As of September 28, 2026, the most articulated robot hands on sale or in production offer 20 to 22 actuated degrees of freedom: Unitree’s Dex5-S (22, from $6,500), Sharpa’s SharpaWave (22), Tesollo’s DG-5F (20) and the Shadow Dexterous Hand (20 actuated of 24 joints). Prices run from $3,500 for the open-source ORCA hand to quote-only research hands, and touch sensing ranges from none to more than 1,000 tactile pixels per fingertip.

Why it matters: the hand is where humanoid cost, reliability and data collection meet. A spec sheet that quotes degrees of freedom without saying how many are motor-driven, what the fingertips can feel or what the hand can lift tells a buyer very little.

Key Facts

  • Unitree Dex5-S: 22 DoF, every joint driven and back-drivable, from $6,500 (launched Sep 21, 2026)
  • SharpaWave: 22 active DoF, 1,000+ tactile pixels per fingertip, 0.005 N sensitivity; mass production announced Dec 16, 2025
  • ORCA: 17-DoF tendon-driven hand, $3,500; ORCA Touch with 351 taxels per hand from $6,100 (Mar 2026)
  • Figure 03 fingertips detect forces as small as 3 grams (Figure, Oct 9, 2025)
  • Tesla’s Optimus hand and forearm contain over 100 screws and small components, still hand-assembled (Electrek, Sep 25, 2026)

What does “degrees of freedom” mean for a robot hand?

A degree of freedom is one independent way a joint can move; the human hand is commonly cited at about 27 including the wrist. The number on a spec sheet can mean three different things. Total joints counts every hinge. Active or actuated DoF counts joints with their own motor, the ones software can command independently. Passive or underactuated joints move by coupling to a neighbour. Unitree’s Dex5-1, for example, has 20 DoF of which 16 are active and 4 passive; the Shadow Dexterous Hand has 24 joints, 20 of them actuated. Inspire’s RH56DFX has 12 joints driven by 6 micro linear actuators, trading dexterity for a 540 g weight.

How the motion is delivered matters as much as the count. Direct-drive hands put a motor at or near each joint, which makes them back-drivable and easy to model but heavy and warm. Tendon-driven hands, such as ETH Zurich’s ORCA and the Shadow hand, move motors into the palm or forearm and pull cables, which slims the fingers but adds stretch, friction and calibration work. Linkage hands, including many low-cost designs, couple joints mechanically to save actuators. Tacta Systems uses a fourth approach: fluid-filled tendons that push pistons, with no motors in the hand at all (our TactaBot report).

How do robot hands compare on DoF, touch and price?

The table uses maker specifications where published. Figures are as of September 28, 2026; “quote” means the maker publishes no list price.

HandMakerDoF (active)Tactile sensingPayload / forcePriceStatus
Dex5-SUnitree (China)22 (all driven)Not disclosedNot disclosedFrom $6,500Launched Sep 21, 2026
Dex5-1PUnitree (China)20 (16 active)94 pressure sensors3.5–4.5 kg; 1.1 kg handQuoteOn sale
SharpaWaveSharpa (Singapore)22 (22)1,000+ tactile pixels per fingertip; 0.005 NNot publishedNot publishedMass production (Dec 2025)
DG-5F-M / DG-5F-STesollo (Korea)20 (20)Not listed on product page20 kg (1.76 kg hand) / 12 kg (880 g hand)QuoteOn sale
Shadow Dexterous HandShadow Robot (UK)24 (20)Fingertip pressure sensors, 40 tendon load sensors; BioTac optionalNot published; 4.3 kg handQuoteOn sale
Allegro Hand V5Wonik Robotics (Korea)16 (16), four fingersFingertip pressure sensors (V5 Plus)15 kg (listing)About $11,000 (listing)On sale
ORCA / ORCA TouchORCA Dexterity (Switzerland)17 (tendon-driven)None / 351 taxels per handNot published$3,500 / from $6,100On sale
RH56DFXInspire Robots (China)12 joints (6 actuators)Force sensing, 0.5 N resolution15 N thumb, 10 N fingersQuoteOn sale
Tacta HandTacta Systems (US)15 (fluidic tendons)MEMS sensors, 1 mm pitch25 N per fingerNot publishedShips early 2027
NEO hand1X (Norway/US)22 per handNot specifiedArm payload 18 lb (robot)Only with NEO ($20,000)Integrated
Figure 03 handFigure AI (US)Not disclosedFingertips sense 3 g; palm cameraNot disclosedNot sold separatelyIntegrated
Optimus Gen 3 handTesla (US)22, plus 3 in wrist/forearm (Tesla engineer, 2024)Reported reliability issuesNot disclosedNot sold separatelyIntegrated

The Allegro payload and price come from a Humanoid.guide listing rather than a Wonik price list. A 25-DoF figure circulating for 1X’s NEO hand does not match 1X’s own spec page, which lists 22 per hand, so we use 22.

