What happened
Electrek reports that Xiaomi Tech posted video on X on November 3, 2026, showing humanoid robots sorting center-console side covers at a workstation in the Xiaomi Auto factory. The robots pull covers from bins, carry them to storage racks, and place them into slots. The video includes a cumulative work clock and the label “Autonomous, 100x Speed.” Electrek’s account says the company described the run as covering the full day’s production needs at that workstation.
Xiaomi says the robots detected and corrected anomalies on their own and coordinated with the production line’s control system. It characterized the run as another step toward real-world industrial deployment. The supplied material does not provide a unit count, success rate, or number of human interventions for this run, and it does not say whether the robots had battery swaps, charging breaks, or another power arrangement during the shift. Those omissions make it difficult to translate the duration claim into a comparable measure of useful output.
The video is described as part of a public program that has added tasks over time. In March, Xiaomi put a humanoid robot into a Beijing EV factory as an “intern” installing self-tapping nuts. Xiaomi reported a 90.2% success rate over three hours and said the robot was fast enough to meet a 76-second station cycle time. By July, Xiaomi said that success rate had reached 98% and added two harder stations, including the side-cover sorting work shown in the new video.
The footage also shows mobile robots carrying bins and a wheeled robot assisting two-legged robots after a dropped part. Electrek notes that Xiaomi CEO Lei Jun said last December the company planned to deploy humanoid robots in large numbers across its factories within five years. Separately, Xiaomi said its open-source Xiaomi-Robotics-1 model, pretrained on 100,000 hours of real-world robot data, had begun trial runs in staff apartments; a 30-minute run completed more than 50 task chains across three rooms. The supplied source gives no pricing, customer access, or deployment terms for the factory robots.
What the archive adds
The archive has covered a different factory-robotics deployment claim: Tesla’s commercial-grade Optimus production announcement. This Xiaomi video is a new event, not a restatement: it offers a specific production-line task and an eight-hour duration claim, but no corresponding output or intervention metrics.
Trend signal
This case weakly supports agent economics moves to completed work: the meaningful question is not how long a robot appears to run, but how much verified work it completes at an acceptable rate. The demonstration gives a duration and task description, while withholding the count and error or intervention measures needed to assess completed work economically.
Why now
The video appeared as Xiaomi was receiving awards at the World Internet Conference in Wuzhen. It follows Xiaomi’s earlier reported factory tests, which had shared success-rate figures for nut installation, so the absence of comparable numbers in the new full-shift claim is especially notable.
Connections to other events
An earlier Telegram post in the archive described Google DeepMind’s Gemini Robotics 2 family, including a base model, Embodied Reasoning 2, and On-Device 2. That was a model-family announcement; Xiaomi’s video instead shows a claimed, task-specific factory application. Together they reflect distinct routes toward embodied systems, but the supplied material does not establish that the Xiaomi robots use Gemini Robotics.
Why it matters
For teams evaluating physical automation, an eight-hour demonstration is useful evidence of sustained activity, but it is not equivalent to a production benchmark. A buyer needs to know output per hour, quality, recovery from faults, human rescue frequency, and uptime—including how charging affects the work cycle. Xiaomi’s previous tests show that the company has published some success and cycle-time figures before; comparable data for this run would let operators judge whether the task beats or complements existing automation.
The footage also highlights a design choice: the work is performed alongside wheeled mobile platforms, and the video includes a wheeled robot helping with a dropped part. On a factory floor, teams should compare the whole system—manipulation, mobility, power, and recovery—not assume that a bipedal body is the most efficient form for every station. The current evidence establishes a vendor-reported milestone, not a validated business case.
Limits and uncertainty
The source does not establish how many covers were sorted, what fraction were placed correctly, whether people intervened, or how charging was handled during the shift. “Autonomous” is Xiaomi’s characterization; the supplied material does not describe the system’s supervision, safeguards, or failure-handling protocol in enough detail to independently evaluate it. The video and company statements also do not show whether performance is repeatable across shifts, stations, or factories, or whether the robots are commercially available. The earlier reported percentages concern a different operation and cannot be treated as the success rate for side-cover sorting.
Reader takeaway
Xiaomi has put humanoid robots into a real EV-factory sorting task and claims they covered a full shift while correcting anomalies. Treat that as a meaningful deployment signal, not proof of production economics: the next useful evidence is verified output, quality, human-intervention rate, and uptime including charging.
Open question
Will Xiaomi publish output and intervention data for the eight-hour run, and how will those figures compare with the workstation’s existing production target?