Soft and underactuated robotic hands: how fewer motors grasp better

From rigid jaws to hands that give way

For decades the standard robot gripper was a pair of rigid jaws driven by one motor. It works well when the object is known in advance and sits in a fixed position, as on an assembly line. It works badly when shapes vary, positions are uncertain or the object is fragile.

The opposite idea was to copy the human hand in full: around twenty joints, each with its own motor or tendon pair, fingertip sensors and a powerful computer to coordinate it all. Such hands showed what is mechanically possible, but they were heavy, expensive and hard to program. A small error in the estimated position of an object was enough for a finger to push it away.

Soft and underactuated hands take a middle path. They accept that the controller does not know everything and let the body of the hand solve part of the problem.

Underactuation: one motor, many joints

In an underactuated hand a single motor, or a small number of motors, drives many joints. A common design uses a tendon that runs through all segments of a finger. When the motor pulls the tendon, the finger starts to close. As soon as one segment touches the object and stops, the remaining segments keep moving until they also make contact. The finger has wrapped itself around the shape without the controller computing any joint angle.

Some designs go further and route one tendon through all fingers, sometimes over pulleys or a differential that splits the force. Then a single command such as “close” produces a whole-hand grasp that adapts to a ball, a bottle or a tool handle. This is also the core idea behind hands built on synergies, explained on our page on hand synergies and how the brain simplifies grasping.

Compliance: why giving way helps

Compliance means that the hand yields when a force acts on it. Engineers build it in with springs in the joints, elastic ligaments, flexible materials or by controlling motors to behave like springs.

Compliance has three big advantages:

• Adaptation. Fingers settle into the shape of the object and spread the contact over a larger area, which needs less force and is gentler on fragile items.
• Robustness. If the hand bumps into a table or the object is not where the camera thought, the fingers bend instead of breaking or knocking it over.
• Safety. A hand that gives way is less dangerous for people working next to the robot.

There is a related research field on robots with variable stiffness. Here the stiffness of a joint can be changed during the motion: soft when contact is likely, stiff when precision or force is needed. Human muscles do something similar when we co-contract them to stiffen the wrist before a hammer blow.

Example designs

Several families of soft hands have emerged in research and industry:

• Tendon-driven hands with elastic joints. Rigid phalanges connected by bands that can stretch, twist and even dislocate safely and snap back. A well-known example is the Pisa/IIT SoftHand, developed in Italy, in which a single motor closes all fingers through one tendon.
• Pneumatic soft grippers. Fingers cast from silicone with internal air chambers that bend when inflated. They are popular for food and parcels.
• Granular jamming grippers. A soft bag filled with granules moulds around an object and then becomes rigid when the air is pumped out.
• Fin-ray style fingers. Plastic fingers with internal ribs that bend toward the object when pressed, inspired by fish fins.

Each design is a different answer to the same question: how much intelligence can be moved from the software into the mechanics?

The limits of soft hands

Soft and underactuated hands are very good at holding things. They are less good at manipulating them inside the hand. Turning a pen between the fingers, typing, or pinching a small screw and rotating it requires independent control of individual fingers. With only one or two motors that is difficult.

Researchers try to bridge this gap by adding a second motor for a different grasp pattern, by locking certain joints, or by using the environment, for instance sliding an object along a table into a better position before lifting it. Another limit is sensing: a hand that adapts mechanically does not automatically know how firmly it is holding something. That is why tactile sensors and force feedback, described in our explainer on haptics and the sense of touch, are often combined with soft designs.

From robots to prostheses

The same arguments that favour soft hands on robots apply to prosthetic hands, only more strongly. A prosthesis has to be light, survive bumps and falls, and be controlled with very few signals from the user’s muscles. A hand that closes around an object on its own reduces the mental effort for the wearer. Our guide to bionic and myoelectric prosthetic hands explains how these ideas show up in devices people wear today.

Summary

Soft and underactuated robot hands replace part of the control problem with clever mechanics. Fewer motors, elastic joints and soft surfaces let the hand adapt to objects it has never seen. The result is robust, affordable grasping, at the cost of some fine dexterity. For an overview of how grasping and touch fit together, go back to the introduction to robot hands and touch.