Three ways to move a finger, and a fourth

Reading Time: 9 minutes

When people look at a robotic or prosthetic hand, they usually ask how many motors it has and where they are. Those are good questions, but they hide the one that decides how a hand behaves: what carries the motion from the motor to the joint.

For us, the question of where the motors go was answered before we started. In a prosthesis the forearm is taken by the residual limb and by the socket that connects the hand to it. There is nowhere else to put the motors: they have to live in the hand, and the hand has to stay light enough to wear all day. So this article is mostly about the second question.

Where the motors sit

Some robotic hands, especially those designed together with a humanoid arm, keep their motors in the forearm and pull the fingers through cables. The fingers stay light and can carry many joints, but the hand is no longer a self-contained part: the forearm has to house the actuators, the cables have to cross the wrist, and hand and arm have to be designed together. When the same team builds both, that is a reasonable trade.

A hand with its motors inside is a module. It can be mounted on different arms, removed and replaced. That is where prosthetics pushed us.

Placement alone, however, does not tell hands apart. Almost all robotic hands sold as standalone products carry their motors inside, and they still behave very differently. What separates them is what sits between the motor and the joint, and there are three classic answers.

Tendons

The human hand is the original tendon-driven hand: most of the muscles that bend the fingers sit in the forearm, and long tendons carry their pull to the joints. Robotic hands that follow this route can have light fingers and many joints, and of the three routes they are the only one that can move the motors off the hand altogether, although not all of them do: some keep their motors in the palm.

A tendon, though, can only pull, so every joint needs a way back: a second tendon, a spring, or a double pulley that lets one actuator wind one tendon while it releases the other. When each tendon has an actuator of its own, controlling every joint takes more actuators than joints, as we saw for the human finger in the previous article: the Stanford/JPL hand used four actuators for each finger with three degrees of freedom (Salisbury and Craig, 1982, International Journal of Robotics Research). The cable also has a life of its own: its lever arm on the joint can change along the motion, the friction along its route changes with the posture of the hand, and cables stretch over time. A little stretch can help, because it adds compliance, but it also means re-tensioning.

Where you want light fingers with many joints, or a hand designed together with its arm, tendons are the right choice.

Tendons: motors in the forearm, and the controller commands every joint.

Joint-level actuation

A hand with joint-level actuation puts a motor at each joint, usually with a gearbox. I call it joint-level actuation rather than direct drive, because in the strict sense direct drive means a motor coupled to the joint with no gearbox at all: the limit case of this family, not the rule. Commanding the motor means commanding the joint, with high control bandwidth, and there is no cable to stretch or re-tension. These hands are precise and reliable, and when you need to command every joint, for instance to study how objects are moved within the hand, they are the natural choice.

The cost is paid in three places. Torque is scarce in a motor the size of a finger. It can be bought with high gear ratios, which make the joint stiff and slow to yield when it meets something unexpected, or the gearing can be kept low enough for the joint to be backdrivable, and then the motor has to stay powered to hold an object. The mass ends up at the far end of the system, in the parts that move fastest. And every motor produces heat where it sits, which in these hands is mostly inside the finger, the part of the hand that touches the object.

Joint-level actuation: a motor at every joint, kept powered to hold the object.

Linkages

The third route is the one I described in the first article of this series: one motor per finger, and a linkage that couples its joints so that they always bend together. The point to add here is a count. However many joints a linkage finger has, it has one degree of freedom, and it closes along a path fixed at design time.

It is also the simplest of the three, repeatable, predictable and robust, which is why most multi-articulating prosthetic hands use it, with one motor per finger. Prosthetics has been asking for more dexterity for decades, and the hands people wear every day have settled on five or six motors and fixed paths: on a hand that has to work all day, on a battery, far from the people who built it, every additional motor has to earn its place. Robotic hands are now going well beyond that count, to twenty motors and more, and I expect them to face the same test.

Linkage: the phalanges bend together, and the fingertip stops short of the object.

A fourth: a way of sharing

Adam’s Hand uses a differential, and a differential is not really a fourth way of carrying torque: it is a way of sharing it. It decides how the motion of one motor is divided among several joints, and it can be built on more than one kind of transmission.

