Robot hands range from orderable research end effectors to dexterous systems available only as part of a humanoid robot. This comparison keeps those buying formats separate and shows source-verified degrees of freedom, actuation, tactile or force sensing, weight, grip metrics, pricing and verification dates. Use the filters to find hands you can buy separately, integrated hands and open research platforms without treating incomplete manufacturer claims as equivalent data.
18 robot hands compared · 14 with documented tactile or force sensing · 14 with a standalone buying route
Compare manufacturer-published figures without collapsing actuated and passive joints into one number. Grip values keep their original metric, because fingertip force, holding force and payload are not interchangeable.
Showing 18 of 18 hands
Figure AI
1X Technologies
Inspire Robotics
Seed Robotics
ORCA Dexterity, Inc.
Carnegie Mellon University
Wonik Robotics
Wonik Robotics
qb robotics
SCHUNK
Tesollo
DexRobot
Shadow Robot
Shadow Robot
Unitree Robotics
Unitree Robotics
mimic robotics
A parallel or adaptive gripper is usually the better production tool when the object set, grasp direction and cycle are predictable. A multi-finger robot hand earns its extra cost and control burden when the job needs in-hand reorientation, tool use, varied object geometry or human-compatible contact. Define the required grasps and failure tolerance before using finger count or human likeness as a shortcut for capability.
Manufacturers count degrees of freedom in different ways. Total DoF can include coupled, passive or wrist joints; actuated DoF counts independently driven motion. A hand with more stated joints is not necessarily more controllable. RoboZaps keeps actuated and total DoF in separate fields, preserves the maker's counting basis on the product record and marks the value unpublished when a source does not make the distinction.
Tendon and cable drives can move motors out of the fingers and support compliant, human-like packaging, but cables stretch and may need tensioning. Direct-drive and geared joints can simplify control and deliver stiffness, while adding distal mass and potential backlash. Pneumatic and soft actuators favour safe contact and compliance but need different infrastructure. Compare actuation architecture with service access, replacement parts and expected duty cycle—not as a standalone ranking.
A contact switch, a motor-current estimate, normal-force taxels and three-axis fingertip sensing provide very different feedback. Ask which surfaces are instrumented, whether the signal measures contact, normal force or shear, the spatial resolution and sampling rate, and whether raw data is exposed through the SDK. The table reports only the sensing mode a source actually documents; a tactile marketing claim without a public measurement stays qualified on the product page.
Vendors publish fingertip force, pinch force, holding force, cylindrical grasp capacity or payload—sometimes as a peak and sometimes as a continuous rating. Those are different tests. RoboZaps preserves the original metric and unit instead of converting them into a misleading strongest-hand leaderboard. For a real shortlist, request the test pose, object geometry, friction material, duration and safety factor behind the number.
Mounting interface, side and wrist orientation, hand mass, centre of gravity, cable routing, power, communications and controller frequency all affect whether a hand works on an arm. Then inspect the software layer: ROS support, low-level command access, simulation models, calibration tooling, example grasps and licence terms. A nominally compatible flange does not make the controls, payload budget or safety case compatible.
Some robot hands can be ordered as complete hardware, some are quote-only research products, some publish open designs that still require assembly, and many headline-grabbing hands are not sold separately from their humanoid. The buying-format field makes that boundary explicit. A whole-robot announcement, preorder or price is never presented as the price or availability of its hand.
Ask for the exact left/right configuration, controller and sensor package included in the quote; continuous and peak force limits; expected cable, skin or fingertip life; replacement procedures; supported communication stack; calibration requirements; warranty exclusions; lead time; export restrictions; and a task-relevant demo using representative objects. Budget for mounting, power, controls integration, spare wear parts and engineering time as well as the hand.
This directory is built as an evidence trail, not a one-number ranking. Product pages retain the source and checked date behind each public specification so readers can audit a claim in context.
Figure AI
The redesigned hand system of Figure's third-generation humanoid, Figure 03, purpose-built for the Helix vision-language-action AI. It ships only as part of the Figure 03 robot and is not sold separately.
1X Technologies
The hands of 1X's NEO humanoid: 25 degrees of freedom (22 in the fingers and palm plus 3 at the wrist), tendon-driven and force-controlled, with tactile sensing across fingertips and surfaces. They ship only as part of NEO.
PSYONIC
PSYONIC's Ability Hand is a 490 g multi-articulated bionic hand, originally a prosthesis, that PSYONIC also sells to robotics research users with full API access.
Inspire Robotics
The RH56DFX is Inspire Robots' humanoid five-finger dexterous hand with 6 degrees of freedom across 12 joints, driven by force-controlled micro linear servo actuators. It integrates force sensing with 0.5 N resolution and is aimed at robot end effectors and prosthetics.
Seed Robotics
The RH8D is Seed Robotics' adult-size dexterous robot hand with 19 degrees of freedom driven by 8 smart actuators contained inside the unit. Its underactuated design conforms to object shapes, and it offers rich per-joint sensing with optional FTS tactile pressure sensors.
ORCA Dexterity, Inc.
