NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

Hundreds of Robots on the Line, Working to Takt Time

A wire harness has no fixed grasp point. Its length, branches and connector positions change with the vehicle model.

2026-11-21 · 4686 words · NeuroAI
Hundreds of Robots on the Line, Working to Takt Time

A wire harness has no fixed grasp point. Its length, branches and connector positions change with the vehicle model. There is no locating hole to reuse, no jig that fits the next car down the line.

The part that resists automation is not steel. It is something soft.

That is where hundreds of robots now work, at a plant in Changzhou run by Aiboge.

The harness is soft

A harness is crimped from wire, terminals and sheathing. Bend it and it holds a new shape. Every unit arrives in a slightly different pose.

Hard parts can be pinned in place. A harness cannot. Inserting a connector also needs alignment and force control: too deep and the terminal deforms, too shallow and it never seats.

Soft material plus force-sensitive insertion is why harness assembly is treated as one of the hardest tasks to automate.

This is the line between a demo and a delivery. A demo lets a person arrange the harness in the friendliest possible position. A production line does not.

One model ships in several configurations, and each rewires the branches; changing configuration recalculates grasp points and insertion order.

Takt belongs to the line

Takt time is the interval at which a line produces one unit. Daily output divided by effective working hours gives the number.

A robot at a station must finish inside that window. Faster buys nothing. Slower stops the line, because the next unit is already on the conveyor.

The window is set by the slowest station, not the fastest.

That is why working to takt tests something different from success rate. It tests several thousand consecutive cycles without a miss. One failure in those thousands is one line stop.

The first question a robot faces on a line is not whether it can do the task. It is how often it fails.

Three numbers, three ledgers

Hundreds of units have entered the Aiboge plant. The contract carries a starting volume of 1,000. The stated target is 10,000.

Three figures, three ledgers. Hundreds are installed. One thousand is written into a contract. Ten thousand sits in a target column.

They cannot be added, and they cannot substitute for each other.

Only the first has happened.

Ten thousand is at least 10 times hundreds: against 999 units the multiple is just over 10, against 200 it is 50.

Ten times is the floor, not the forecast.

The gap between hundreds installed and 1,000 contracted is the gap between a site and a delivery schedule. Count the site and you can point at machines; count the contract and you wait.

Demo versus the nine-thousandth repetition

A demo is done once, for an audience. Takt is done correctly in every window. The first allows a retry. The second does not.

Two engineering problems sit behind those requirements: tuning a motion until it works, and driving failure rate down to what a line can absorb. The second takes longer and shows badly on a stage.

Launch events review the first attempt. A line reviews the 9,000th.

Tashi Zhihang, the supplier, describes the deployment as running at "real takt." Those two words land on one metric: minutes of line stoppage.

Line-stop time is the first field on an acceptance sheet.

Two checks of $500 million

Genesis AI and Agile Robots each raised $500 million on the same day. SoftBank led the Agile Robots round.

Both went into physical AI. Together, $1 billion.

Most projects in this category are still at launch, trial or planning stage, with few confirmed cases of scaled production use. That sets the order of questions: confirmed production use first, round size second.

A billion dollars buys time, not takt.

Money reaches a valuation in an afternoon. A line is installed machine by machine, shift by shift.

Nor does $500 million convert into units. Unit price and order value are undisclosed. The confirmed endpoints are the round and the line; the distance between them cannot be filled in from outside.

Eyes and hands

Tieke Yingmai, working with Huawei, showed an onboard inspection system: a 500 mm body that identifies seven defect types in real time, including loose fasteners, rail-surface scuffs, mud pumping and transponder faults.

Inspection is looking. Assembly is doing. Looking covers fasteners, surfaces, signals; doing covers gripping, aligning, inserting, tightening.

A missed defect can be reviewed later. A bad insertion has to be reworked, and a rework can stop the line. Inspection fears false negatives. Assembly fears downtime.

A camera can be a beat slow. A hand cannot.

The 500 mm is a body dimension. What harness assembly needs is force control in a pair of hands. One design makes the machine small. The other makes it steady.

Inspection is graded on miss rate. Assembly is graded on takt and rework rate. Same machine, different line, different exam.

The 10,000 column

Ten thousand is written in the target column.

Getting there means passing 1,000 first. Every additional thousand recalculates spare-part types and inventory depth: each new station adds a calibration set, a spares list and a maintenance path.

These numbers are not public, so they are not estimated here.

The Aiboge line runs at its own takt now. A robot that keeps up is on the line. One that does not is still in demonstration.

Ten thousand also needs spares, maintenance, training and a stoppage plan. Spares are stocked by model, maintenance is scheduled by shift, training is broken down by station. Miss one and the takt breaks.

None of that appears at a launch. All of it appears on a handover sheet.

Hundreds are on the line. Ten thousand is still a target.

*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*

Related coverage

Cite this article

NeuroAI (脑机网), "Hundreds of Robots on the Line, Working to Takt Time," 21 November 2026, https://neuroai.site/a/na-corp-embodied-robot-line-1121

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Companies & Stack” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories