PaperlayerView exact revision on arXiv

Source-linked AI summary

RBO Hand 3 -- A Platform for Soft Dexterous Manipulation

Steffen Puhlmann, Jason Harris, Oliver Brock

arXiv:2201.10883v1cs.RO

TL;DR

The paper targets a soft hand for dexterous manipulation, transfer of human strategies, and sustained real-world experimentation. It presents RBO Hand 3, combining pneumatic actuation with intrinsic compliance and human-like functionality. The hand achieves the highest possible Kapandji score, all 33 GRASP taxonomy postures, replicates common human grasping strategies, and rotates objects in-hand.

  • Problem

    The paper seeks a soft robotic hand that supports dexterous manipulation, transfer of human strategies, and extensive real-world experimentation.

    Source: §I · p. 1, A. Enabling Dexterous Manipulation · p. 2

  • Method

    RBO Hand 3 combines many actuated degrees of freedom with intrinsic compliance and anthropomorphic features in a pneumatic hand.

    Source: §I · p. 1, 1) PneuFlex Actuator · p. 4

  • Results

    The hand achieves the highest possible Kapandji score and all 33 GRASP taxonomy postures, replicates common human grasping strategies, and rotates objects in-hand.

    Source: §VIII · p. 15

  • Takeaways & Limitations

    Intrinsic compliance absorbs contact dynamics and allows passive adaptation to objects and the environment during manipulation.

    Source: §VIII · p. 15

    The authors identify sensorization and improved pneumatic actuation as future work.

    Source: §VIII · p. 15

Abstract

from arXiv · show

We present the RBO Hand 3, a highly capable and versatile anthropomorphic soft hand based on pneumatic actuation. The RBO Hand 3 is designed to enable dexterous manipulation, to facilitate transfer of insights about human dexterity, and to serve as a robust research platform for extensive real-world experiments. It achieves these design goals by combining many degrees of actuation with intrinsic compliance, replicating relevant functioning of the human hand, and by combining robust components in a modular design. The RBO Hand 3 possesses 16 independent degrees of actuation, implemented in a dexterous opposable thumb, two-chambered fingers, an actuated palm, and the ability to spread the fingers. In this work, we derive the design objectives that are based on experimentation with the hand's predecessors, observations about human grasping, and insights about principles of dexterity. We explain in detail how the design features of the RBO Hand 3 achieve these goals and evaluate the hand by demonstrating its ability to achieve the highest possible score in the Kapandji test for thumb opposition, to realize all 33 grasp types of the comprehensive GRASP taxonomy, to replicate common human grasping strategies, and to perform dexterous in-hand manipulation.

I. INTRODUCTION

RBO Hand 3 extends earlier soft hands with greater dexterity and robustness, guided by the goals of dexterous manipulation, human-strategy transfer, and sustained real-world experimentation. Its dexterity framework combines compliance with actuation to rearrange physical constraints, manipulate objects, and compensate for uncertainty.

I. INTRODUCTION

  • RBO Hand 3 extends earlier compliant, underactuated pneumatic hands with substantially increased dexterous abilities and robustness for real-world experimentation.The first generation relied on compliance for robust grasping, while the second added actuation and anthropomorphic features.Source: §I · p. 1
  • The design targets dexterous manipulation, transfer of human strategies, and a reliable platform for extensive real-world experiments.The authors report that the hand served as an effective, extremely reliable in-hand manipulation platform in their lab over the last year.Source: §I · pp. 1–2

A. Enabling Dexterous Manipulation

  • Manipulation funnels use physical constraints to guide object state and reduce uncertainty, supporting versatile and robust manipulation.The hand can rearrange its own features as well as exploit constraints supplied by the environment.Source: A. Enabling Dexterous Manipulation · p. 2
  • Funnel-based dexterity requires compliance and sufficient actuation to modify compliance, rearrange constraints, and actuate the object.These purposes of actuation overlap during real-world manipulation.Source: A. Enabling Dexterous Manipulation · p. 2

B. Leveraging Insights From Human Manipulation

The hand's anthropomorphic design is intended to support transfer of human manipulation strategies, while robust components and repair simplicity support prolonged experimentation. In related-work comparisons, the authors position RBO Hand 3 as combining compliance, many actuated degrees of freedom, and human-like functionality.