The ORCA Hand is an open-source, tendon-driven anthropomorphic robotic hand that originated in the Soft Robotics Lab at ETH Zurich and is sold by ORCA Dexterity, Inc. Fully assembled hands ship with Feetech or Dynamixel actuators, and the Touch variant adds 351 tactile taxels across all five digits.
From $3,500Listed price
Carnegie Mellon University
LEAP Hand is a low-cost, anthropomorphic four-finger robot hand for machine-learning research, presented at RSS 2023 by Carnegie Mellon University researchers. It is fully open source, with CAD, URDF, simulation environments, and APIs released, and assembles from off-the-shelf parts for under $2,000.
Wonik Robotics
The Allegro Hand V5 Sense is a four-finger (three fingers plus thumb), 16-DoF research hand equipped with 16 pressure-sensing channels across fingertip and joint sensors for detailed force-distribution feedback.
Wonik Robotics
The Allegro Hand V5 is a three-finger dexterous research hand with a 9-DoF structure of independent current-controlled joints, equipped with 360° omnidirectional pneumatic tactile fingertip sensors.
qb robotics
The qb SoftHand2 Research is an anthropomorphic soft robotic hand with 19 degrees of freedom driven by two motors through two synergies, capable of both power and precision grasps and in-hand manipulation.
SCHUNK
The SCHUNK SVH is a series-production anthropomorphic five-finger gripping hand with nine drives moving 20 degrees of freedom, with all control, regulator and power electronics integrated in the wrist.
Tesollo
A fully actuated five-finger robot hand similar in size to an adult male's hand, composed of 20 independently controlled BLDC-driven joints, rated for 2.5 kg pinch and 10 kg envelop payloads.
DexRobot
A commercial five-finger dexterous hand with 19 DoF (12 active, 7 passive), tendon-driven transmission, five tactile sensing modalities, and active thermal management for 24/7 operation.
Shadow Robot
A three-finger, 12-DoF dexterous hand developed in collaboration with Google DeepMind for high-cycle machine learning research, with tendon transmission, active compliance, and dense tactile sensing.
Shadow Robot
A five-finger, tendon-driven robot hand with 20 DC motors, 20 actuated DOF plus 4 under-actuated movements across 24 joints, and over 100 sensors running at up to 1 kHz.
Unitree Robotics
The tactile-sensing variant of Unitree's five-finger Dex5-1 dexterous hand, with 20 degrees of freedom (16 active) and 94 pressure sensors per hand.
Unitree Robotics
A three-finger dexterous hand with seven motor-driven degrees of freedom and 33 pressure sensors per hand, designed to pair with Unitree's G1 humanoid robot.
mimic robotics
Swiss-made dexterous robotic hand with 21 total DoF, tactile sensing, and a >25 kg steady-state power-grasp payload.
A robot gripper is an end effector designed to secure objects, often with two jaws, suction or a task-specific mechanism. A robot hand uses multiple articulated digits to support a wider grasp set and sometimes in-hand manipulation. Hands offer flexibility at the cost of more actuators, sensing, control complexity and maintenance; predictable production tasks often remain better served by a gripper.
Some dexterous hands are sold as standalone research or integration products, while others are available only through a quote, an open hardware build or as part of a complete humanoid. Use the Buying format filter and confirm what the quote includes. RoboZaps does not treat a whole robot's sales route as proof that its hand is sold separately.
Public pricing is inconsistent: a few makers and open-hardware projects publish a component price, many commercial research hands require a quote, and integrated humanoid hands have no separate price. The table shows a price only when it applies to the hand and has an auditable basis; otherwise it says to contact the maker rather than borrowing the host robot's price.
There is no universal minimum. Simple grasp sets can work with a few coupled joints, while in-hand manipulation benefits from more independently actuated motion. Compare actuated DoF, coupling, joint range and control access against the target grasps. Total DoF alone can overstate controllability when several joints move together.
Not automatically. Five fingers can expand anthropomorphic grasp and tool-use options, but a well-designed three-finger hand may be lighter, stronger, easier to control and more reliable for the target objects. Task coverage, independent actuation, sensing, friction surfaces and control software matter more than matching the human finger count.
Match the sensing to the control problem. Contact detection may be enough for basic grasp closure; force regulation needs calibrated normal-force data; slip-aware manipulation benefits from shear or vibration signals and adequate sampling. Check where sensors are located, their resolution, refresh rate, durability and whether the SDK exposes raw or processed readings.
They can be excellent for research, controls development and custom integration, but an open design is not the same as a supported production component. Review the hardware and software licence, bill-of-material continuity, assembly calibration, safety validation, replacement parts and who owns reliability. Commercial deployment may still require substantial engineering and a support plan.
No public benchmark supports a defensible universal winner. Dexterity depends on the task, joint independence and range, sensing, force control, speed, software and reliability—not the largest DoF number. Shortlist hands against a representative object and manipulation test, then compare sourced specs and access constraints before arranging a demonstration.
The Robot Sourcing Brief is US$1,950 fixed: a scored shortlist of three platforms with sources, US legality per model, and real deployment costs. Credited in full if you procure through RoboZaps.
Not ready to spend anything? Start with the free US deployability report.