Wrist Torque: A 67.5x Spread Across 89 Robots We Track

Rated payload tells you what a robot can lift in a straight line. It tells you nothing about what the wrist can twist. Those are two different specs, and the gap between them is wider than most buyers assume: of the 336 robots in our database, only 89 (26.5%) publish a wrist torque figure alongside their rated payload, and among those 89, the ratio of wrist torque to payload spans from 0.26 Nm/kg to 17.29 Nm/kg, a 67.5x spread. A robot’s payload number will not tell you which end of that range it sits on.

What does wrist torque actually measure, and why isn’t it just payload?

Rated payload is a mass figure: how much weight the flange can hold and move through a trajectory without exceeding the arm’s structural and motor limits. Wrist torque (and its companion spec, wrist inertia) measures something else entirely: how much rotational, or moment, load the last three joints (J4, J5, J6) can sustain. Torque is a rotational force, force applied at a distance from a pivot point, measured in newton-meters, and it is not interchangeable with a straight-line weight rating.

The distinction matters because of where the load sits, not just how much it weighs. A gripper mounted dead-center on the flange, holding its payload close to the wrist’s rotation axis, puts almost no extra moment load on the wrist. The same mass mounted off-center, or extended out on a long tool, or asymmetric in shape, creates a much larger moment arm, and the wrist has to resist that twisting force on every move, not just carry the weight. As the general robotics explainer at amdmachines.com puts it, the moment and inertia loads at the wrist “depend on the center of gravity of whatever you’ve mounted,” which is exactly why a gripper’s shape and offset can matter more than its weight.

Our companion piece, what payload rating doesn’t tell you, walks through the underlying physics, the J = m x r^2 moment-of-inertia calculation you can run by hand to check a specific tool. This post is not that explainer. It is the data: how many robots actually let you check that math against a published spec, and how far the real numbers spread once you do.

How many robots actually publish it?

Not many, and coverage is not even close to uniform across robot classes. Wrist torque coverage by type, out of the robots in each category (AMRs excluded, they have no tool flange):

SCARA robots and delta robots both flatline at zero: neither type publishes a single wrist torque figure in our database, likely because their kinematics are built for planar or vertical pick-and-place, not for absorbing an off-axis moment load, so the spec rarely gets tested or printed. Articulated arms are the only type where a buyer can reliably find this number, and even there it is under half the catalog (65 of 136, 47.8%). Welding robots (6 of 6) look perfect, but six robots is not a market, it is a coincidence of a small, spec-heavy niche where every datasheet happens to be thorough.

How wide is the wrist-torque-to-payload spread?

Among the 89 robots that publish both figures, dividing wrist torque (Nm) by rated payload (kg) gives a rotational-capacity-per-kilogram number, and it swings by a factor of 67.5 from one end of the database to the other.

The low end, the robots with the least wrist torque relative to what they can lift, meaning the least margin for an off-center or asymmetric tool:

The high end, the robots with the most rotational headroom per kilogram of rated payload:

Source: our analysis of the 89 robots in the Industrial Robotics Hub database publishing both performance.wristTorqueNm and performance.payloadKg.

The headline number: Yaskawa PL80 sits at 0.26 Nm/kg, the lowest ratio in the dataset. FANUC M-2000iA/1700L, which also happens to carry the highest rated payload in our entire database at 1,700 kg, tops the ratio table at 17.29 Nm/kg. Divide the two and you get 67.5x. That spread exists entirely among robots that already publish both numbers; it says nothing about the roughly three-quarters of the database that publishes neither, where the ratio could sit anywhere.

The Yaskawa pattern is the more interesting thread here. The same brand holds both the single lowest ratio in the whole database (PL80, a palletizer, 0.26) and two entries that sit near the top: the Yaskawa SP100 welding robot (6.96) and the Yaskawa MPX3500 painting robot (15 kg payload, 93.2 Nm torque, ratio 6.21). Three Yaskawa robots, three different types (palletizer, welding, painting), and they land at opposite ends of a 67.5x spread. That is not a brand-quality signal. It is a job signal. A palletizer moves a fixed, centered case straight up and down and needs almost no rotational margin at the wrist; a welding torch or paint gun needs fine, continuous wrist dexterity and gets built with torque to spare. The ratio tracks what the wrist is asked to do, not which logo is on the base or how much the arm can lift.

