Why the thumb is the hardest digit to copy

Each of the four fingers is, mechanically, a planar chain: three joints that bend in the same plane plus a small side movement at the knuckle. The thumb is built differently. Its base, the carpometacarpal (CMC) joint, is a saddle joint with two axes, so the whole digit can swing away from the palm and rotate until its pad faces the fingers. Two more joints, the MCP and IP, then flex it. A review comparing human and robot hands counts four thumb degrees of freedom, controlled by eight muscles.

That base joint is what makes opposition possible, and opposition is what makes the hand useful. Hand surgeons put the thumb at roughly 40% of hand function, which is why losing it is treated as a major injury. For a humanoid the consequence is the same: a hand without a capable thumb can hook, push and wrap, but it cannot pinch a screw, hold a pen or turn a key.

Two to five actuators: how hands budget the thumb

Every robot hand has to decide how many of those human axes to copy, and how many get their own actuator. The answers cluster. Low-cost commercial hands with six actuators, such as the Inspire RH56DFX, sold as an option for Unitree humanoids, and the PSYONIC Ability Hand on Apptronik’s Apollo, give one actuator to each finger and two to the thumb: one to rotate it across the palm, one to flex it. Unitree’s three-finger Dex3-1 gives its thumb three direct-drive joints out of seven.

High-DoF hands spend more. The Wuji Hand puts four independently controlled joints in every digit, thumb included, with the motors inside the fingers. The Shadow Dexterous Hand E, a long-standing research reference, gives the thumb five joints with its own motor for each; the only coupled joints in that hand are in the fingers. The latest Optimus hand is reported to have 22 degrees of freedom driven by tendons from forearm actuators, but Tesla has not published how many of them are in the thumb.

Coupling: one motor, several joints

The two-actuator thumb works because joints can share a motor. In six-actuator linkage hands such as the RH56DFX, the flexion drive bends more than one thumb joint through a linkage, so the thumb curls in a fixed pattern instead of each joint moving freely. The SyLink research hand takes the idea further and publishes how: one motor at the thumb base flexes the CMC, MCP and IP joints together through two stacked crossed four-bar linkages, and a separate servo in the palm drives the side axis. Two actuators move four joints.

Coupling buys space, mass and cost, and it matches how people usually curl the thumb in a power grasp. The price is that some poses become unreachable. A coupled thumb cannot keep its tip joint straight while bending the base, which matters for flat pinches and for pushing buttons with the pad. The researchers behind an earlier two-axis thumb design found a related limit: a thumb–index grasp failed because the two fingers involved shared an actuator and could not flex differently.

Range of motion: robots can out-bend the human thumb

Range is the other half of the design. The DexLink hand, a 16-DoF linkage-driven research hand published in 2026, lists its thumb ranges next to human reference values. Its thumb exceeds the human figures on three of four axes and falls short only at the tip joint, adding up to 365° of total joint range against 302.4°. Unitree publishes its Dex3-1 thumb joints at −60 to 60°, −35 to 60° and 0 to 100°. Shadow’s base rotation spans 120°.

Larger range does not automatically mean better opposition, because what counts is where the thumb tip ends up relative to the other fingertips. A 2026 kinematic study of palm design found that adding two palm joints, which bend the outer fingers toward the thumb, raised the workspace the thumb shares with the little finger by more than 1,000% compared with a rigid palm. Part of a good thumb, in other words, is in the palm.

Measuring opposition: the Kapandji test

Because DoF counts and ranges do not say whether a thumb works, robot-hand researchers have borrowed a clinical test. The Kapandji score, used by hand therapists, grades opposition from 1 to 10 by the furthest point on the hand the thumb tip can touch: the side of the index finger, then each fingertip in turn, then the creases of the little finger and finally the palm. DexLink reports the maximum score; SyLink reports the highest score in its own test without giving a number.

For buyers, the test is a useful shorthand. A hand that scores 6 can oppose every fingertip; one that cannot reach the ring and little fingers is effectively a three-finger gripper with extra fingers for wrapping. Commercial suppliers rarely publish the score, so it is worth asking for a demonstration.

What to check on a hand spec sheet

• Thumb actuators, not just hand DoF. A 6-DoF hand with a 2-actuator thumb and a 20-DoF hand with a 5-actuator thumb are different machines. Ask how many of the thumb’s axes have their own drive and which are coupled.
• Rotation range at the base. The side axis decides whether the thumb pad can face the fingers at all. The RH56DFX lists 65°; Shadow’s base rotation spans 120°.
• Thumb force versus finger force. In a pinch the thumb balances the fingers. Check that the thumb is rated at least as high as the fingers it opposes, as in the RH56DFX’s 15 N against 10 N.
• A Kapandji score or a reach test. Research hands now report it. A commercial supplier should be able to show which fingertips and palm points the thumb tip reaches with real contact force.

FAQ

Sources