B. Leveraging Insights From Human Manipulation

  • Morphological and functional resemblance to the human hand is intended to facilitate transfer of observed human manipulation strategies.The authors use human-hand functionality to guide selection of relevant actuated degrees of freedom.Source: B. Leveraging Insights From Human… · p. 3

C. Support Extensive Real-World Experiments

  • Design changes to features and manufacturing processes were intended to support hundreds of hours of continuous grasping and manipulation experiments without failure.The design also prioritizes robust components and minimizing repair complexity because research labs cannot match commercial products' reliability.Source: C. Support Extensive Real-World… · p. 3

III. RELATED WORK

  • Unlike underactuated soft hands that lack broad reconfiguration, RBO Hand 3 combines actuation and compliance with anthropomorphic functionality for dexterous manipulation and strategy transfer.The paper reports versatility and dexterity comparable to the 26-DOF BCL-26 hand, despite RBO Hand 3 having fewer actuated degrees of freedom.Source: §III · p. 3
  • The BCL-26 hand achieves the highest possible Kapandji score and all GRASP taxonomy types; RBO Hand 3 has fewer actuated degrees of freedom but similar reported versatility and dexterity.The cited BCL-26 has 26 actuated degrees of freedom.Source: §III · p. 3

IV. REALIZATION OF THE DESIGN OBJECTIVES

The design objectives call for actuation, intrinsic compliance, human-hand functionality, and reliability in sustained experiments. The section also points readers to figures showing the hand's feature nomenclature and component manufacturing.

IV. REALIZATION OF THE DESIGN OBJECTIVES

  • The design objectives are dexterous manipulation, transfer of human manipulation strategies, and extensive real-world experimentation.The hand is intended to combine sufficient actuation, intrinsic compliance, relevant human-hand features, and uninterrupted experimental operation.Source: §IV · p. 4
  • Figure 2 provides the nomenclature correspondence between human-hand features and RBO Hand 3 features, while Figure 8 illustrates manufacturing.Source: §IV · p. 4

A. Compliant Actuation

RBO Hand 3 combines intrinsically compliant pneumatic actuation with 16 independently actuated degrees of freedom to support dexterous manipulation and replicate relevant human-hand functionality.

A. Compliant Actuation

1) PneuFlex Actuator:

  • PneuFlex actuators bend on inflation because a thread helix constrains radial expansion while inextensible palmar fabric limits extension.Their geometry can adjust inflation profile, strength, and stiffness.Source: 1) PneuFlex Actuator · p. 4

2) Bellow Actuator:

  • Stackable bellow actuators use thin inflatable pouches to produce large rotations with negligible bending radius in a small form factor.Multiple pouches can share one tube, and bellow strength scales with chamber area, limiting use where finger cross-sections are constrained.Source: 2) Bellow Actuator · p. 4

3) Pneumatic Control:

  • Air-mass control regulates preset behavior without changing actuator air masses during contact-based deformation, enabling open-loop specification of desired compliance.Air flow is regulated using a linear forward model and pressure-sensor input for valve switching.Source: 3) Pneumatic Control · p. 5

B. Enabling Dexterous Manipulation

  • 16 independent actuated degrees of freedom support rearranging constraints, changing compliance, and actuating the manipulandum.These actuation purposes overlap during real-world manipulation.Source: B. Enabling Dexterous Manipulation · p. 5

1) Two-Compartment Finger:

The hand’s two-compartment fingers and multi-actuator opposable thumb provide distinct movements, workspaces, and force capabilities. Their designs replicate selected human-hand functions while supporting varied manipulation.

1) Two-Compartment Finger:

  • Two finger chambers independently actuate the MCP joint and coupled PIP/DIP joints, supporting precision grasping with flexed MCP and extended IP joints.The little finger is shorter to coordinate this behavior with the actuated palm.Source: 1) Two-Compartment Finger · p. 5
  • Two independent finger actuations create a palmar workspace resembling the human finger’s two arcs and allow forces across substantial workspace regions.Figure 4 maps workspace and force directions under different chamber inflation patterns.Source: 1) Two-Compartment Finger · p. 5, Figure 4 · p. 5
  • The finger’s maximum force is ca. 8.3 N when fully extended and 0 N when fully flexed.The force depends on the gap between the contacted fingertip position and its no-contact position at a given inflation level.Source: 1) Two-Compartment Finger · p. 5