Why does the cobot segment publish this least, when cobot buyers need it most?

This is where the coverage gap stops being an annoyance and starts being a real buying risk. Of the 120 cobots in our database, only 15 (12.5%) publish a wrist torque figure. Worse: of the 16 distinct brands that sell a cobot in our database (ABB, AUBO, Dobot, Doosan, FANUC, Han’s Robot, JAKA, Kawasaki, KUKA, Mitsubishi, ROKAE, Siasun, Staubli, Techman, Universal Robots, Yaskawa), only 4 (ABB, FANUC, Kawasaki, Yaskawa) have even one cobot model with a published wrist torque number. The other 12, including Universal Robots, the market’s highest-volume cobot brand, plus Doosan, AUBO, Dobot, Han’s Robot, JAKA, Mitsubishi, ROKAE, Siasun, Staubli, and Techman, publish zero wrist torque data on any cobot in their entire lineup.

That is backward from where the risk actually sits. Cobots have small, PFL-limited wrists by design, built to stay within force and power limits for safe human proximity, and cobot buyers are disproportionately the ones bolting on a third-party gripper, a vision bracket, or a long-reach tool that was never engineered as a matched pair with the arm. An off-center gripper or a part with real rotational inertia is exactly the load that stresses a wrist joint beyond what its payload rating implies, and it is exactly the buyer who is most likely to encounter that load who has the least published data to check it against. Heavy, fenced articulated arms sold through integrators who demand a complete datasheet before quoting a cell get this number far more reliably (47.8% coverage) than the cobot segment that ships direct-to-floor with a snap-on tool ecosystem.

Here are the 15 cobots that do publish it, the practical shortlist if wrist torque is a hard requirement for your application:

Source: our analysis of the Industrial Robotics Hub database, cobots publishing both performance.wristTorqueNm and performance.payloadKg, sorted by ratio.

Notice these 15 cobots cluster in a much narrower ratio band (1.28 to 3.80 Nm/kg) than the full 89-robot dataset (0.26 to 17.29). That is partly a real design pattern, cobot wrists are engineered with less rotational overhead by nature of the collaborative force limits, and partly just the effect of a small sample from a handful of transparent brands. Either way, if your build calls for a heavy or off-center end effector on a cobot, this table is close to the entire public dataset available to check against.

Which brands are actually transparent about it?

Restricting to brands with at least 15 robots in the database, the share of each brand’s own lineup that publishes wrist torque:

Yaskawa leads at 90.0% of its own lineup (27 of 30), which lines up with the ratio spread it produces on its own: a brand that publishes the number consistently across palletizers, welding arms, and painting robots is also the brand that shows you how much that number varies by job. FANUC (68.0%) and Kawasaki (62.5%) follow. ABB and Estun both publish it on under a fifth of their catalogs (17.9% and 13.8%) despite large overall lineups, and three brands with 15+ models each, Epson, KUKA, and ROKAE, publish it on nothing at all.

What should you actually check before you buy?

If your build calls for an off-center gripper, a long lead-through tool, or any end effector with real rotational inertia rather than just mass sitting on the flange, the rated payload number will not tell you whether the wrist survives it. The 67.5x spread between the lowest and highest wrist-torque-to-payload ratios in our database proves the two specs do not scale together, not even loosely, across robot types, brands, or payload classes.

Do not assume it. Ask the integrator or manufacturer for the wrist torque and wrist inertia spec directly, the same way you would ask for payload at full reach instead of trusting the headline rated-payload number. If you want to check your own tool against a candidate robot’s published figures by hand, our companion piece on what payload rating doesn’t tell you walks through the J = m x r^2 calculation. And if you are speccing a cobot cell specifically, treat the absence of a published wrist torque number as the default, not the exception: only 4 of 16 cobot brands publish it at all, and the 15-robot table above is close to the entire public reference you will find.