Robotics · 31 October 2026 · 4 min read

ABB’s fashion cell claims a 75 percent cut in setup. That is not last week’s 90 percent

The company says Fashion Inductor, with AI Robot Trainer built in, is its first end-to-end vision module for fashion logistics. Throughput is up to 1,300 items an hour per cell. Picking accuracy is a target above 99.5 percent, not a result a customer has published. The 75 percent is ABB’s figure for engineering and commissioning.

A white industrial robot arm holds a wrapped item above a table, between two grey totes, with blue lights and a black curtain behind. — ABB’s photograph of a Fashion Inductor cell: an arm over a tote, holding a wrapped item. It is the company’s demonstration setup. It is not a customer’s warehouse, and it is not a counted 1,300 items.

ABB Robotics has posted a release, without a date on the page, for what it calls its first end-to-end vision-based industrial physical-AI module: Fashion Inductor, with AI Robot Trainer built in. The company put the product up on 15 October, and the product manual carries that date. The job is sorting fashion goods, where the item, the packaging and the light keep changing. ABB says engineering and commissioning effort falls by up to 75 percent. That figure is the company’s. The release does not name the comparison, or a warehouse that has run the cell.

The performance lines are also the company’s, and one of them is written as a target. Throughput is up to 1,300 items an hour per cell. Picking accuracy “targets above 99.5 percent.” A target is not a miss rate someone else has counted. This is not the announcement of 17 October. That one attached AI Robot Trainer to a different vision product, and claimed up to 90 percent less commissioning time and up to 50 percent less cycle time. Seventy-five and ninety are two of ABB’s figures, about two jobs. They do not add up to a third.

The new piece ABB describes is a digital twin it calls “an exact replica of the module software,” connecting its application controller, automatic path planning, the trainer and RobotStudio. It says an integrator can change the setup, including grippers from another maker. It also describes a loop that watches the cell, finds anomalies, makes new training pictures and sends an updated model back to the robot. That is a workflow the vendor is selling. It is not an independent test of how often the arm misses when the garment changes.

Craig McDonnell, managing director of the industries business at ABB Robotics, said physical AI “fundamentally changes what robots can do,” and that variable work such as fashion has been hard to automate. The sentence that would change a buyer’s sum is a named site, a counted hour, and an accuracy someone other than ABB measured. None of those is in the release. Up to 1,300 items is a ceiling on a cell. It is not a day’s work already done.