It has been built on tendons for a long time. In 1978 Hirose and Umetani described a gripper made of many segments, driven by a single pair of wires running over pulleys, that wrapped itself around objects (Hirose and Umetani, 1978, Mechanism and Machine Theory). The SDM Hand of Dollar and Howe closes four fingers and eight joints with one actuator, through pulleys that keep the tendon tension equal between the fingers (Dollar and Howe, 2010, International Journal of Robotics Research). Hannes, a prosthetic hand developed at IIT, distributes motion through a differential on tendons (Laffranchi et al., 2020, Science Robotics).

We built ours on gears, for four reasons:

  • The mechanism can be built in modules, so a repair means replacing a module, not rebuilding the finger.


What matters most is that the mechanism imposes no trajectory. The motor decides how far the hand closes; the object decides how that motion is shared.
Each joint stops where contact stops it. With three motors, Adam’s Hand has ten driven degrees of freedom: the two of the thumb have a motor each, and the eight joints of the long fingers share one motor through the differentials, with contact deciding where each of them stops.

The price is paid in control. No joint of the long fingers has a force that can be commanded on its own, and precision grasps at the fingertips, where you would want to choose exactly how hard each finger presses, are where this architecture is weakest.

Adam’s Hand: one motor, and contact decides where each joint stops.

Infographic comparing four robotic hand transmission architectures: tendon-driven, joint-level actuation, linkages, and Adam's Hand differential, with strengths and limitations.
Four ways to bring the motion of a motor to a finger.

More capability is not more usable capability

The dominant direction in robotic hands has been to add: more fingers, more joints, more motors, more sensors. It is easy to see why. A hand that resembles ours should handle our objects more naturally. In the previous article I asked how many of the things we do with our hands actually need twenty joints controlled one by one; here the question is what each of those joints costs.

But more mechanical capability does not automatically become more usable capability. Every degree of freedom with a motor of its own has to be commanded, measured, powered, protected and maintained. In a demonstration, each one is an asset. In a deployment, each one is also something that can drift, wear or fail.

Energy and heat follow the same count. As I described in the first article, hands that stay compliant through their motors have to keep them powered for as long as they hold something, and consumption and heat grow with the number of motors.

So do failures, and for a fleet the arithmetic is simple:

failures per day = robots × hours worked per day × actuators per robot ÷ MTBF, the mean time between failures of one actuator.

Take a fleet of 1,000 humanoids working two shifts, sixteen hours a day. A humanoid typically has seven actuators in each arm, six in each leg and a few in the torso and neck: around thirty before the hands. Just as a reference, give every actuator an MTBF of 10,000 hours:

  • With two hands of twenty actuators each, every robot carries seventy actuators in total, and the fleet should expect 112 failures a day, 64 of them in the hands.
  • With two hands of three motors each, like Adam’s Hand Plus, every robot carries thirty-six, and the fleet should expect about 58 failures a day, fewer than 10 of them in the hands.

A different MTBF changes the totals, not the comparison: same robots, same reliability per actuator, half the failures, and the hands go from more than half of them to one in six.

Put the same numbers in a plant with 1,000 people at work: 112 of them have to stop every day, 64 for the same reason. Any plant manager would go after that reason first. With robots, the reason is the hands, and a failed hand usually has to come off the arm to be fixed.

Fewer actuators are not a free pass, though: when one actuator drives several joints, its failure affects more of the hand, so it has to fail rarely, and gradually rather than all at once.

None of this makes high-dexterity hands wrong. For research on in-hand manipulation, or for tasks that need fingers moving independently, they are the right tool. The question is how much useful manipulation a hand delivers for the complexity it brings with it.

We call this Manipulation Efficiency: the variety of objects, tasks and contacts a hand can manage for a given level of control complexity. It is a design criterion, not a measured quantity, and its simplest indicator is how many driven joints a hand has for each independent command: ten for three in Adam’s Hand Plus.

A higher number is not better in itself, because it also means less independent control of each finger.

