Industry Brief, September 30, 2026: A tugger reorder, AI-built vision apps, robot shifts

By MillBrief Editorial · September 30, 2026

Four stories today, all from company announcements, so read them as the vendors’ side of the story. An autonomous tugger maker says a large customer has ordered its product for another job. A machine vision software company will show AI agents that draft vision programs from a plain-language request. A robot startup says the number that matters is how long a robot works before someone has to help it. And a Chinese maker of robots for work at height is taking a coating robot for ship hulls and tanks abroad for the first time. As always, the writing is ours, each item links to the original outlet, and vendor numbers are labelled as vendor numbers.

A tugger customer comes back for another job

Cyngn, a Nasdaq-listed developer of autonomous vehicle technology for industrial sites, says a Fortune 100 heavy equipment manufacturer has placed a follow-on order for its DriveMod Tugger, according to a company press release published by Robotics Tomorrow. The new deployment will move components to a heavy equipment assembly line at the customer’s plant, a new material-handling application that adds to its earlier use of Cyngn’s technology. Cyngn does not name the customer, citing competitive reasons, and the release gives no unit count or order value.

Cyngn’s figures for the product: the tugger tows up to 12,000 pounds (our conversion: about 5,440 kg), works indoors and outdoors, needs no magnetic tape or guide wires on the floor, and targets a typical payback period of under two years. DriveMod is currently available on Motrec MT-160 tuggers. Cyngn presents the order as an example of its approach: start with one repetitive route, measure the value, then expand. Its EasyLaunch programme is built around that, and it offers a route assessment on its website.

Why it matters: For a mid-size plant, the useful model here is the small first step, one fixed route that a person drives many times a shift, rather than the unnamed customer’s size. “Under two years” is a target, not a guarantee, so work out your own payback from what that route costs you today, including the driver’s time and any damage or near-miss history. Our ROI and payback guide walks through the sums, and what to automate first covers why a dull, repeatable job is usually the right place to start.

Source: Robotics Tomorrow, September 29, 2026 (Cyngn press release).

AI agents that draft machine vision programs

MVTec, the Munich machine vision software company, will show a prototype called the AI Vision Solver at the VISION trade show in Stuttgart, October 6 to 8, 2026, the company said in a release carried by Robotics Tomorrow. A developer describes the task in plain language and with images. AI agents then pick operators from MVTec’s HALCON library and generate an executable vision application. When the software finds syntax errors or logical gaps, it feeds them back to the agents, which try again. The agents draw on a knowledge base supplied by MVTec. Booth demonstrations will include barcode reading, matching and optical character recognition.

The prototype runs in HDevelopEVO, a new development environment that MVTec says will gradually replace HALCON’s current one starting in 2027. MVTec has used deep learning for classification and inspection since 2016; the new step applies AI to building the application itself. The company is careful on responsibility. Its Director of Marketing and Business Development, Dr. Maximilian Lückenhaus, says AI does not take full responsibility for an industrial application and the final assessment stays with the user.

Why it matters: Engineering time is often a large part of the cost of a custom vision station, so tools that shorten development could lower quotes or speed up changeovers. But this is a prototype with no release date, and a program that compiles is not the same as one that holds up under your lighting, part variation and line speed. Whoever builds your system, insist on a validation run with your real parts, good and bad, and written acceptance criteria before sign-off. Our RFQ guide shows how to put those criteria in the request for quote.

Source: Robotics Tomorrow, September 28, 2026 (MVTec announcement).

A robot maker says to judge robots by the whole shift

DYNA Robotics has launched DYNA 2.1, a semi-humanoid robot with an upper torso, two arms that take parallel-jaw grippers or dexterous hands, and four steerable wheels, according to a company release published by Robotics Tomorrow. DYNA says it can complete a whole commercial laundry shift without human help: loading and starting washers and dryers, unloading towels, folding and stacking them by size, shelving the stacks and recovering from mistakes such as a dropped towel. It is being deployed in hotels, laundromats and restaurants. The company says its underlying AI model was trained on a million hours of human and robot data.

The more interesting part is the yardstick. DYNA says it optimises for mean time between interventions (MTBI): how long the robot runs before a person has to step in. Co-founder Jason Ma argues this reflects commercial use better than the success rate on single, isolated tasks. The release does not give an MTBI figure for DYNA 2.1, nor a price.

Why it matters: The settings DYNA names are hotels, laundromats and restaurants, not factories, but its metric is worth borrowing. Whether you are looking at a mobile robot, a bin-picking cell or anything sold as “AI”, ask the vendor how often a person had to intervene per shift in comparable deployments, and make it an acceptance test you measure on your own floor. A 98% pick success rate can still mean someone walking over every few minutes (our arithmetic: at 600 picks an hour, 2% is 12 misses an hour, one every five minutes, if each one needs a person). Frequent rescues are one of the costs we list in the hidden costs of automation.

Source: Robotics Tomorrow, September 29, 2026 (DYNA Robotics press release).

A magnetic robot that coats steel hulls and tanks

RobotPlusPlus, a Chinese developer of robots that work at height, showed its HighMate series at SMM 2026 in Hamburg, according to a company release in Robotics Tomorrow. The lead product, the HighMate AP-F-CO coating robot with a VOC recovery system, is entering international markets for the first time. It attaches by permanent magnet, handles curves as tight as a 3 m radius (about 9.8 feet), carries two spray nozzles and collects overspray as it works. The company quotes a coating rate of 300 to 500 square metres per hour (our conversion: about 3,230 to 5,380 square feet).

Also on show: the HighMate AP Ex series, which RobotPlusPlus says is certified for ATEX Zone 1 explosive atmospheres, takes interchangeable hydro blasting, cleaning, coating and abrasive blasting modules, and delivers more than 40 square metres per hour with over 99% wastewater recovery; a C20 robot for cleaning cargo holds; and a magnetic crawler for surface preparation on curved surfaces such as ship hulls. The company was founded in 2015 and says it holds more than 480 patents.

Why it matters: This is for fabricators and maintainers of large steel structures (tanks, vessels, hulls), not for general machine shops. There, coating and blasting at height are dangerous, messy and dependent on a skilled painter, so a robot that keeps people off the staging has a safety case as well as a cost one. The quoted rates are the vendor’s and will depend on surface shape and preparation, so ask for a trial on your own work. As a Chinese supplier new to many export markets, check who will supply spares and service near you, and whether tariffs or import rules affect the landed price. Our integrator question list covers the support terms to get in writing.

Source: Robotics Tomorrow, September 28, 2026 (RobotPlusPlus press release).