2) Dexterous Opposable Thumb:

  • Four thumb actuators comprise a tip PneuFlex actuator and three bellows that produce anteposition, abduction, and flexion movements.The actuator arrangement is intended to maximize thumb opposability.Source: 2) Dexterous Opposable Thumb · p. 6, Figure 5 · p. 6
  • At 20° opening and 250 kPa, the thumb’s Proximal, Middle, and Distal bellows exerted 4.4 Nm, 3.2 Nm, and 1.9 Nm, respectively.Measured torque decreased with opening angle.Source: 2) Dexterous Opposable Thumb · p. 6

3) Palm Hollowing:

The actuated palm imitates human ring- and little-finger CMC flexion, while finger abduction broadens workspace and improves grasping versatility.

3) Palm Hollowing:

  • Human ring- and little-finger CMC flexion improves thumb opposition and lets the palm adapt its shape to objects or the environment.Little-finger CMC flexion can reach up to 30°.Source: 3) Palm Hollowing · p. 7
  • The palm bellow’s additional actuation greatly improves thumb opposition and the hand’s dexterity and versatility.It imitates simultaneous flexion at the human ring- and little-finger CMC joints.Source: 3) Palm Hollowing · p. 7

4) Finger Abduction:

  • Inflatable abduction bellows between neighboring fingers increase workspace and improve hand dexterity and versatility.Finger spreading also helps the hand encompass objects and apply forces from different directions.Source: 4) Finger Abduction · p. 7
  • A soft layer further increases the inherent mechanical compliance of the fingers and palm.Its overall shape and material are inspired by human-hand functionality.Source: 5) Soft Layer · p. 7

5) Soft Layer:

The RBO Hand 3 combines soft, passively adapting contact surfaces with modular construction and actuator changes intended to support durable, extensive real-world experimentation. Its modularity facilitates repair and design exploration, while the authors estimate at least 300 hours of continuous use but could not calculate a reliable mean time to failure.

5) Soft Layer:

  • Soft silicone finger pulps and a glove adapt to object shapes, increasing contact area; the pulps also reduce the flexion cavity that caused slipping in earlier versions.The softer contact material (shore hardness 00-30) is distinct from the harder actuator silicone (shore hardness A-30), which withstands high pressures and maintains structural integrity.Source: 5) Soft Layer · p. 7
  • The hand’s 16 compliant pneumatic degrees of actuation comprise eight in the fingers, four in the thumb, three for finger abduction, and one for palm hollowing.A soft layer on the palm and palmar sides of the fingers enables passive shape adaptation.Source: 6) Summary: Enabling Dexterous… · p. 7

1) Modularity:

  • Modularity reduces distinct parts and allows broken components to be replaced quickly, while supporting physical evaluation of new parts and digit combinations.The hand uses connected scaffolds, screw-mounted components, and a removable silicone glove.Source: 1) Modularity · p. 7

2) Improved Robustness:

  • Thicker actuator walls, increased adhesion area and ridges, and more thread-helix turns were introduced to reduce rupture, bond failure, and silicone cutting under inflation.The changes target stronger fingers and repeated high-pressure use.Source: 2) Improved Robustness · pp. 7–8

C. Support Extensive Real-World Experiments

  • The authors estimate at least 300 hours of continuous use for the two-compartment fingers, but report too few failures to calculate a reliable mean time to failure.The estimate corresponds to more than a month of eight-hour days.Source: 2) Improved Robustness · p. 8

V. ACTUATOR CHARACTERIZATION

The actuator characterization measures fingertip workspace and forces for the two-compartment finger and torques for the bellows. The results characterize their output across inflation conditions, with low variation across repeated measurements.

V. ACTUATOR CHARACTERIZATION

  • Actuator characterization controls air pressure because, unlike air mass, pressure does not change in the presence of external forces and specifies maximum inflation states more suitably.The hand itself is controlled based on air mass.Source: §V · p. 8

A. Characterization of the Two-Compartment Finger

  • Approximately 0.06 N was the average standard deviation across five repetitions when measuring two-compartment finger forces at sampled fingertip positions.The measurements covered 36 combinations of chamber pressures from 0 to 250 kPa, with 50 kPa intervals.Source: A. Characterization of the… · p. 8

B. Characterization of the Bellow Actuator

  • Less than 0.1 Nm was the average standard deviation of bellow torque across five repetitions per inflation pressure and opening angle.The measurements indicate torque is proportional to air pressure and air-chamber area.Source: B. Characterization of the Bellow… · pp. 8, 10

VI. EVALUATION OF THE RBO Hand 3

The evaluation tests thumb opposability and grasp-posture versatility, alongside the hand’s ability to reproduce human grasping behaviors. It reports the maximum Kapandji score and repeated performance of all 33 GRASP taxonomy types.

