# Dexterous Robot Hand Anatomy

By greenlig

Topic: Machines

[Open the interactive Jig](https://ask.jig.com/j/qdhNGDEE89HN)

## Guide

### 1. An Anthropomorphic, Tendon\-Driven Hand

This dexterous hand copies the human layout: four fingers and an opposable thumb on an adult\-sized palm\. Forearm actuators, antagonistic tendons and a sensor array work together\. Drag the background to inspect it from any side\.

The proportions follow research hands such as the Shadow Dexterous Hand, which has 24 joints and 20 actuated degrees of freedom\.

### 2. Each Finger Is a Serial Kinematic Chain

Each finger is three rigid links \(proximal, middle and distal phalanx\) joined by revolute joints, each rotating about a single axis\. Watch the index finger flex from the metacarpophalangeal \(MCP\) joint distally\. The MCP joint carries a second axis for abduction and adduction, giving it two degrees of freedom\.

The distal two joints are coupled, as in the human DIP–PIP linkage: one actuator drives both, so each finger has 4 joints but 3 independent degrees of freedom\.

### 3. Remote Actuation in the Forearm

The fingers contain almost no actuators\. As with the extrinsic flexors of the human hand, all 20 motors sit in the forearm, which keeps distal mass and inertia low\. Each motor turns a spool that winds in or pays out a tendon cable, converting torque τ to tension T = τ / r for spool radius r\.

Remote actuation trades compactness for cable friction, stretch and routing compliance that the controller must compensate\.

### 4. Antagonistic Tendon Pairs

Tendons run from the spools through the palm to each phalanx\. Tension in the palmar \(orange\) flexor closes the joint; tension in the dorsal \(blue\) extensor opens it\. Joint torque equals tension times the tendon's moment arm, τ = T · r\_j\. Use the slider to load the index flexor\.

A cable transmits tension only, never compression, so each degree of freedom needs at least two opposing tendons \(an n\+1 or 2n arrangement\)\.

### 5. Thumb Opposition

The thumb has five joints, more than any finger, including the carpometacarpal joint that rotates it in front of the palm\. Opposition brings the thumb pad against a fingertip, forming a precision pinch with two opposing contact normals\. This is the grasp that lifts the small cube\.

Precision pinch relies on fingertip contact friction; the cube is held stable when both contact forces lie inside their friction cones\.

### 6. Closed\-Loop Tactile Sensing

During a grasp, fingertip tactile sensors report contact and joint\-angle sensors report each joint's position\. The controller samples this feedback at up to 1 kHz and limits grip force once contact is detected\. The result is a stable grasp that does not crush a compliant object\.

The Shadow Dexterous Hand carries more than 100 sensors, each sampled at up to 1 kHz\.

### 7. Grasp Synthesis

Tap any finger to flex or extend it, or choose a grasp below to see how the joints coordinate\. Joint speed is limited to roughly 120° per second, so closing a full power grasp takes about one second\. Compare the power grasp, which wraps the ball with the palm, with the fingertip precision pinch\.

### 8. The Open Problem Is Control

Joints, tendons and sensors are necessary but not sufficient; the difficult part is learning control policies for a high\-dimensional, contact\-rich system\. Researchers train policies in simulation with domain randomisation and transfer them to the physical hand\. Using this approach, one team taught a dexterous hand to solve a Rubik's Cube one\-handed\.

The human hand has 27 bones and roughly 17,000 tactile mechanoreceptors in the glabrous skin; robotic hands remain far from this density\.