What kinds of touch sensing do robot hands use?

Most hands sense force in one of three ways. Joint or tendon sensing infers contact from motor current or cable tension; the Shadow hand carries 40 tendon load sensors, and Inspire’s RH56DFX reports force at 0.5 N resolution. Pressure arrays place discrete sensors on the pads, such as the 94 on Unitree’s Dex5-1P. High-density tactile skins read many “taxels” per fingertip: ORCA Touch outputs a 3D force vector from each of 351 taxels at 0.1 N sensitivity, and Sharpa claims more than 1,000 tactile pixels per fingertip. Figure takes a mixed approach on Figure 03, pairing fingertip sensors that detect about 3 grams with a camera in each palm.

Density is not the same as usefulness. A taxel array has to survive repeated contact, and its data has to reach a control policy fast enough to change the grip. Tesla is the cautionary case: Electrek reported on September 25, 2026, citing The Information, that Optimus touch sensors have reliability issues and that Tesla plans a replaceable “sensing glove” for 2027.

Why are hands the bottleneck for humanoid robots?

Because they are the densest, least durable part of the robot. Tesla’s Optimus hand and forearm contain over 100 screws and small components, and workers still assemble them by hand, according to the same Electrek report; that is the step that limits a production ramp (our Optimus tracker). Wonik Robotics’ backers put hands and actuators at about 70% of humanoid hardware cost, the argument behind Korea’s ₩350 billion investment in a Wonik hand plant.

Durability is the other gap. Tesollo, now preparing an IPO, has said no manufacturer is close to a certified 10,000-hour mean time between repairs for a dexterous hand. Makers are responding in different ways: Tacta rates its motorless mechanism at 30 million cycles and treats fingertips as consumables, while Unitree makes every Dex5-S joint back-drivable with torque protection so impacts do less damage (our Dex5-S report). Hands also compete for the same small motors and reducers as the rest of the robot, a supply issue covered in our report on the actuator gear shortage.

Published durability data is thin, and what exists is modest. ETH Zurich’s ORCA paper reports more than 10,000 continuous operating cycles, about 20 hours, without hardware failure, for a hand built from under CHF 2,000 of materials. Tacta’s 30-million-cycle rating is a company specification, not a customer result. Weight is moving faster: Tesollo’s DG-5F-S comes in at 880 g against 1.76 kg for the DG-5F-M, which matters because every gram at the end of the arm costs payload and battery on a mobile humanoid.

The practical reading for buyers: compare active DoF, not total joints; ask for payload and cycle-life data, which half the hands in the table do not publish; and treat tactile specifications as unproven until a customer reports them after months of use.

Frequently Asked

How much does a dexterous robot hand cost?

Published prices run from $3,500 for the 17-DoF ORCA hand and $6,500 for Unitree’s 22-DoF Dex5-S to about $11,000 listed for Wonik’s Allegro Hand V5, as of September 2026. Shadow, Tesollo, Inspire and Sharpa sell on quotation.

Which robot hand has the most degrees of freedom?

Among hands with published specs, Unitree’s Dex5-S and Sharpa’s SharpaWave each have 22 actuated degrees of freedom, and 1X lists 22 per hand on its NEO humanoid. Tesla engineer Milan Kovac said in November 2024 that the Optimus hand has 22 DoF plus 3 in the wrist and forearm.

What is the difference between active and passive degrees of freedom?

Active degrees of freedom have their own actuator and can be controlled independently; passive ones move by mechanical coupling to another joint. Unitree’s Dex5-1 has 20 DoF, of which 16 are active and 4 passive.

Which robot hand has the best touch sensing?

By published density, Sharpa claims more than 1,000 tactile pixels per fingertip on SharpaWave with 0.005 N sensitivity, and ORCA Touch has 351 taxels per hand. Figure says Figure 03 fingertips detect forces as small as 3 grams.