Citation: Feix, T., Romero, J., Schmiedmayer, H., Dollar, A., and Kragic, D. (2016). “The GRASP Taxonomy of Human Grasp Types.” IEEE Transactions on Human-Machine Systems, 46(1), 66–77. DOI: 10.1109/THMS.2015.2470657
Abstract
In this paper, we analyze and compare existing human grasp taxonomies and synthesize them into a single new taxonomy (dubbed “The GRASP Taxonomy” after the GRASP project funded by the European Commission). We consider only static and stable grasps performed by one hand. The goal is to extract the largest set of different grasps that were referenced in the literature and arrange them in a systematic way. The taxonomy provides a common terminology to define human hand configurations and is important in many domains such as human–computer interaction and tangible user interfaces where an understanding of the human is basis for a proper interface. Overall, 33 different grasp types are found and arranged into the GRASP taxonomy. Within the taxonomy, grasps are arranged according to 1) opposition type, 2) the virtual finger assignments, 3) type in terms of power, precision, or intermediate grasp, and 4) the position of the thumb. The resulting taxonomy incorporates all grasps found in the reviewed taxonomies that complied with the grasp definition. We also show that due to the nature of the classification, the 33 grasp types might be reduced to a set of 17 more general grasps if only the hand configuration is considered without the object shape/size.
I. Introduction
Understanding the way humans grasp objects, knowing the kinematic implications and limitations associated with each grasp, and knowing common use patterns is important in many domains ranging from medicine and rehabilitation, psychology, and product design, among many others. In human–computer interaction and tangible user interfaces, the hand is used to interact with technology, so understanding grasp posture — and how it adjusts to task demands — matters for interface design. Grasp-type classifications have also been used in programming by demonstration (mapping human grasps to robot grasps), as a functional layer mapping human hand kinematics to artificial hands, and in prosthetics, where hands are often designed around a discrete set of target grasp types.
The hand has 15 joints (excluding carpal/metacarpal joints), giving over 20 degrees of freedom, making direct modeling of hand shape difficult. However, the ways the hand interacts with grasped objects — grasp types — are far more limited and can be broken into subclasses. Many grasp classifications have been proposed across robotics, medicine, and biomechanics, but there is little consensus on the full range of grasp types humans commonly use. This paper compares all human grasp taxonomies in the literature, finds the largest set of distinct grasps, and synthesizes them into a single classification: the GRASP taxonomy (named after the EU GRASP project in which the initial version was developed). An earlier workshop version of this taxonomy has been widely used, inspiring taxonomies of microinteractions, robotic hand design studies, vision-system affordance definitions, and experiments on human hand function.
II. Background
A. Power, Intermediate, and Precision Grasps
Each grasp can be classified by its need for precision or power. Napier’s original distinction was extended by Landsmeer into “power grip” (rigid hand-object relation; all object movement comes from the arm) versus “precision handling” (hand performs intrinsic movements on the object without arm motion). A third category — intermediate/link grasp — was later added, where power and precision elements are present in roughly equal proportion.
B. Opposition Types
There are three basic directions, relative to the hand coordinate frame, in which the hand applies force to hold an object securely:
- Pad Opposition — between volar (palm-side) surfaces of fingers and thumb, near/on the pads (e.g., holding a needle or small ball).
- Palm Opposition — perpendicular to the palm (e.g., grasping a large hammer or screwdriver).
- Side Opposition — transverse to the palm (e.g., holding a key between thumb and radial side of fingers, or a cigarette between finger sides).
C. Virtual Finger (VF)
In many tasks several fingers work together as a functional unit, the “virtual finger.” Fingers belong to the same VF if they apply force in a similar direction and act in unison; VFs oppose each other in the grasp (like the jaws of a gripper or vice). E.g., in pad opposition the thumb maps to VF1 and the index finger to VF2; in palm opposition the palm is VF1 and the four fingers together are VF2. A third VF (VF3) is assigned to fingers opposing a task-related force/torque, if present.
