People often ask me why Adam’s Hand drives its four long fingers with a single motor, when most multi-articulating prosthetic hands put a motor in every finger. The short answer is that a person wearing a prosthesis cannot steer four fingers one by one. The longer answer involves a gear mechanism that most of us already own without thinking about it, because it sits in the axle of every car.
I want to use this article to explain that mechanism properly, step by step, because it is the idea the whole hand is built on.
Four accelerator pedals
A prosthetic hand is controlled through a very narrow channel. In a typical fitting, two electrodes on the forearm read the electrical activity of the residual muscles. From those two signals the user has to open, close and choose how to grasp. They can express an intention, but they cannot command each joint.
Now picture a hand with one motor per finger. Every finger has to be told what to do, and when the hand meets an object someone has to decide which finger stops, which one keeps pushing and how hard. It is a bit like driving a car with four accelerator pedals, one for each wheel. You could do it in theory, but nobody would call it driving.
With Adam’s Hand there is one pedal. The user asks the hand to close, and the mechanism decides how that motion is shared among the fingers.
What a differential does
When a car takes a corner, the inner wheel travels a shorter path than the outer wheel. If both wheels were locked to the same shaft, one of them would have to slip. The differential solves this: it takes the motion coming from the engine and delivers it to two outputs that are free to turn at different speeds.
Inside it there is a carrier, driven by the engine, and on the carrier two small bevel gears called satellites. The satellites always turn with the carrier, orbiting around the axle, and that is how they push both wheels forward. On a straight road that is all they do, and the whole assembly turns as one block. In a corner they also start spinning on their own pin, taking a little speed from the inner wheel and giving it to the outer one. The total motion is the same; only the split changes.
That is the property we needed for a hand: one input, several outputs, and the freedom for each output to go its own way when something holds it back.
From four wheels to four fingers
A four-wheel-drive car uses three differentials: a central one that splits the engine’s motion between the front and rear axle, and one on each axle that splits it again between the two wheels. One motor, four outputs.
Adam’s Hand uses the same tree. A motor in the palm drives a first differential stage. Its two outputs drive two more differentials, and each of those drives two fingers. When the hand closes around an object, the first finger that touches it stops. Because of the differentials, the motion that finger can no longer use goes to the others, which keep closing until they touch as well. Each finger settles where the object stops it.
No controller is computing the shape of the object. The grasp takes that shape because of how the gears are connected. This is what we call inter-finger underactuation: fewer motors than joints, with the difference resolved by the mechanism.
Inside each finger
Adapting across the fingers is only half of it. Each long finger has two phalanges and two degrees of freedom, and inside the proximal phalanx there is a train of idle gears. The last gear of the train is fixed to the distal phalanx.
While the finger closes freely, both phalanges move together. When the proximal phalanx touches the object and stops, the gear train keeps transmitting motion, so the distal phalanx keeps rotating and wraps around the object. The finger ends up enveloping the shape instead of touching it at one point.
There are five gears in our proximal phalanx, and the count is not arbitrary: it has to be odd. With an even number, the distal phalanx would turn the opposite way and open instead of closing. Beyond that rule the gear train is easy to adapt, which is what makes the finger itself modular.
Many multi-articulating hands use a linkage in each finger instead. The two phalanges are coupled and always bend together along a path fixed at design time, whatever the object. When the base of the finger touches, the whole finger stops, and the fingertip often stays away from the surface. Linkages are compact, strong and repeatable, and where the objects are known that is a sound choice. It is simply not the choice you want when every object is different.
Putting the two levels together gives the hand one more property, which is easy to miss. A differential does more than share out motion: it also keeps the torque equal on its two outputs. When a finger touches the object, it starts to meet resistance. The differential cannot load that finger any harder than the other one, so before the grip can tighten, the same resistance has to be reached on the other side as well. The other finger is still free, and the only way it can get there is to keep moving, and to move faster: the carrier always turns at the average speed of its two outputs, so when one output stops, the other has to turn twice as fast as the carrier. The free fingers hurry onto the object, and the force builds up only once they are all in contact. The gear train inside each finger does the same between its two phalanges.
This does not make the hand any weaker. The grip can be just as firm, and the pressure at each contact can still be high. What changes is how that force is managed: no finger squeezes harder while the others are still on their way, and no contact is overloaded while another carries almost nothing. The force at each contact still depends on where it sits along the phalanx, but the load is balanced across the whole grasp. That is what makes the difference with a ripe fruit or a paper cup, and it also makes the grip more stable. As with everything else here, no sensor and no controller are involved. It is the equilibrium of the gears.
