Ask five suppliers for a dexterous hand and you get five different answers about what matters — degrees of freedom, fingertip force, tactile resolution. Almost nobody opens with the thing that decides all of them: dexterous hand actuation types. Whether the fingers move on linkages, on tendons, or on motors sitting in the joints themselves sets your DOF ceiling, your control interface, your maintenance interval and how much of the hand you can repair in the field.
This is the comparison we work through before anyone commits to a sample. It is worth reading next to the Linker Hand range, where all three architectures sit in one catalogue and the differences show up directly in the model lineup.
What actuation decides, and what it does not
Actuation is not a marketing label. It fixes four things you cannot recover later in software.
- Where the mass sits. Motors in the palm give you a compact hand with the load close to the wrist. Motors in the fingers push mass outward, and the elbow and shoulder pay for every gram in torque and speed.
- How many axes are independent. A linkage couples several joints to one actuator. Often that is exactly what you want — but it means quoted joint count and controllable DOF are two different numbers.
- How transparently you feel contact. Backdrivability depends on everything between the motor and the fingertip: gearing, cable routing, friction.
- What fails first, and what the spare costs. Every architecture has one part that wears out ahead of the rest, and that part determines your maintenance contract.
The three architectures
Linkage-driven, including worm-gear hybrids
A motor drives a train of bars and pivots that close the finger. One actuator flexes several joints in sequence, so the finger wraps around whatever it touches without needing to sense anything. That is underactuation, and it turns a single control channel into an adaptive grip.
Worm-gear variants add self-locking: the finger holds position with the motor off, which is how a value-oriented hand can quote a large payload on a modest power budget. The trade is backlash accumulating over time and poor backdrivability.
Tendon-driven
Motors sit in the palm or forearm and pull cables routed through the fingers, close to human anatomy. You get the best dexterity per unit volume available: many joints inside a human-sized envelope, with the heavy parts kept proximal where the arm can carry them.
The costs are honest ones. Cables need tension maintenance, stretch changes joint calibration as the hand ages, and a snapped tendon puts that finger out of service until it is re-strung. Tendon hands are also where fingertip tactile arrays are usually offered, because nothing else is competing for fingertip volume.
Direct drive, or motor per joint
A small motor with a miniature gearbox sits in or immediately behind each joint. Every axis is independent and directly commandable, and joint torque can be estimated from motor current without adding another sensor.
Packaging is the limit. One motor per axis makes the hand larger, lower-DOF or expensive, so in practice direct drive shows up on the one or two axes that need true force transparency rather than across all twenty.
Head to head
| Linkage / worm-gear hybrid | Tendon-driven | Direct drive per joint | |
|---|---|---|---|
| DOF per actuator | Several — the coupling is the feature | One cable per axis; joints share the volume well | One |
| Independent control | Limited by the linkage design | High | Full |
| Adaptive grip without sensing | Excellent | Partial | None — you close the loop yourself |
| Backdrivability | Low; a worm gear self-locks | Moderate, depends on routing and friction | Highest |
| Mass location | Palm, proximal | Forearm or palm | In the finger |
| First thing to wear | Gear and pivot backlash | Cable tension, cable stretch | Gearbox and motor heat |
| Field repair | Swap a finger module | Re-string the cable | Replace a joint unit |
| Right answer when | Utility grasping, high payload per gram | High DOF in a human-sized envelope | Precision pinch and force control |
Reading the DOF number honestly
Vendors quote degrees of freedom two ways, and the gap between them is where most disappointment comes from:
- Total joints — every hinge in the hand, whether or not anything commands it individually.
- Active DOF — axes that have their own controllable actuator.
A hand whose passive DOF roughly equals its active DOF is buying you mechanical adaptation, not twenty commandable channels. For irregular objects that often outperforms twenty motors. If your task needs in-hand reorientation, controlled squeeze rate or individual fingertip placement, it is the active count you have to buy — and you should ask a supplier to confirm which number they are quoting before you compare two datasheets.
Choosing by task
| Task | What actually matters | Architecture that fits |
|---|---|---|
| Pick and place across mixed objects | Consistent wrap, low cost per grasp, repeatability over cycles | Linkage, usually with passive DOF doing the adaptation |
| Handheld tool use | Holding force while the motor is warm | Worm-gear hybrid — self-locking does part of the work |
| In-hand manipulation | Independent axes plus compliance | Tendon-driven, or direct drive on the critical axes |
| Teleoperation and imitation data | Human-hand fidelity plus fingertip sensing | Tendon-driven with tactile, evaluated alongside a motion capture glove |
| Unknown or fragile objects | Detecting contact before damage | Any hand with fingertip tactile; direct drive where you estimate torque from current |
On that fourth row: if you are capturing demonstration data, put glove latency and joint mapping into the same evaluation window as the hand. Discovering a mapping problem after the hand ships costs far more than testing them together.
Integration checks before you commit
| Check | Why it bites late |
|---|---|
| Bus and frame rate | A high-DOF hand reporting dozens of channels at several hundred hertz can crowd out the rest of the robot on a classic CAN bus. Ask for payload size and update rate before you put the hand and the arm on one segment. |
| Wrist payload budget | The hand typically consumes roughly 10–20% of available wrist payload as a rule of thumb — treat that as approximate and work backwards from the arm rather than forwards from the grip. |
| Calibration drift | Any compliant transmission moves over time. Ask whether joint offsets get re-zeroed in software or mechanically, and how often. |
| Wear parts and lead time | Ask what wears, what the spare costs and how long it takes. Swapping a finger module and re-stringing a cable are very different maintenance commitments. |
| Software before hardware | Confirm the SDK, ROS packages and protocol support before comparing anything else. Our seven-day sample evaluation protocol covers what to test in week one. |
Rules of thumb
- Uncertain task, objects vary: start with linkage. Cheapest per grasp and it forgives poor sensing.
- DOF density or fingertip sensing is the requirement: tendon-driven, with a maintenance plan agreed up front.
- One or two axes need force transparency: direct drive there, not everywhere.
- Always ask for active DOF, not total joints, and get the answer in writing.
- Buy the sample and the production units on the same enquiry so interface, firmware and spares match when you scale.
Bring us the grasp, not the datasheet
Actuation is cheaper to get right before the sample arrives than after. Tell us the grasp types, payload and control interface, and we will shortlist across every factory we buy from — start from the Linker Hand range or the Linkerbot overview, then move to a quote.
Ask us to spec a hand for your arm
Specs quoted from manufacturer datasheets; final figures confirmed at quotation.