humanoidsdata.com

Search

Search datasets, articles, and glossary terms for humanoids and embodied AI.

← All glossary terms

Hardware & control

Dexterous manipulation

Dexterous manipulation is the controlled, skillful reconfiguration of an object through coordinated motion and contact, often using multiple fingers. It can involve changing an object's pose within a hand, regrasping, sliding, rolling, finger gaiting, or making precise contact with the environment. The term describes capability, not a fixed minimum number of fingers or joints.

Also known as: robot dexterous manipulation, dexterous robotic manipulation

Updated

Dexterity is more than holding an object

A stable grasp keeps an object from moving unintentionally. Dexterous manipulation deliberately changes the object or its contact configuration. Bicchi's influential robot-hand survey describes dexterity in terms of moving a manipulated object from one position and orientation to another chosen configuration within the hand workspace.

That definition covers in-hand rotation, but current usage is often broader. Regrasping, finger gaiting, sliding, rolling, tool use and precise insertion can all be called dexterous when they require controlled contact changes. In-hand manipulation is therefore an important subset, not a synonym for every dexterous task.

A dexterous hand does not guarantee dexterous behaviour

Many actuated joints can increase the available motions, but hardware complexity alone does not establish dexterity. Actuation, sensing, controllability, compliance, friction, calibration and the policy all affect what the system can perform reliably.

The field has no universal degree-of-freedom threshold. Recent survey work compares hands under different embodiments, sensory configurations and evaluation protocols, which makes the claimed capability and test task more informative than the label “dexterous hand”.

Contact-rich data carries the missing detail

Vision can record an object's gross motion while hiding fingertip pressure, incipient slip and changes in contact mode. Dexterous datasets benefit from synchronised hand state, commanded and measured motion, tactile or force signals, object pose, contact annotations and failures.

The hand model and control interface also matter. A demonstration collected with one tendon routing, fingertip geometry or action space may not transfer directly to another hand. Dataset documentation should separate task-level dexterity from embodiment-specific commands and report success over repeated physical trials.

Sources