Holding without spending energy
There is one more element between the motor and the differentials: a worm gear. A worm gear can drive the mechanism, but the mechanism cannot turn the worm back. So once the hand has closed, the motor stops completely. The grasp is held by the transmission itself and draws no current.
This came straight from prosthetics. A user cannot keep a muscle contracted for minutes just to hold a glass, and a battery that must last the whole day cannot keep a motor powered for the same reason. Holding had to be free.
The important detail is where the worm gear sits. Because it is placed before the differentials, the part of the mechanism that adapts is still free to move after the motor has stopped. With the input held still, the fingers can only move by compensating each other: if one opens, the others have to close by the same amount. So when a held object is nudged or twisted in the hand, the fingers on one side give way, those on the other side close in, and the whole grasp resettles around the object without letting it go. It also means a single finger cannot be moved on its own when all the fingers are fully closed, which helps the hand keep a firm hold on the object.
Holding and adapting are separated: holding costs nothing, and adapting is done by the gears. While the hand is closing, the adaptation is driven by the motor. Once the motor has stopped, it becomes passive.
Many robotic hands take the opposite route. To be compliant they keep their motors engaged and use impedance control: sensors measure the interaction and the controller adjusts the motor torque to make the finger behave like a spring. It works, but the motors stay on for as long as the hand holds something, which means continuous power and heat, and more of both as the number of motors grows. Active control also has a limited bandwidth, and against stiff surfaces it can become unstable, which is why the gains are usually set conservatively in real deployments. A mechanical compliance reacts as fast as the parts can move and stays stable on any surface, because there is no loop to go unstable.
The same passive compliance also makes the hand more robust. When a finger is hit, it can yield, and the mechanism shares the load with the other fingers, instead of the shock being taken by a rigid transmission. That is a big part of why Adam’s Hand is so durable.
Three motors, ten degrees of freedom
The whole hand uses three motors. One, in the palm, drives the four long fingers through the differentials, which gives eight driven degrees of freedom. The other two move the thumb. The thumb has a single phalanx, and one motor sits inside it to drive flexion and extension. The other sits in the palm and rotates the thumb. The rotation is continuous, but in everyday prosthetic use it is offered as a set of selectable positions, because stopping a rotation at a precise angle through two electrodes is hard. Three positions serve as references: lateral, for key grip; facing the tip of the index finger, for precision grip; and in opposition, for tripod grip.
That split is deliberate. The thumb carries most of the function in precision grasps, and opposition is also the hardest movement to actuate. That is why two thirds of the actuation goes to the thumb, and one motor, with the help of the mechanism, handles the rest. The result is ten degrees of freedom driven by three motors.
Beyond prosthetics
The two problems prosthetics forced us to solve first are the same two problems robotics faces as robots leave controlled factory cells: getting rich manipulation out of a limited control interface, and surviving uncontrolled contact with the real world. A robot arm usually gives its gripper a single open or close command. That is very close to what a prosthesis receives from its user.
Because the adaptation happens in the gears, it is identical whether the command comes from a person or from a robot controller. And the architecture is not limited to the human form factor: the same compliant concept can be applied to end-effectors with two, three, four or more fingers, each with two or more phalanges.
More than 250 people use Adam’s Hand every day. The mechanism described here is what they rely on each time they pick something up without thinking about which finger should stop first.
Mechanical Intelligence
At BionIT Labs we have a name for this: Mechanical Intelligence. We don’t mean a mechanism that thinks. We mean a physical architecture that solves part of the control problem through its own behavior. In Adam’s Hand it does that in two complementary ways:
- Actively, while the hand closes: one motor sets the intent, and the differentials share motion and force across the fingers and the phalanges until the grip fits the object;
- Passively, once the hand is holding: with the motor off, the fingers compensate each other, absorbing impacts and redistributing loads without any additional command.
If I had to sum up the design in one rule, it would be this: before programming adaptation, engineer it into the mechanics. Software is still needed, and it matters. But every contact the gears can resolve on their own is one less thing the user, or the controller, has to worry about.
As a mechanical engineer, I have always been fascinated by the differential. Seeing our team carry that idea into a bionic hand is one of the achievements I’m most proud of.
