PaperlayerView exact revision on arXiv

Source-linked AI summary

A Quasi-Direct-Drive Underactuated Asymmetric Hand for Dexterous and Efficient Grasping and Manipulation

Benjamin Davis, Chase Kidder, Hannah S. Stuart

arXiv:2609.27240v1cs.RO

TL;DR

QDD actuation supports force transparency, but its low torque density makes human-like strength difficult to achieve without size, weight, and heating constraints. The QUAD Hand combines radial QDD actuation with an underactuated, compliantly coupled ulnar finger driven by a shared larger motor. The hand achieved 29 out of 33 possible Feix taxonomy grasps, and the single shared actuator reduced heat energy by up to 96×, even when normalizing for force per finger.

  • Problem

    QDD actuation supports force transparency, but its low torque density makes human-like strength difficult to achieve without size, weight, and heating constraints.

    Source: §I · p. 1

  • Method

    The QUAD Hand combines radial QDD actuation with an underactuated, compliantly coupled ulnar finger driven by a shared larger motor.

    Source: A. Overview · p. 2, 1) Ulnar Transmission Linkage · p. 4

  • Results

    The hand achieved 29 out of 33 possible Feix taxonomy grasps, and the single shared actuator reduced heat energy by up to 96×, even when normalizing for force per finger.

    Source: 1) Grasp Taxonomy · p. 5, 2) Sustained Load Performance · p. 7

  • Takeaways & Limitations

    The evaluations support role-based finger asymmetry as an approach to human-like grasp diversity, force sensitivity, and thermal efficiency.

    Source: §VI · p. 7, §VII · p. 8

    The evaluations are not comprehensive: grasp-taxonomy and Kapandji tests assess static poses, force evaluations cover one joint position, and printed ulnar links yield before actuator stall torque.

    Source: §VI · p. 7

Abstract

from arXiv · show

In this paper, we present the Berkeley QUAD (Quasi-direct-drive, Underactuated, Asymmetric Design) Hand, a four-finger anthropomorphic robotic hand with 11 degrees of freedom and 8 degrees of actuation. The design utilizes QDD actuation at the base of each finger, enabling high force transparency for dexterous, adaptive performance. However, the low torque density of these actuators traditionally presents major issues with size, weight, and thermal limits. We overcome this by applying bio-inspired asymmetry, delegating dexterity to the radial fingers through individual QDD actuation, and strength to the ulnar finger through an underactuated, compliantly coupled transmission driven by a larger QDD motor. A novel preloaded, linkage-based transmission permits this ulnar coupling in a way that preserves human-like workspace reachability. Under light loads, the ulnar motor drives the third (middle) finger directly for dexterity while the fourth (ring) finger mirrors its motion. However, under larger loads, the middle finger complies while the motor drives the ring finger further downwards and inwards toward the center of the grasp to apply better closure forces. Hardware evaluations validate this architecture, demonstrating that the hand achieves 29 out of 33 Feix taxonomy grasps and exhibits backdrive forces as low as 50 g for delicate interactions. Additionally, the underactuated fourth finger improves grasp closure and provides the spatial efficiency necessary for larger actuation, yielding up to a 96-fold reduction in heat generation during sustained loading. Webpage: https://benudavis.github.io/berkeley-quadhand/

I. INTRODUCTION

QDD actuation supports force-transparent, adaptive contact, but scaling it to a human-like hand creates weight, size, and heating constraints. The QUAD Hand assigns dexterity to radial fingers and strength and closure to mechanically coupled ulnar fingers.

I. INTRODUCTION

  • Passive mechanical compliance adapts to disturbances without a software loop but limits fine manipulation, while active force feedback requires sensing and is constrained by backlash and power dependence.QDD force transparency is presented as an alternative for safe contact.Source: §I · pp. 1–2
  • QDD motors offer low reflected inertia and friction, but human-like strength makes them heavy, bulky, and prone to rapid heating; fully articulated QDD hands are therefore practically infeasible.Existing QDD approaches are limited to three fingers, losing grasp stability provided by the ulnar fingers.Source: §I · p. 2
  • Radial fingers perform diverse, dexterous motions, whereas ulnar fingers contribute strength, stability, and closure, and their movements are highly correlated.These role differences motivate assigning distinct actuation strategies to the two groups.Source: §I · p. 2