VI. EVALUATION OF THE RBO Hand 3

  • The evaluation uses a mixing board to regulate actuator inflation mass, making the demonstrated behaviors easier for novice operators to reproduce.The hand’s reported behaviors include thumb opposition and GRASP taxonomy postures.Source: §VI · p. 10

A. Thumb Opposability

  • The hand achieved the highest possible Kapandji score; actuating the palm bellow to rotate the ring and little fingers toward the thumb was necessary to reach locations five through ten.The test was performed by replaying pre-recorded poses for the thumb to touch the ten locations.Source: A. Thumb Opposability · p. 10

B. Grasp Postures

  • The hand performed all 33 GRASP taxonomy types in three consecutive successful trials, integrating many actuated degrees of freedom with compliant passive shape adaptation.Its predecessor achieved 31 types; the soft finger pulps reduced a cavity associated with two failed grasps, and the more dexterous thumb enabled the scissors grasp.Source: B. Grasp Postures · pp. 10–11, Figure 11 · p. 11

C. Grasp Strength

The hand withstands directional pull-out forces and reproduces three frequent human tabletop grasping strategies. The strategy demonstrations include human-like top and edge grasps, while the flip grasp retains the underlying strategy but uses an atypical contact posture.

C. Grasp Strength

  • 39 N was required to pull the object out in the distal direction, compared with 23 N in the palmar direction; radial and ulnar pull-out required 30 N and 32 N, respectively.Standard deviation was below 3 N in each direction; the authors attribute the directional differences to which fingers, thumb, and palm form barriers to pull-out.Source: C. Grasp Strength · p. 11

D. Replicating Human Grasping Strategies

  • The three demonstrated strategies were selected because they were the most frequent tabletop strategies observed in humans, appearing in more than 78% of 3,400 prior-work trials.The strategies are top grasp, flip grasp, and edge grasp.Source: D. Replicating Human Grasping Strategies · p. 12
  • A single pre-recorded grasping synergy for each strategy was replayed in open-loop control, and compliance allowed the hand to adapt to objects and the tabletop.The hand could grasp a large variety of objects with these predefined motions; the authors connect this to reducing uncertainty and exploiting constraints.Source: D. Replicating Human Grasping Strategies · p. 12
  • The hand reliably replicated top, flip, and edge grasps; top and edge behavior closely resembled human behavior, while the flip grasp used the back of the fingers.For the flip grasp, simultaneous high inflation of both finger compartments produced the fingertip forces required for the strategy, which still rotated the object around an axis formed by the thumb.Source: D. Replicating Human Grasping Strategies · p. 12, Figure 12 · p. 12

E. In-Hand Manipulation

The RBO Hand 3 rotates diverse objects inside the hand using shared actuation patterns and compliant constraints. Its platform robustness is supported by repairable components, although lifetime estimates remain uncertain because too few failures have occurred.

E. In-Hand Manipulation

  • The hand rotates three objects about three axes using the same actuation pattern for each rotation type.Only a subset of actuators is inflated; the remaining compliant hand forms constraints that adapt to the object's motion.Source: E. In-Hand Manipulation · p. 12, Figure 13 · p. 13
  • The manipulations are described as robust funnels: actuated forces reconfigure constraints while compliant contact passively helps reduce uncertainty in object position and orientation.Most hand motion and constraint rearrangement is attributed to passive compliant constraints that try to retain their original pose.Source: E. In-Hand Manipulation · p. 13

