployment as an “internship” for the machines. The trial, announced through the company’s official Weibo channel, saw the robots work autonomously for three consecutive hours at a die-casting workshop, hitting a 90.2% success rate on a self-tapping nut installation station while keeping pace with the line’s 76-second takt time. The stunt-like branding masks a serious industrial bet. Xiaomi is stress-testing whether AI-driven bipedal machines can handle real production conditions, not just lab demos.
The experiment arrives as Chinese automakers are racing to differentiate their EV factories with higher automation and in-house technology. Xiaomi’s EV program has already drawn attention from sector-focused outlets such as CNTechPost, which track how software-centric companies are entering car manufacturing. By putting humanoid robots directly onto a live station, Xiaomi is signaling that it wants to leapfrog traditional automation, not only with more robots, but with more flexible, AI-native ones that could in theory move between tasks with software updates rather than mechanical reengineering.
What the Robots Actually Did on the Line
The trial took place at a self-tapping nut feeding and installation station inside a die-casting workshop at one of Xiaomi’s EV plants. According to metrics shared on the company’s social feed, the humanoid robots operated continuously for three hours without human intervention, completing dual-side nut installations at a 90.2% success rate. That figure matters because the station demands precision: self-tapping nuts must be seated at exact torque levels to secure die-cast structural components, and any misalignment can compromise part integrity downstream. Even a small percentage of mis-seated fasteners can cascade into costly scrap or time-consuming rework later in the line.
The robots also matched the line’s fastest cycle, completing each task within a 76-second takt time. Takt time is the maximum allowable interval per unit to meet production targets, so keeping pace means the robots did not slow the station’s throughput during the trial. A 90.2% success rate, however, still implies roughly one in ten attempts required correction or manual backup. In high-volume automotive manufacturing, that gap between 90% and the near-100% reliability expected of fixed automation is significant. The “intern” label, then, is not just marketing whimsy. It signals that Xiaomi treats this as a learning phase rather than a finished deployment, with human workers still ultimately responsible for quality.
The AI Model Behind the Hardware
The robots run on Xiaomi-Robotics-0, a vision-language-action model that the company’s research team developed and described in an open technical paper. Unlike traditional industrial robots that follow rigid pre-programmed paths, a VLA model fuses camera input with language-based task understanding and translates both into physical actions in real time. The arXiv paper, authored by researchers affiliated with Xiaomi, details the training pipeline and deployment strategy, with a strong emphasis on real-time execution speed and closed-loop perception, which are essential for keeping up with a moving production line where parts may not be perfectly fixtured every time.
Open-sourcing the model is also a deliberate strategic choice. By publishing the architecture and benchmarks, Xiaomi invites external researchers to probe, critique, and potentially improve the system, while positioning itself as a platform leader rather than just a hardware vendor. The practical question is whether a general-purpose VLA model can reliably handle the narrow tolerances and strict uptime requirements of automotive assembly. Lab evaluations described in the paper show what the model was tested on, but the factory trial represents the first public evidence of Xiaomi-Robotics-0 operating under genuine production constraints, with real parts, real cycle times, and real consequences for failure when a nut is misaligned or a thread is stripped.
Why Xiaomi Chose the “Intern” Frame
Calling the robots “interns” is a calculated piece of corporate communication. Coverage from industry-focused outlets like Gasgoo’s daily brief highlighted the language as noteworthy, and for good reason. The framing sets public expectations low: interns make mistakes, need supervision, and improve over time. If the robots fail a task or require human backup, the narrative absorbs the setback. If they exceed expectations, the story writes itself as a tale of precocious “trainees.” This approach also deflects immediate concerns about job displacement by implying the machines are supplementary learners, not replacements for skilled assemblers.
The framing also reflects real technical limits. A 90.2% autonomous success rate is impressive for a humanoid robot performing fine-motor industrial work, but it is nowhere near the reliability threshold needed for unsupervised mass deployment. Xiaomi appears to be managing that gap publicly rather than overpromising. The company has suggested that the robots can perform tasks across multiple factory scenarios, yet the disclosed data covers only one station in one workshop. Scaling from a single nut-installation task to dozens of varied assembly operations (each with different tools, materials, and safety constraints) will require extensive iteration on both hardware and software. The intern metaphor buys time for that process and gives Xiaomi room to report incremental improvements without declaring victory prematurely.
Mass Deployment Plans and the Five-Year Horizon
Xiaomi has signaled ambitions well beyond a single pilot. According to coverage of the company’s roadmap, Xiaomi aims for mass deployment of humanoid robots within the next five years. That timeline would place large-scale rollout around 2031, roughly aligning with the period when many Chinese EV makers expect to operate at or near peak capacity in new plants. The economic logic is straightforward: as labor costs rise and production volumes grow, humanoid robots that can adapt to varied tasks without retooling entire lines could become attractive, especially for tasks that are ergonomically difficult or involve repetitive strain for human workers.
Still, a five-year forecast for mass deployment of humanoid robots in automotive settings deserves scrutiny. Much of the reporting on the Xiaomi trial treats the company’s own performance metrics and timeline as established fact, but the core data comes from a single social media announcement and subsequent write-ups on channels such as CnEVPost’s news feed. Independent verification of the 90.2% success rate or the three-hour autonomy window has not been made public. The gap between a controlled factory demo and a fleet of robots running 24-hour shifts across multiple plants is enormous. Xiaomi’s EV rivals have invested heavily in conventional automation, yet they have not publicly committed to humanoid platforms for line work at this scale. If Xiaomi’s approach proves viable, it could pressure those competitors to follow; if the robots plateau at demo-level reliability, the five-year target may quietly slip into a longer-term aspiration.
What a 90% Success Rate Actually Means for Workers
The most immediate question for workers is not whether robots will take over the line, but how a 90.2% success rate changes their day-to-day reality. In practice, every failed attempt by the robot becomes someone else’s problem: a technician who must clear a fault, an inspector who must flag a defect, or an operator who must redo the installation to preserve quality. Rather than erasing human labor, a partially reliable humanoid system can shift it into oversight, exception handling, and maintenance. In the short term, that may create hybrid roles where workers supervise one or more robots, intervene on edge cases, and provide feedback that engineers use to refine the AI model.
Over time, if Xiaomi succeeds in pushing reliability closer to that of fixed automation, the balance could tilt. The company’s decision to test humanoids on a precision fastening task suggests that it is targeting jobs that are both repetitive and physically demanding, where management can plausibly argue that automation improves ergonomics and safety. Yet even in that best-case framing, the same technology could reduce the number of entry-level positions on the line, especially for tasks that do not require complex judgment. For now, the “intern” robots are more dependent on human colleagues than the branding suggests—but the whole point of the internship is to learn, and Xiaomi’s five-year horizon makes clear that this is not a one-off experiment but the beginning of a long-term shift in how work is organized inside its EV factories.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
