Industry Brief, September 21, 2026: Humanoid tally, fenceless guarding, high-mix cobots

By MillBrief Editorial · September 21, 2026 · Retrospective edition, published September 23, 2026

This is a retrospective edition. The brief did not run on September 21, 2026; we wrote it on September 23, 2026 using only reporting that was published on or shortly before September 21, 2026, and it makes no forecasts.

This is a retrospective edition covering September 21, 2026, using only reporting published on or up to two days before that date. Today: the first industry-wide count of humanoid robot sales, and how small it is next to ordinary robot arms; a camera-based system that lets a robot cell run without a fence; a FANUC engineer’s case that high-mix automation starts with the parts, not the robot; and a Korean robot maker betting on heavier, faster cobots. Our own words, links to the original outlet, vendor numbers labelled as vendor numbers.

About 7,000 humanoids were sold worldwide last year, and many are not doing productive work

Reuters, in a report carried by BusinessWorld, says around 7,000 humanoid robots were sold worldwide in 2025 for industrial and professional service use, according to figures compiled by the International Federation of Robotics (IFR) that Reuters reviewed before publication. The IFR is collecting humanoid data for the first time. To count, a robot has to look human and work autonomously in spaces built for people; it does not need legs. Consumer and military humanoids are excluded.

For scale, the report compares that with roughly 542,000 conventional industrial robots installed in 2024 and an estimated 199,000 service robots sold that year. The IFR’s secretary general, Susanne Bieller, told Reuters that many of the humanoids sold in 2025 were bought by research institutions or by companies using them to generate training data for AI models, rather than to do productive work. Carmakers, the early adopters, are running pilots with single-digit or sometimes double-digit numbers of robots per plant. Bank of America Global Research, by contrast, estimates 90,000 humanoid shipments this year and 1.2 million by 2030.

Why it matters: The gap between 7,000 humanoids and more than half a million conventional robots is the useful number here. Humanoids are still mostly in labs and pilots, so if a vendor pitches one for your line, ask for references from paying customers doing production work. For a plant weighing its first project, a proven arm or cobot remains the default; start with whether automation is worth it for a small manufacturer.

Source: BusinessWorld (Reuters), September 21, 2026.

A certified camera system that lets a robot cell run without a fence

Machine Design profiles SR-1, a fenceless guarding system from Sensory Robotics, a Cincinnati startup. Instead of hard fencing, four to eight 3D time-of-flight cameras surround the cell and build a live model of the space around the robot. Walls, pillars and known equipment such as conveyors or mobile robots are marked as expected and ignored, and a CAD model of the robot tracks the real arm. If a person or unknown object enters the inner zone, the robot stops; an outer zone can slow it instead, and it resumes once the space is clear.

The company’s figures: it detects objects down to 50 mm moving at up to 2 m/s, reacts in 250 ms, and needs about 30 percent less floor space than hard guarding. These are vendor claims. The article reports that SR-1 meets ANSI/RIA R15.06 and ISO 10218, is certified to PLd Category 3 under ISO 13849, and received a cULus listing (UL 1740) in May 2026. It works with robots from different makers and can be added to existing cells. An optional subscription, SR-Insight, logs events outside the approved risk assessment, such as a robot running faster than it should or equipment moved into the cell. The article also raises the obvious worry: without a physical barrier, people may walk in more often and trigger stops.

Why it matters: Fencing costs money and floor space, and every gated entry costs minutes of downtime. A certified alternative is worth a look for a crowded shop, but the risk assessment is still yours and your integrator’s. Ask which standards the system is certified to, who signs off the risk assessment, and how nuisance stops will be tracked. Those belong on your list of questions for a system integrator.

Source: Machine Design, September 21, 2026.

FANUC’s high-mix advice: count part families, not SKUs, and start with the high runners

Machine Design writes up an Automate 2026 session by Jerry Perez, executive director of global accounts at FANUC America, on high-mix, low-volume cobot work. His first point: the number of SKUs is not the real question. Look at what actually changes from part to part (geometry, size, handling, cycle time, quality requirements, work instructions). Thousands of SKUs may collapse into a handful of part families that a robot can handle.

He also warned that cobots are easier to program, not easy to deploy. Mounting, wiring, setting the tool centre point, and accounting for payload, inertia and safety settings can all take real engineering, and some problems only appear once the project is underway. On safety, he pointed to the revised ISO 10218 and ANSI/A3 R15.06, which assess the collaborative application as a whole rather than labelling the arm itself as safe. His practical advice: automate the high runners first, the small set of parts that makes up much of the volume, and do not aim for 100 percent of a process on day one.

Why it matters: This is a FANUC session, but the advice is vendor-neutral and matches what we tell buyers. Sort your parts into families before you talk to anyone, pick the highest-volume family as the first job, and budget for integration work even if the arm is a cobot. Our guide on what to automate first walks through the same selection.

Source: Machine Design, September 21, 2026.

HD Hyundai bets on heavier, faster cobots and a one-ton arm

Seoul Economic Daily reports, citing industry sources, that HD Hyundai Robotics plans to release the HDT1000-33, an industrial robot that can lift up to one ton with a 3.3 m working radius, in January. It is the largest of 14 new HDT-series robots, aimed at heavy handling in car and battery plants, a segment where the report names Kuka, Fanuc and China’s Estun as competitors.

The more relevant news for smaller plants is its cobots. In July it released three models, the HDC25-18, HDC35-18 and HDC50-17, its first new cobots in about six years, after pulling an earlier model and recalling delivered units. It calls them “hybrid” cobots: meant for heavier parts than the 10 to 15 kg class typical of cobots, at speeds closer to industrial robots. At a recent demonstration at its Yongin reliability centre, robots fitted with the company’s SafeSpace 2.0 system and radar sensors slowed and then stopped as a worker approached, and returned to full speed once the worker moved away. Its CTO says it is in supply talks with more than 30 potential customers, plans to deliver more than 100 units next year, and aims for over half of South Korea’s high-payload cobot market by 2030. (The article is an AI translation from Korean.)

Why it matters: Heavier cobots blur the line between cobot and industrial robot. A 35 or 50 kg cobot running fast will almost certainly need a proper risk assessment and sensing, much like the fenceless system above. When comparing quotes, compare payload, speed and the safety system as one package, not the arm alone. Our cobot vs industrial robot comparison sets out the trade-offs.

Source: Seoul Economic Daily, September 21, 2026.