F. RBO Hand 3 as a Platform

  • The authors estimate at least 300 hours of continuous use for two-compartment fingers and at least 60 hours for pouch actuators, but lack enough failures for reliable mean-time-to-failure analysis.They consider these estimates likely insufficient for a productized hand and therefore designed for easy repair.Source: F. RBO Hand 3 as a Platform · p. 13
  • Finger or pouch replacement takes at most ten minutes after the operation has been performed once or twice, using ordinary wrenches.Observed failure modes include displaced finger threads, damaged finger bases, and broken pouch seams.Source: F. RBO Hand 3 as a Platform · p. 13
  • The authors characterize the hand as an academic research platform with unprecedented availability through more robust components and quick repairs.This claim is framed for academic research, not productized use.Source: F. RBO Hand 3 as a Platform · p. 13
  • The bellow actuator withstands air pressures up to 300 kPa during frequent use, with an estimated continuous-use lifetime of at least 60 hours.The authors report insufficient failures for a reliable mean-time-to-failure estimate.Source: F. RBO Hand 3 as a Platform · p. 14

VII. LIMITATIONS AND FUTURE WORK

The section describes lower forces, sensorization, actuation precision, and modeling as challenges for soft robotic hands. It argues that compliant interactions and suitable control can support robust manipulation without relying on precision in the sense of accuracy and repeatability.

VII. LIMITATIONS AND FUTURE WORK

  • Intrinsic compliance is presented as improving robustness and facilitating dexterous grasping and manipulation.The authors argue that suitable control and planning can address many perceived shortcomings of soft-material robotics.Source: A. The Ambivalence of Soft Material… · p. 14

A. The Ambivalence of Soft Material Robotics

  • Soft pneumatic fingers tend to produce lower maximum forces than rigid counterparts, and high air pressures can break soft actuators.Embedding a rigid skeleton can provide kinematic stiffness and force transmission, while thicker actuator walls can withstand higher pressures.Source: A. The Ambivalence of Soft Material… · p. 14
  • Compliant-hand sensorization requires flexible sensors that preserve compliance, constraining sensor designs and often requiring complex fabrication or costly components.The authors report ongoing work on soft sensor technologies intended to have little effect on compliance or actuator design space.Source: A. The Ambivalence of Soft Material… · p. 14
  • Manufacturing differences and accumulated air-mass estimation errors make actuation less accurate and long-term hand movements difficult to repeat and predict.Tasks requiring high precision and repeatability are beyond the hand's stated scope.Source: A. The Ambivalence of Soft Material… · p. 14
  • Accurately modeling soft-hand interactions, especially material deformation, remains difficult, making traditional grasping and manipulation approaches inapplicable.The authors argue that robust compliant interactions, rather than accurate models and precise control, can provide a different kind of precision.Source: A. The Ambivalence of Soft Material… · p. 14

B. Sensorization of the RBO Hand 3

The paper reports grasping and manipulation despite open-loop control, while identifying sensory feedback as a route to more responsive behavior. It also describes pneumatic-control changes under development and concludes that compliant, actuated constraints support manipulation across varied objects.

B. Sensorization of the RBO Hand 3

  • The hand performs grasping and manipulation under pure open-loop control, while sensory feedback could further improve dexterity and enable responsive autonomous behaviors.The authors report work on pressure-sensing feedback control and other soft sensor technologies.Source: B. Sensorization of the RBO Hand 3 · p. 14
  • The authors are investigating liquid-metal strain sensing, piezoresistive fabric, flexible printed circuits, and acoustic sensing for proprioception and contact information.They plan to investigate whether feedback from these sensors improves dexterity when integrated into the hand.Source: B. Sensorization of the RBO Hand 3 · p. 14

C. Pneumatic Control

  • Binary pneumatic valves can cause movement tremor, which the authors found to have only minor effects on manipulation performance.They plan to use proportional valves and precise mass-flow sensors to improve movement smoothness and accuracy.Source: C. Pneumatic Control · p. 14

VIII. CONCLUSION

  • The hand combines many degrees of actuation with compliance in a robust, anthropomorphic design and is built modularly from low-cost, accessible materials.The conclusion identifies sensorization and improved pneumatic actuation among future work.Source: §VIII · p. 15
  • The hand achieves the highest possible Kapandji-test score, performs all 33 GRASP-taxonomy postures, and holds an object against external forces up to 39 N.The paper also reports replication of commonly observed human single-object tabletop grasping strategies.Source: §VIII · p. 15
  • The hand rotates different objects inside itself by reconfiguring compliant, actuated constraints, with the same actuation pattern across objects.The authors attribute robustness, versatility, and dexterity to compliance that absorbs contact dynamics and adapts passively to object and environment shapes.Source: §VIII · p. 15
Loading 2201.10883v1…