After deployment, serviceability

Over the last few years robotic hands have moved from demonstrations to deployment: from showing what a hand can do to showing that it can do it every day. We believe the next step will be serviceability, because we have seen it happen before.

In prosthetics, a hand that cannot be repaired locally has to travel back to its maker, and for several weeks a person is left without it. That is why we designed Adam’s Hand in modules, fingers included, so that the prosthetist can repair the most common failures on site and the user leaves with a working hand.

Robotics is starting to feel the same pressure, and some robotic hands already have fingers that can be replaced in the field. In a lab or a pilot cell, sending a hand back to its manufacturer is an inconvenience. When a hand stops a production line, it is a cost, and the question becomes how the hand is repaired, not only how well it grasps.

The architecture decides how hard that repair is. In a tendon hand the cable route is a wear item, and re-tensioning is part of maintenance; when the motors sit in the forearm, a failure can mean opening both the actuator pack and the routing. In a hand with joint-level actuation the failure stays local, and how easily it is fixed depends on whether the finger was designed to be opened. In a hand built from self-contained modules, the part that fails is the part you replace.

Downtime is the second number to watch, because what a fleet sells is availability, not the absence of failures:

Robots in repair at any moment = failures per day × hours each one takes to fix ÷ 24

Say a failure in the arms, legs or torso takes a shift, eight hours: that alone keeps 16 robots out of 1,000 in repair at any moment, whatever hands they carry. A hand failure is quicker to fix only if there is a spare hand to swap in, and even then taking a hand off and mounting another takes at least an hour, while the hand that came off still has to be repaired.

With twenty-actuator hands that is 64 swaps a day, eight technicians doing nothing else, and with two days on the bench some 128 spare hands in rotation.

With Adam’s Hand Plus (three motors) it is about 6 swaps a day and 13 spare hands, because two of the six motors on such a robot, the ones that flex the thumbs, sit inside the thumb modules: when one fails, the thumb is replaced on the robot in minutes, like any other finger module. With spares, both fleets stay close to 98 percent availability: the difference is paid in technicians and spare hands.

Without spares, a failed hand waits at least two days for parts: at any moment 128 robots stand still because of their hands, against 13, and the fleet’s availability falls to about 86 per cent, against 97. In prosthetics that difference is measured in weeks without a hand; on a line, in robots standing still, multiplied by the size of the fleet.

1,000 humanoids over a working day: failures, then robots standing still without spares. Illustrative scenario, assumptions on screen.

Between the motor and the object

In each of the four architectures the motor sits in a different place, but that is not what sets them apart. What sets them apart is what happens between the motor and the object. For us, that is where Mechanical Intelligence lives, and after three years of Adam’s Hand on the market it is still the choice I would make.

The same finger, the same object, four ways to bring the motion of a motor to it.

  • All
  • News
  • People
  • Tech

Three ways to move a finger, and a fourth

Twenty-seven degrees of freedom, and how many a hand actually uses

One motor, four fingers: the mechanism behind Adam’s Hand

US Reimbursement for Upper Limb Prosthetics: HCPCS Coding for BionIT Labs Solutions

Evolving Our Identity: Engineering Human Potential

BionIT Labs at OTWorld 2026: Introducing Adam’s Hand Plus, PRISM and PULSE

BionIT Labs Opens U.S. Headquarters in Chattanooga: Expanding Access to Adam’s Hand Across North America

Disability Pride Month: Building a More Inclusive Workplace with Abile Job

How to Navigate the Rehabilitation Journey After an Upper Limb Amputation: Expert Guide by Diane Atkins, OTR – Amputee Clinical Specialist

New Small-Size Adam’s Hand: Smaller. Faster. Even More Durable.

Amputation and Rehabilitation: Everything You Need to Know Before and After Surgery, with Prof. Alexander Gardetto

Adam’s Hand on Tour: New Demos and Trials in Italy and Europe!

BionIT Labs Secures Top Spot at Cybathlon 2024 with Adam’s Hand!

Just one control method for infinite possibilities with Adam’s Hand!

Adam’s Hand got PDAC approval!