Industry Brief, September 17, 2026: Order books, spec sheets, and verification

By MillBrief Editorial · September 17, 2026

Trade-show week, which means spec sheets. The useful exercise for a buyer is separating what a vendor has shipped from what it has announced, and both from what an outside party can confirm. Today: a humanoid maker whose order book and whose filed revenue tell different-sized stories, a cobot platform refresh with numbers you can check against the model names, a robotics editor giving an unusually straight answer about safety, and a panel preview that gets the hard part of Robots-as-a-Service right. Our own words, links to the original outlet, vendor numbers labelled as vendor numbers.

Agility’s Digit 5: the specs are claims, and the order book deserves a second look

Agility Robotics announced Digit 5, with reworked legs and knees, new batteries, and a rebuilt safety architecture. The company’s figures: a 50 lb (22.7 kg) repeated lift, 90 minutes of runtime against a 9-minute full charge (a 10:1 run-to-charge ratio, up from 2:1 on Digit 4) and more than 20 working hours in a 24-hour day. It stands 5 ft 11 in (1.81 m), weighs 284 lb (129 kg), and reaches up to 7.2 ft (2.2 m), against 5.5 ft on the previous generation. Early access is slated for the first half of 2027, general availability by the end of 2027, with the first international deployments in the EU and UK. All of those are Agility’s numbers; no independent testing accompanies them.

The number worth pausing on is commercial. CEO Peggy Johnson cites more than $300 million in multi-year orders. The Robot Report sets that against the company’s SEC filings, which show $1.8 million in net sales for 2025, alongside a $2.5 billion SPAC valuation. Those figures are not contradictory (multi-year orders are booked over years and a 2025 filing predates this announcement), but the distance between them is the whole story of humanoids right now.

Why it matters: An order book is a statement of intent; recognized revenue is a statement of delivery. For a buyer, the practical question is which one your supplier’s support organization is staffed against. A vendor shipping $1.8M of product is not yet running the service network that a $300M installed base will eventually demand; that network is something they build with your deployment, not before it. Ask where you sit in the delivery queue and what the support commitment looks like in year three. This is the same supplier-durability question we put first on our integrator checklist, and it bears directly on the line items in the hidden costs of automation.

Source: The Robot Report, September 15, 2026.

Universal Robots’ Gen 7, and a spec-reading tip hiding in the model names

Universal Robots used IMTS to launch its seventh-generation platform: three new arms, a rebuilt controller, and a tool flange aimed at AI workloads. The arms are the UR10g-1750, UR17g-1300 and UR18g-950, and the naming convention is worth internalising: per the company’s own announcement, the models are “named after their payload (kg) and reach (mm).” So the UR10g-1750 is a 10 kg payload with 1,750 mm of reach; the UR18g-950 trades reach for an 18 kg payload. The CB7 Core controller is claimed at 40% more compute in a 30% smaller footprint than previous generations, and the TP7 Core teach pendant at more than 20% lighter. The new flange adds built-in force-torque sensing, impedance control, and Real-Time Data Exchange for wrist-mounted cameras and high-bandwidth sensors. The controller is stated as compatible with existing e-Series and UR Series force/power-limited arms. No pricing was announced. Every number here is Universal Robots’.

Why it matters: Two buyer-side notes. First, payload and reach trade against each other within a single generation: if a vendor’s flagship number is reach, check what payload you actually get at that reach, which is the spec that decides whether your part and gripper fit the envelope. Second, “AI-ready” in this announcement means concrete plumbing (sensor bandwidth at the flange, force-torque sensing, a real-time data channel), not autonomy you can buy. Read it as infrastructure for a future integration project and budget it that way. If you are sizing a first cell, what to automate first is the better starting point than a spec sheet, and cobot vs industrial robot covers the class decision underneath it.

Sources: The Robot Report, September 14, 2026; Robotics Tomorrow, September 15, 2026.

“There are no easy answers”: a straight answer on humanoid safety

IEEE Spectrum’s Evan Ackerman makes a point that the category mostly avoids: humanoid robot companies have been reluctant to discuss safety “because there are no easy answers.” His read on Agility’s approach is that it sidesteps the hard problem rather than solving it: Digit prevents contact instead of surviving it, autonomously avoiding people, stopping, setting down its cargo, and assuming a seated position as humans approach. Ackerman’s verdict is pointed: “It may not be elegant, but it works, which is more than can be said for any other commercial humanoid that I’m aware of.” Worth noting what the piece does not contain: no ISO or ANSI standard is named, and no detail is given on redundancy, safety-rated controllers, speed-and-separation monitoring, or force limiting. Ackerman also flags that power, safety, and battery life actively work against one another, and that the “drop-in replacement for human labor” framing lacks historical support.

Why it matters: For fenced industrial robots and for cobots, the safety conversation has decades of standards behind it and a conformity assessment you can demand paperwork for. For humanoids it is currently a design philosophy plus a vendor’s word. That is not a reason to dismiss the category, but it does change what you ask for: request the specific safety architecture, the standards conformity claimed, and who performed the assessment, and treat “the robot avoids people” as a design choice to be validated on your floor, with your traffic patterns, not as a certification. Our note on what engineers wish buyers knew covers the general habit of asking for evidence rather than assurances.

Source: IEEE Spectrum, September 15, 2026.

RaaS is a service business wearing a hardware costume

Ahead of a RoboBusiness panel next month, The Robot Report’s Mike Oitzman sets out why Robots-as-a-Service is harder than it looks: a successful RaaS business “is not just a robot sold on subscription, nor is it simply a rental financing model.” The dimensions he lists are all operational rather than technical: pricing structure, fleet maintenance and service coverage, uptime guarantees and SLAs, customer success metrics, sales compensation, and an organisation built around services rather than a pure engineering OEM. The pitch to buyers is the familiar one: convert a large capital investment into a recurring operating expense. This is a panel preview, not a study; the session (with Locus Robotics, Roboworx and Aescape on the panel) runs October 20, 2026, and the article carries no unit-economics figures.

Why it matters: The capex-to-opex swap is genuinely useful when cash is the constraint, but it is not free money: you are buying the vendor’s service organisation as much as the hardware, and the SLA is the actual product. Before signing, price the full contract term against outright purchase plus a maintenance contract, and read what happens at renewal, on early termination, and if utilisation comes in under plan. Note too that the failure modes Oitzman lists are the vendor’s, not yours, which is precisely why a RaaS provider’s service maturity deserves the scrutiny you would otherwise spend on the robot. Our ROI and payback guide sets out the comparison, and integrator rates covers the support costs that survive either structure.

Source: The Robot Report, September 14, 2026.