On November 17, at the APEC CEO Summit in Shenzhen, Xiaomi founder Lei Jun spent part of his speech on robots.
He said the company has been building robots for six years, and that its embodied AI robot recently joined a Xiaomi factory as an intern — working a full 8-hour shift and recharging itself twice along the way.
What 8 hours means
In manufacturing, 8 hours is normally a shift. A robot that holds a full shift is no longer a demo; it is scheduled into production.
The detail Xiaomi disclosed: the station sorts centre-console side panels, the task ran 8 hours, the robot corrected its own errors when they occurred, and it coordinated with the line's control system.
Earlier milestones came in steps. In March, at a station fitting self-tapping nuts, the robot worked 3 hours unattended with a 90.2% success rate on two-sided installation, meeting a 76-second takt time. By July, the same station reached 98%, and the scope expanded to panel sorting and box folding and recovery.
From 90.2% to 98%, the failure rate fell from 9.8% to 2% — nearly a fivefold improvement, measured at the same station, so the comparison holds.
Changing stations is harder than adding hours
Three hours to 8 hours looks like 2.7 times, but the two figures come from different stations and cannot be lined up.
A 76-second takt means one unit every minute and a bit. Panel sorting is not a takt problem; it is a materials problem. The panels are flexible, they slide, they stack, and the grasp point differs every cycle. That job has belonged to people, because a small error in vision or force control means a missed pick.
Holding that station for 8 hours suggests the tolerance band on perception and force control has widened.
At home, a different approach
Outside the factory, Xiaomi ran a second trial in its staff apartments. An agent system built on the Xiaomi-Robotics-1 foundation model ran for more than 30 minutes, covered 3 rooms and completed over 50 tasks, including tidying, waste removal and room inspection.
Thirty minutes, 3 rooms, 50 tasks — under 40 seconds each. That density does not look like a stage demo.
The difference between the two settings is not difficulty but variables. A station's materials are fixed; in an apartment every table is different, with one more cup today and one fewer stool tomorrow. Hence two systems: vision and force control for the station, a foundation model for the home.
Three numbers nobody gave
Lei Jun offered a judgement: the imagination around embodied robots should exceed the car industry.
He also offered the other half. Over four years, investment in large models has been enormous, while the number of companies that have genuinely landed deployments, closed a commercial loop and made money is relatively small.
Xiaomi supplied a baseline of its own: 11.6 billion devices connected to its IoT platform, and factories already running without workers.
But three numbers are still missing on the robot line: how many units are deployed, what one costs, and how many jobs the stations replaced. Without them, 8 hours is a technical validation, not a business.
What suppliers read
For component makers, the value of the announcement is the station list. Panel sorting, box folding and recovery, self-tapping nut fitting — every station added to that list means a set of parts must now be delivered to automotive-grade standards.
The longer the list, the more specific the demand for reducers, torque sensors and dexterous hands.
From validation to production
A full shift proves endurance at one station. Production asks more: mean time between failures, spare parts, and who is called when the robot stops at 2am.
None of that was disclosed. Nor was whether the robot ran the same station for 8 hours on consecutive days, which is the difference between a record attempt and a schedule.
The comparison inside Xiaomi's own timeline is more useful than any cross-company one. The same nut-fitting station went from a 3-hour run at 90.2% to 98% in four months. That is the pace at which these systems actually improve — incremental, station by station.
Xiaomi has an advantage most robot startups lack: its own factories and its own retail network. A station can be tested without a customer's permission, and a working result can be shown to buyers without a sales cycle. That vertical integration is why the company can publish station-level metrics at all.
*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*