III. Comparison of Taxonomies
A. Grasp Definition
“A grasp is every static hand posture with which an object can be held securely with one hand, irrespective of the hand orientation.” This implies the grasp must withstand force in every direction (scaled appropriately) and excludes: in-hand motion (object not in constant relation to the hand), bimanual tasks, and gravity-dependent grasps (e.g., Hook Grasp, Flat Hand Grasp) where hand orientation is essential to stability rather than merely to the task.
B. Comparison Table
The authors compared 22 existing taxonomies spanning robotics, medicine, and biomechanics literature. Grasps were registered in a large comparison table (22 publications × 47 grasp columns), with grasps considered equivalent if hand configuration, grasped object size, and contact surfaces were similar. Overall, 211 relevant grasp examples were found across the literature, arranging into 47 candidate grasp types. Of these, 5 were excluded for violating the grasp definition (e.g., hook grasp, platform grasp, push grasp), and 8 were merged as minor variations of already-counted grasps. This left 33 final valid grasp types.
The most consistently named grasp across authors was the “Lateral Grasp” (#16, 19 entries). Full hand wrap grasps split into three related-but-distinct types (medium wrap #3, large diameter #1, small diameter #2) which combined were referenced by 21 of the 22 publications. Taxonomy sizes in the literature varied widely — from as few as 3 grasps to as many as 21 — due to differing scope, precision of description, and treatment of object size/shape (e.g., Cutkosky’s taxonomy distinguishes grasps by object size; many others do not).
C. Analysis of Power and Precision Grasps
9 of the 22 surveyed taxonomies classified grasps into power/intermediate/precision (PIP) categories. Power and precision grasps were about equally represented (40 and 41 listings respectively); intermediate grasps were rarer (8 listings). Where at least two publications classified the same grasp, agreement was high (13 of 19 comparable grasps had full consensus). The Lateral grasp (#16) was notably inconsistent — classified as precision, intermediate, or power by different authors — reinforcing that it sits at an intermediate stage between power and precision; the GRASP taxonomy places it in the intermediate category.
IV. Synthesis of Grasp Taxonomies
The 33 valid grasp types were arranged into a matrix (Fig. 4 in the paper). Columns are organized by power/precision/intermediate requirement, then by opposition type (Palm, Pad, or Side), which also determines the VF1 assignment (palm → VF1 for Palm Opposition; thumb → VF1 for Pad/Side Opposition, except the Adduction Grip where the thumb doesn’t contact the object). Rows are differentiated by thumb position — abducted (thumb can oppose fingertips) or adducted (thumb applies force to the side of the fingers, or is “out of the way”) — a new classification dimension introduced by this taxonomy.
Compared to prior taxonomies (the largest of which listed 20–21 grasp types; Cutkosky’s widely-used robotics taxonomy lists 15), the GRASP taxonomy’s 33 types represent a larger, more complete set.
A. Merging of Grasps Within One Cell
Because many grasps share opposition type and thumb position and differ mainly in object shape, several taxonomy cells contain multiple grasps. Merging same-cell grasps into a single prototypical grasp reduces the 33 types down to 17 more general grasps — trading classification precision for simplicity depending on task needs.
B. Completeness of the Taxonomy
A follow-up study observing two housekeepers and two machinists (≈4700 grasps, 7.45h/subject) using the GRASP taxonomy found all 33 grasp types were used except Tripod Variation (#21) and Distal Type (#19) — both highly specialized (scissors, chopsticks-like use) and simply not present in the observed activities.
C. Statistics of the Cells in the Taxonomy
Using frequency/duration and inter-rater confusion data from the housekeeper/machinist study: most populated cells have meaningful real-world frequency; a few rare cells (adduction grip #23, distal type #19, tripod variation #21) contain a single specialized grasp still used in important niche cases. Picking the most frequent grasp per cell as the “prototype” (the 17-grasp reduced set) still covers 83.4% of grasp duration and 75.8% of grasp frequency in the observed dataset.
