LEAP Hand

The LEAP Hand is the cheapest credible route into 16-degree-of-freedom manipulation. Carnegie Mellon published the design in 2023 with a paper at RSS, then released CAD, firmware and a bill of materials. A US dealer sells it as a kit or assembled, which is how most labs outside Pittsburgh actually buy one.

Two versions are in circulation and they are not a straight upgrade. The v1 kit lists at $2,066 and drives all 16 joints with 16 Dynamixel XC330 servos. The v2 lists at $4,550, weighs less than half as much, assembles in under two hours instead of three, and adds soft joints that absorb contact. It does that with eight motors and a tendon per finger, so four of the 16 joints follow the others rather than moving independently.

What the price difference actually buys

If you rank hands by dollars per actuated joint, v1 wins and v2 looks like a regression: 120% more money for half the motors. If you rank them by mass at the wrist, assembly time, or how the hand behaves when it hits something, v2 wins. Both statements are true, which is why a single ranked list of hands is not useful and this site publishes records instead.

The 2023 paper reported beating the Allegro Hand across the authors' experiments at roughly one eighth of the cost. That claim is worth checking against a dated price, because the Allegro denominator moves between $13,000 and $21,450 depending on which listing you read. The side-by-side comparison keeps both numbers visible.

What the spec sheet does not settle

• Mass. One dealer blog says 0.5 kg for the LEAP Hand; the same dealer's later guide says 1.2 kg for v1 and 0.5 kg for v2. The likely explanation is that the blog described v2 or a Lite build, but neither page states it, so both figures stay in this record.
• Force. The 5–15 N range is configuration dependent. Fingertip force changes with servo choice, tendon tension and joint angle, and the vendor does not publish a test protocol.
• Simulation parity. v1 has published Isaac Gym and Isaac Lab environments. We found no equivalent statement for v2, which matters if your work depends on sim-to-real transfer rather than hardware alone.

Who should not buy it

Buy it if you have someone who can solder, tension tendons and debug a Dynamixel bus. Do not buy it as a first robot hand for a demo that has to work next week, and do not buy it expecting tactile sensing: there is none, and adding fingertips changes the mass and the grasp geometry. A lab that needs five fingers, sealed hardware and a support contract should look at the Inspire RH56 instead, and one that needs per-joint torque sensing should look at the Allegro.

Specification

The right-hand column records where each field comes from: vendor documentation, a dealer page, or a conflict between sources. Fields marked unverified have no attributable public value.

Price observations

Tier A: documented list price. B: sources disagree by more than 15%. C: estimate or press figure. D: quote only. See methodology.

What is not verified

Questions buyers ask

Is the LEAP Hand v2 an upgrade over v1?

Not on every axis. v2 is lighter, faster to assemble and more compliant, but it uses 8 motors instead of 16, so four joints per hand follow the tendon rather than moving independently.

Why does the bimanual pair cost exactly twice a single kit?

Because it is two kits in one order. The dealer lists no pair discount, and $4,133 is exactly 2 × $2,066.

Does the LEAP Hand have tactile sensing?

No. There is no force or contact sensing in the fingertips. Joint current can be read from the servos, which is not the same measurement.

Sources

1. Carnegie Mellon LEAP Hand project page. Design files, firmware, parts list. 2026-11-05
2. Shaw, Agarwal, Pathak. LEAP Hand, RSS 2023. 2023
3. US dealer LEAP guide and 2026 use-case playbook. Listed prices, mass, assembly time. 2026-11-05

Record log

• 2026-11-05Record created. Dealer prices and v1/v2 fields captured.