On November 3, Xiaomi Tech posted on X that its “full-size humanoid robot takes on an 8-hour task at the Center Console Side-Cover Sorting Workstation at Xiaomi Auto Factory.” Electrek reported it the next day. The number in the post is eight hours. The post gives no success rate, and Electrek notes that Xiaomi did not say how many times a person had to step in.
This is a follow-up to our August 18 piece, when Xiaomi put per-station success rates into its quarterly results. The delta is duration. The rate is still missing.
What was shown
According to Electrek, the robot pulled side covers from bins, carried them to storage racks and placed them into slots, and “covered the full day of production needs at that workstation.” That last phrase is the outlet’s wording, not a Xiaomi figure. The video carries a running clock of cumulative working time and an “Autonomous, 100x Speed” label, so the footage is compressed. Electrek also describes a segment with two legged robots and a wheeled upper-body humanoid, but gives no total robot count.
Nothing is said about power: no battery swaps, tether or charging breaks. Nothing is said about parts count either, so there is no way to turn eight hours into cycles.
Where eight hours sits
Per Electrek, Xiaomi’s earlier disclosures run like this. In March, a humanoid worked the self-tapping nut station at its Beijing EV plant and reported a 90.2% success rate over three hours, fast enough for the station’s 76-second cycle. In July, Xiaomi said that rate was up to 98% and added two harder stations, one of them the side-cover sorting job. The August 18 results then listed the nut station at 98.0% and console side-panel sequencing and crate folding at 90% each. The side-cover station in this video appears to be the same one as the 90% station, though Xiaomi’s post does not say so.
So the progression is three hours at one station in March, to an eight-hour run at a harder station in November. That is a real step. Duration is the number a plant manager asks about second. The first is how often someone has to walk over.
At 90%, a station fails one cycle in ten. An eight-hour clock tells you the robot did not stop for eight hours. It does not tell you whether it recovered by itself from the misses, or whether someone cleared them. Without that count, eight hours of runtime and eight hours of unattended work are different claims.
The same gap, one day earlier
On November 3, AgiBot ran six G2 humanoids in a Shanghai convenience store for 24 hours, logged 1,730 tasks, and passed four of six on its own review, with no intervention count either. Two Chinese companies published endurance runs within a day of each other. Both gave a duration and a task total, and neither gave the figure that decides staffing: interventions per hour.
Xiaomi’s run is narrower than AgiBot’s, one robot at one station against six robots across a store, and more consequential. A convenience store tolerates a dropped cake. An automotive line with a takt time does not tolerate an empty slot.
What it means for careers
Sorting parts from bins into rack slots is line-side logistics and kitting work. It sits between the assembly operator and the forklift driver, and it is the layer conventional automation never took, because bins are loose and parts face different ways.
The Xiaomi disclosures so far do not show a job going away. They show a station being held for a full shift by a robot, with the success rate and intervention count withheld. The signal that moves staffing is a number the company has not yet published for this station: its success rate over a full shift. The next place it could appear is Xiaomi’s quarterly results, where it published the station-level rates in August.
Until then, the honest reading is that a plant now has a humanoid that can fill a workstation for eight hours, and that it is not yet known how many humans were needed to keep it there.