Analysis
What hand spec sheets leave out, and why it costs money
Fifty kilograms of load. Twenty newtons of fingertip force. A million cycles. Three numbers, three different kinds of claim, one table.
The short version
Spec sheets mix measured values, structural ratings and laboratory claims in the same column. The omissions that matter most are the test condition, the configuration and the service life, and all three change the purchase decision.
A LinkerBot O6 page states a 50 kg load, a 20 N maximum thumb fingertip force and a 70 N five-finger grasp. Those three numbers sit within four lines of each other and cannot be compared. One is probably a structural rating, one is a pinch force, one is a grasp force. Nothing on the page says which is which.
This is normal in a young component category, and it is expensive. A buyer who specifies "50 kg load" writes a requirement no vendor can be held to, and a buyer who compares it with another vendor's 15 N fingertip force is comparing a structure with a measurement.
Three kinds of number
| Claim | What it usually means | What is missing |
|---|---|---|
| 50 kg load | Structural rating or suspended load | Test condition, direction, duration |
| 15 N fingertip | Measured pinch force | Joint angle, contact area, protocol |
| 1 million cycles | Laboratory fatigue test | Load profile, sample size, failure criterion |
| ±0.2 mm repeatability | Positioning repeat | Payload and speed at which it holds |
PaXini does this better than most. Its English page states 15 N and one million cycles and adds a footnote saying the figures come from the company's own laboratory and vary by application. That footnote is worth more than the numbers: it tells you the claim is a lab result, not a datasheet guarantee.
The configuration problem
The second omission is which build the number describes. The Unitree Dex5-1 and Dex5-1P share every mechanical field and differ by 94 pressure sensors. The Inspire RH56 family has five variants between 540 g and 790 g, and a dealer quote that says "RH56" has not told you which one.
This is where money leaks. A purchase order written against a family name can be filled with the cheapest member of that family, legally, and the buyer discovers it during commissioning. The RFQ template puts the exact model string in field one for that reason.
The omission that costs the most
Service life. Almost no vendor publishes a replacement interval, and the ones who do publish a cycle count without a load profile. Meanwhile an industry estimate puts hand repair or replacement at roughly every three months at 10,000–20,000 RMB per event. For a fifty-robot fleet that is a line item in the tens of millions of RMB a year, and it appears in none of the spec sheets used to choose the hand.
The practical fix is to ask for three numbers that are not on the page: the price of one finger assembly, its lead time, and the failure rate the vendor sees in the field. A vendor who answers is selling a product. A vendor who does not is selling a shipment.
Questions
Are vendor load figures unreliable?
Not necessarily. They are usually accurate for a condition that is not stated, which makes them unusable for comparison rather than wrong.
How do I compare two hands fairly?
Ask both vendors for fingertip force in newtons at a stated joint angle, the exact model string, and the price of a replacement finger. Compare those three together.