D. Grasp-Type Properties
Follow-up studies compiled per-grasp-type properties (object mass, size/hand aperture, rigidity, and whether grasp force is dominated by object weight vs. task interaction force). Most grasps are used on objects under 500g requiring ≤5cm hand aperture — i.e., most grasps don’t operate the hand near its force/aperture limits. Most grasps are used on rigid objects; power sphere (#11) and precision disk (#12) are notable exceptions, mainly used on “floppy” (easily deformable) objects.
Perceived vs. real frequency: comparing how often a grasp is referenced in the literature vs. how often it’s actually observed in real-life video data shows only weak correlation (Pearson r = 0.36) — e.g., palmar pinch (#9) and tip pinch (#24) are over-represented in the literature relative to real usage, while the lateral tripod (#16) is under-represented in the literature relative to its real frequency.
V. Discussion
Classification depends not just on hand pose but also on hand-object contact type — e.g., medium wrap (#3) and prismatic 4-finger (#6) have similar hand shapes but differ in whether the palm contacts the object, changing their opposition type (palm vs. pad). This makes classification easy for humans but harder for automated grasp-recognition systems, which may need multiple measurement modalities (e.g., joint angle + contact sensing) to correctly classify grasps in this taxonomy.
Thumb position (abducted/adducted) is a new taxonomic feature: the thumb must be abducted to oppose the fingertips (as in most pad grasps), and is adducted mainly when opposition is against the side of the finger (e.g., lateral grasp) or when the thumb isn’t involved in opposition (e.g., fixed hook, palmar).
Grasp frequency varies considerably by environment (different objects/tasks afford different grasps), which matters for HCI systems trained to interpret hand actions — such systems may not generalize well across environments where grasp-type frequency shifts.
The authors anticipate the taxonomy will be useful for robotic grasping: a common way to reduce control complexity for high-DOF hands is to define a discrete set of grasp types for the hand to execute, and the GRASP taxonomy can guide selection of the most useful/appropriate types — including as an intermediate representation in learning-by-demonstration pipelines (human motion → grasp type → robot execution).
VI. Conclusion and Future Work
From a comprehensive literature review, 211 grasp examples were found and reduced to 33 unique prehensile grasp types, arranged in a novel taxonomy by opposition type, VF assignment, power/precision/intermediate category, and thumb position (reducible to 17 grasps at coarser granularity). Future directions include: examining taxonomy completeness across a larger diversity of environments (potentially via data-driven methods), and extending the taxonomy to nonprehensile grasps or dynamic in-hand manipulation — currently excluded by the static-grasp definition, and possibly best left as separate classifications given the added complexity. The taxonomy and supplementary documents are available at grasp.xief.net.
The 33 Grasp Types (numbered, from Fig. 4/Table I)
Power grasps: 1 Large Diameter, 2 Small Diameter, 3 Medium Wrap, 4 Adducted Thumb, 5 Light Tool, 10 Power Disk, 11 Power Sphere, 15 Fixed Hook, 17 Index Finger Extension, 18 Extension Type, 19 Distal Type, 26 Sphere 4 Finger, 28 Sphere 3 Finger, 30 Palmar, 31 Ring
Intermediate grasps: 16 Lateral, 21 Tripod Variation, 23 Adduction Grip, 25 Lateral Tripod, 29 Stick, 32 Ventral
Precision grasps: 6 Prismatic 4 Finger, 7 Prismatic 3 Finger, 8 Prismatic 2 Finger, 9 Palmar Pinch, 12 Precision Disk, 13 Precision Sphere, 14 Tripod, 20 Writing Tripod, 22 Parallel Extension, 24 Tip Pinch, 27 Quadpod, 33 Inferior Pincer
Each grasp is further indexed by opposition type (Palm/Pad/Side), virtual finger (VF) assignment, and thumb position (abducted/adducted) — see Fig. 4/5 in the paper for the full matrix and per-grasp object mass/size/rigidity/force statistics.