A. Overview

  • The QUAD Hand uses a hybrid QDD and underactuated architecture to balance human-inspired dexterity, strength, stability, and compliance.A mechanically intelligent transmission drives the fourth finger through flexion and cupping motions, coupled to the middle finger without dedicated actuation.Source: A. Overview · p. 2
  • The fourth finger’s compliant coupling frees palm volume for a larger actuator, while a novel linkage lets it sweep inward toward the palm for improved grasp security.The architecture is designed to provide an economical hand with human-like functional workspace and strength.Source: A. Overview · p. 2

II. RELATED WORK

Prior asymmetric hands often reduce ulnar-finger capabilities to simplify the design, while palmar actuation commonly requires extra motors or tendons. The QUAD Hand instead uses asymmetric actuation and linkage coupling to retain grasp-relevant ulnar motion.

A. Finger Asymmetries

  • Earlier asymmetric hands commonly remove abduction/adduction, couple MCP and IP joints, or fully couple ulnar fingers, treating asymmetry as a compromise that diminishes ulnar capabilities.The QUAD Hand frames asymmetry as a feature, using mechanical intelligence and consolidated actuation to improve adaptability and strength.Source: A. Finger Asymmetries · p. 2

B. Palmar Actuation

  • Flexible CMC joints give human ring and little fingers palmar cupping, opposition, and adaptability, but robotic implementations often require extra actuation or tendon coupling.Tendon-driven coupling can suffer hysteresis, friction, and cable stretch; the QUAD Hand uses a linkage transmission instead.Source: B. Palmar Actuation · p. 2

III. DESIGN

  • The QUAD Hand has four fingers, 11 DoF, and 8 DoA, with asymmetric joint actuation and a ulnar coupling mechanism that enables the ring finger to sweep inward when the middle finger is obstructed.The hand omits the little finger because its grasping role and motion correlate with those of the ring finger.Source: §III · p. 2, Figure 2 · p. 3

A. Hybrid Actuation Strategy

The hand combines QDD actuation near the palm with a larger shared ulnar motor and compliant linkage coupling. Its transmission changes the ring finger’s motion when the middle finger encounters resistance, while the kinematic analysis addresses reachable workspace.

A. Hybrid Actuation Strategy

  • The thumb and index use compact, relatively weak QDD actuators, while coupling the middle and ring fingers makes room for one larger actuator to drive both.QDD actuation near the palm preserves proximal backdrivability while limiting distal mass and bulk.Source: A. Hybrid Actuation Strategy · p. 3

B. Finger Transmissions

  • The ulnar motor drives a preloaded compliant transmission connected to the middle MCP and ring CMC, MCP, and IP joints; the ring finger’s serial underactuated linkage allows adaptive wrapping.A spring keeps the ring IP joint straight in free motion so the finger moves as one rigid body about the MCP; contact at the proximal link then permits distal-link closure.Source: B. Finger Transmissions · p. 3
  • Below an approximately 3.2 N middle-fingertip force, the spring remains undeflected and the linkage moves the middle and ring joints on parallel planes.The threshold is tunable through initial spring tension.Source: B. Finger Transmissions · p. 3
  • Above the middle-fingertip force threshold, spring deflection drives the ring finger inward about its CMC joint, placing its motion on an intersecting plane with the middle finger for cupping.The output link’s path is constrained by equal-length links.Source: B. Finger Transmissions · p. 3

IV. KINEMATIC WORKSPACE ANALYSIS

  • The kinematic analysis derives reachable workspace for each finger and depicts the coupled middle/ring chains alongside the index finger’s separate flexion and abduction linkages.The figure identifies the ulnar transmission, middle MCP linkage, ring base spatial linkage, and underactuated ring linkage.Source: §IV · p. 3, Figure 3 · p. 4

A. Ulnar Kinematics

The ulnar linkage analysis uses a closed vector loop to determine linkage position and then derives an angle solution from the constrained chain. With that angle known, it can determine the position of point L.

