Grasp Taxonomy
- Also called: grasp classification
- Category: Planning & manipulation › Manipulation
- Tags: grasping, human-hand, dexterous-manipulation, prosthetics, hand-design
A structured classification of the ways a hand can hold objects, such as power, precision and lateral grasps, used to analyze human grasping and to design and control robot hands.
Definition
A grasp taxonomy is a classification of hand grasps into a discrete set of named types, organized by criteria such as which parts of the hand oppose each other, how many fingers are involved and whether the grasp emphasizes stability or fine control. The oldest and most widely used distinction is between power grasps, where the object is held against the palm, and precision grasps, where it is held between the fingertips and the thumb, introduced by Napier in 1956. Later taxonomies refine this into many named types such as large and small wrap, tripod, lateral pinch, hook and disk grasps.
Two taxonomies are especially common in robotics. Cutkosky (1989) studied machinists working with tools and organized grasps into a tree that runs from power to precision grasps and from object shape to task requirements, with the aim of guiding the design of robot hands for manufacturing tasks. The GRASP taxonomy of Feix et al. combined earlier taxonomies into 33 grasp types arranged in a table by opposition type (pad, palm or side opposition), by whether the thumb is abducted or adducted, by the number of virtual fingers and by the power, intermediate or precision character of the grasp. Opposition type and virtual fingers, ideas developed by Iberall and colleagues, describe a grasp by which hand surfaces apply opposing forces rather than by individual finger positions.
For robots, taxonomies serve as a shared vocabulary rather than a control law. They are used to label human demonstrations so that a robot can recognize which kind of grasp a person used, to choose among a small set of preshapes for a multi-fingered hand, to compare how many human grasp types a robot or prosthetic hand can reproduce, and to structure grasp datasets. A grasp type says little about exact contact locations or forces, so it is normally combined with geometric grasp planning or learned controllers that turn the chosen type into an actual grasp on a specific object.
Why it matters
Grasp taxonomies connect human hand use with robot hand design and control. They give engineers a compact way to specify what a hand must be able to do, to evaluate dexterous and prosthetic hands, and to turn human demonstrations into grasp choices a robot can act on.
Key applications
- Designing and benchmarking robotic and prosthetic hands
- Labeling human grasp demonstrations for imitation learning
- Selecting preshapes for multi-fingered grasp planning
- Recognizing grasp types from images, gloves or EMG signals
- Ergonomics and hand function assessment
In practice
A team builds a system that learns kitchen tasks from people demonstrating in virtual reality with tracked data gloves. Each demonstration is segmented into actions, and each grasp is labeled with a type from the Feix taxonomy, such as medium wrap for a bottle, tripod for a spice jar lid and lateral pinch for a card. When the robot reproduces a task, it maps the labeled type to one of a few preshapes available on its multi-fingered hand and runs a grasp planner restricted to that preshape on the target object. The same labels let the team count which grasp types their hand cannot reproduce and decide whether a different hand design would cover the tasks they care about.
Common misconceptions
- A grasp type fully specifies a grasp — it names the hand configuration family but not the contact points, forces or object pose, which still have to be planned
- There is one standard grasp taxonomy — several are in use, including Napier's, Cutkosky's and Feix et al.'s, and they differ in categories and criteria
- Grasp taxonomies apply only to human-like hands — opposition types and virtual fingers can also describe grasps made by parallel-jaw and underactuated grippers
Who should know this
- Robot hand and prosthesis designers
- Dexterous manipulation researchers
- Engineers building learning-from-demonstration systems
- Biomechanics and human-motion researchers
Related technologies
- GRASP Taxonomy (Feix et al.)
- Cutkosky Grasp Taxonomy
- Data Gloves
- Hand Pose Estimation
- Electromyography (EMG)
Related terms
- Grasp Planning
- Dexterous Manipulation
- Eigengrasp
- Underactuated Hand
- Motion Retargeting
- Task-Oriented Grasping
Related projects
- Gymnasium-Robotics: Robotics RL benchmark environments for Gymnasium, from the Farama Foundation
- mink: Python inverse kinematics library built on MuJoCo supporting 20+ robot platforms
- TriFinger: Open-source 9-DOF tri-finger dexterous manipulation platform for RL and robotics research
- TRLC-DK1: Bimanual robot platform for humanoid data collection with LeRobot
- ORCA Hand: Open-source tendon-driven robotic hands with tactile sensing from ETH Zurich
- RoboKit: GPU toolkit that solves thousands of robot IK, planning and retargeting problems at once