1) Ulnar Transmission Linkage:

  • The ulnar transmission linkage determines the middle MCP and ring MCP and CMC angles, with its global position shifted by the rack-and-pinion input.A pretensioned spring biases the sliding input link toward minimum length; its extension translates the linkage and transfers motion through point L to flex the ring CMC joint.Source: 1) Ulnar Transmission Linkage · p. 4
  • The parallelogram constraint makes θFE constant, allowing the analysis to solve the remaining AHGFECA chain without explicitly solving the internal linkage.Fixed angles θAH, θHG, and θAC are used in the chain analysis.Source: 1) Ulnar Transmission Linkage · p. 4
  • After separating real and imaginary parts and simplifying to harmonic form, the analysis solves for θGF using a tangent half-angle substitution.With θGF known, the fixed orientation of DEFL makes point L’s global position straightforward to determine.Source: 1) Ulnar Transmission Linkage · p. 4

2) Middle MCP Slider-Crank:

The middle MCP angle is modeled as a constant offset from θKJ, with the slider-crank input formed by summing the motor and compliant-joint translations.

2) Middle MCP Slider-Crank:

  • The slider-crank input OH is the sum of the motor-position translation OA and compliant-joint translation AH; the middle MCP angle is a constant offset from θKJ.Eqs. 2 and 3 are used to solve θKJ before calculating the middle finger MCP angle.Source: 2) Middle MCP Slider-Crank · p. 4

3) Ring CMC and MCP Spatial Linkage:

The spatial chain determines the ring CMC and MCP angles from point L, but its 3D geometry and spherical joint motivate a numerical solution. The method matches two product-of-exponentials estimates of point M.

3) Ring CMC and MCP Spatial Linkage:

  • The ring CMC angle θPN and underactuated ground-link angle θNM are deterministic for a given position of L through the closed-loop chain OLMNP.Because the chain is three-dimensional and includes a spherical joint at M, the authors avoid a high-order analytical solution.Source: 3) Ring CMC and MCP Spatial Linkage · p. 4
  • Two product-of-exponentials solutions estimate point M, one from palm origin O and one from transmission point L.The formulation uses spatial screw axes and home-configuration matrices for each origin, with TOL relating O to L.Source: 3) Ring CMC and MCP Spatial Linkage · pp. 4–5
  • The error vector e(θ) is formed between the two point-M estimates, with θ = [θPN, θNM, θLM]T as the variables to minimize.The optimization uses Levenberg–Marquardt via scipy.optimize.root, initialized from the prior timestep.Source: 3) Ring CMC and MCP Spatial Linkage · p. 5

4) Underactuated Ring Finger Linkage:

The ring finger uses a two-segment underactuated linkage: a four-bar rotates its second segment about the IP joint, while its ground-link angle derives from θNM.

4) Underactuated Ring Finger Linkage:

  • The four-bar linkage NQRS in the first segment rotates the second segment about the ring finger’s IP joint.The ground-link angle NQ is a constant offset from θNM, and the loop is solved relative to that ground link.Source: 4) Underactuated Ring Finger Linkage · p. 5

B. Index Kinematics

The index uses separate linkages for MCP abduction and flexion, and the derived kinematics characterize finger workspaces and grasp reachability.

B. Index Kinematics

  • The index MCP abduction linkage solves θO’C from input θAB, while its parallelogram flexion linkage maps θDE to the flexion angle with an offset.The abduction angle also uses a fixed offset.Source: B. Index Kinematics · p. 5

C. Finger Workspace

  • The passive ring-finger workspace sweeps inward as the ulnar transmission complies, while the thumb can oppose every other finger.Figure 4C–D shows passive workspace at fixed middle MCP positions; thumb-ring opposition is enhanced by the ring CMC joint during grasping.Source: C. Finger Workspace · p. 5, Figure 4 · p. 5

V. EXPERIMENTAL EVALUATIONS

  • Hardware experiments assess grasp-motion diversity, QDD force sensitivity, and efficiency from consolidated ulnar actuation.Source: §V · p. 5

A. Reachable Poses

  • The reduced actuation is evaluated for reachability across the Feix grasp taxonomy and Kapandji test.The section frames these tests as checking whether reduced actuation preserves practical dexterity.Source: A. Reachable Poses · p. 5

1) Grasp Taxonomy:

The QUAD hand demonstrates broad grasp reachability despite reduced actuation, while retaining low backdrive forces at the radial fingers. Its ring finger also supports both cylindrical and spherical grasp patterns through compliant coupling.

1) Grasp Taxonomy:

  • 29 of 33 Feix taxonomy grasps are achieved, with the underactuated ring finger supporting both cylindrical and spherical grasp patterns.For cylindrical objects, it flexes mainly at the MCP joint; for spherical objects, the transmission lets it sweep inward about the CMC joint.Source: 1) Grasp Taxonomy · p. 5, Figure 5 · p. 6

2) Kapandji Test:

  • 10 of 11 Kapandji postures are achieved; the sole miss is pose #5, which requires a fifth digit.The result supports full palmar opposition and thumb–ring opposition without flexing the passive ring CMC joint.Source: 2) Kapandji Test · p. 5, Figure 5 · p. 6

1) Minimum Backdrive Force:

  • Minimum backdrive force is measured by slowly pressing a force gauge against a fingertip with the hand rigidly mounted and joints at zero current.Peak forces are recorded over 15 trials and reported as means with standard deviations.Source: 1) Minimum Backdrive Force · p. 6, Figure 6 · p. 7
  • About 50 g is sufficient to backdrive the thumb and index base joints, while the middle and ring fingers require about 250 g.The radial-finger sensitivity supports delicate interactions; the middle and ring sensitivity is described as sufficient for many everyday grasping tasks, but potentially inadequate for extreme precision.Source: 1) Minimum Backdrive Force · pp. 6–7, Figure 6 · p. 7

2) Sustained Load Performance:

Sustained-load tests show that the shared ulnar actuator draws little current and dissipates substantially less heat than the smaller radial actuators. The authors attribute the heat difference to both current draw and phase resistance.

2) Sustained Load Performance:

  • Sustained-load evaluation applies 3 N at one fingertip or 6 N across the middle and ring fingers, then derives resistive power loss and cumulative heat from quadrature-axis current.The phase resistance used in the calculation is manufacturer-provided.Source: 2) Sustained Load Performance · p. 7
  • The ulnar actuator draws the least current; during individual loading, it draws about twice as much current through the middle finger as through the ring finger.The authors suggest finger-length differences and transmission-linkage kinematics as likely contributors.Source: 2) Sustained Load Performance · p. 7
  • At a combined 6 N load on the middle and ring fingers, current draw is similar to loading the middle finger alone at 3 N, suggesting most load transfers through the ring finger.The tests apply a sustained fingertip load for five minutes.Source: 2) Sustained Load Performance · p. 7
  • Up to a 96× heat-energy reduction is reported with the shared actuator, even when normalized for force per finger.The smaller radial actuator’s phase resistance is more than 17 times that of the ulnar actuator; current and phase resistance both contribute to heat dissipation.Source: 2) Sustained Load Performance · p. 7

VI. DISCUSSION AND LIMITATIONS

The evaluations support the QUAD Hand’s design objectives, while the authors emphasize that they are not comprehensive and identify several directions for further development. The conclusion argues for role-based finger asymmetry as a direction for anthropomorphic hand design.

VI. DISCUSSION AND LIMITATIONS

  • The QUAD Hand’s evaluations support its intended objectives, including high force sensitivity and thermal efficiency for practical manipulation.Linkage-driven coupling consolidates ulnar actuation into a stronger, more efficient motor while maintaining attainability of common human grasp poses.Source: §VI · p. 7, §VII · p. 8
  • The evaluations are not comprehensive: grasp tests cover static poses, force tests use one position per joint, and printed ulnar linkages yield before actuator stall torque.The authors suggest metal for critical loadbearing parts; non-constant linkage kinematics also mean performance varies with position.Source: §VI · p. 7
  • Future work includes optimizing linkage torque profiles and compliant ring-finger trajectories, adding tactile sensing, evaluating the distal IP joint, and integrating modern manipulation control.The authors identify these as topics for further exploration.Source: §VI · p. 7

VII. CONCLUSION

  • The conclusion advocates role-based finger asymmetry over strict symmetry, using underactuation and compliant coupling to capitalize on the distinct advantages of QDD and mechanical intelligence.The authors frame this as a direction toward hands suited to unstructured human environments.Source: §VII · p. 8
Loading 2609.27240v1…