PaperlayerView exact revision on arXiv

Source-linked AI summary

ARTiS: An Adaptive Robotic Gripper for Enhanced Tool Manipulation in Disassembly Applications

Roman Mykhailyshyn, Yukiyasu Domae, Kensuke Harada

arXiv:2609.03362v2cs.RO

TL;DR

Robotic disassembly requires grippers that can hold tools securely while using them, beyond simply grasping them. ARTiS combines an adaptive jamming palm with reconfigurable fingers and pose-specific fingertips; reported success rates vary across grasping, reorientation, and tool use.

  • Problem

    Tool grasping alone is insufficient for disassembly, where grippers also need adaptability, robustness, and fixation during tool use.

    Source: §I · pp. 1–2

  • Method

    ARTiS uses a cooperative tool-fixation architecture combining an adaptive jamming palm with reconfigurable fingers and pose-specific fingertips.

    Source: §I · p. 2

  • Results

    97% grasping, 85% grasping and tool reorientation, and 68% using different tools.

    Source: §VI · p. 14

  • Takeaways & Limitations

    The paper frames robotic disassembly tool handling as a tool-fixation, tool-use, and human-teaching problem.

    Source: §VI · p. 14

    Tool oscillations can disrupt palm contact and stiffness control; the 3D-printing fabrication method also constrains gripper dimensions.

    Source: §VI · p. 14

Abstract

from arXiv · show

Grasping and holding tools while using them presents a considerable challenge not only for robots but also for humans. Such a challenge is particularly noticeable in processes involving assembly and disassembly, where efficiency and consistency depend on performing rapidly adaptive tasks. Nonetheless, contemporary robotic grasping technologies that can securely manipulate tools during operation frequently have significant constraints. In this paper, introduce ARTiS (Adaptive Robotic Tool Gripper in Disassembly Systems), a novel gripper that combines the adaptability of soft grippers, the dexterity of anthropomorphic hands, and the robustness of rigid mechanisms with a soft palm and fingertips. This unique combination makes it possible to hold tools securely in a variety of situations through using active jamming in the palm and fin-ray adaptation in fingertips. Furthermore, high finger dexterity is achieved through the seven degrees of freedom design, which enables the fingertips to orient to any surface, both for automated solutions and collaborative tasks. A comprehensive evaluation was conducted using a range of conventional disassembly tools to assess the gripper's compliance, durability, and functional versatility. More information, hardware instructions, and videos at https://romanmykhailyshyn.github.io/artis/

I. INTRODUCTION

Robotic disassembly needs grippers that can adapt to different tools and hold them securely during use, not merely grasp them. ARTiS combines an adaptive palm and reconfigurable fingers, with tool-use and collaboration evaluations described alongside the design.

I. INTRODUCTION

  • Existing grippers may grasp tools but lack the combined dexterity, adaptive contact, and active palm fixation needed to hold tools during use.An adaptive three-finger gripper can let a screwdriver spin without a locking palm or larger frictional contact area.Source: §I · pp. 1–2
  • ARTiS pairs an adaptive jamming palm with reconfigurable fingers and pose-specific fingertips in a cooperative tool-fixation architecture.The contributions include a cup-shaped palm with a cutout for perpendicular tool holding and 3D-adaptive fingertips.Source: §I · p. 2
  • The evaluation includes grasping, fixing, reorienting, and operating screwdrivers, hammers, and drills, plus object-generalization and human teaching-mode studies.The teaching mode lets an operator configure and record gripper states for repeated collaborative tasks.Source: §I · p. 2

II. RELATED WORKS

  • Three-finger design alone does not make a gripper suitable for hand-tool use; Table I compares structures and disassembly-relevant mechanical properties.The comparison covers robustness, grasping versatility, task versatility, and tool use.Source: §II · p. 2, Table I · p. 3

1) Insufficient Finger Reorientation and Fingertip Adaptation:

  • Grippers with at most 2 DoF per finger cannot adequately control finger position or orientation; ARTiS aims for continuous reorientation and adaptive fingertips.The stated goal is to let fingers adjust tool-tip position or orientation while fingertips adapt contact.Source: 1) Insufficient Finger Reorientation… · pp. 2–3

2) Insufficient Tool Fixation:

No existing gripper satisfies all properties identified for robotic disassembly; ARTiS combines continuously oriented fingers, adaptive fingertips, and an active jamming palm. Its design addresses tool fixation, grasping versatility, and task versatility.

2) Insufficient Tool Fixation:

  • Most surveyed three-finger grippers lack a palm that can fix tools or provide a surface against which to press them, so the palm must carry the main load.The paper distinguishes tool fixation from palm designs aimed primarily at conforming to objects for pick-and-place stability.Source: 2) Insufficient Tool Fixation · p. 3
  • No surveyed gripper satisfies all required disassembly properties; ARTiS combines continuously oriented fingers, adaptive fingertips, and an active jamming palm for tool use.The paper identifies shortcomings in existing designs including limited palm size or shape, finger-opening constraints, fragile fingers, or insufficient fingertip force.Source: 2) Insufficient Tool Fixation · p. 3

A. Justification of Gripper Properties

  • A tool-use gripper must adapt to tool surfaces, fix the tool for power holding, adjust its position, and support simple control strategies for different tasks.The human-hand model combines palm adaptation and fixation with fingertip adaptability for contact, friction, and tool-tip control; the two grasping methods are power sphere and power wrap.Source: A. Justification of Gripper Properties · p. 4, §III · p. 4

2) Grasping versatility:

  • Grasping versatility means handling tools and parts across sizes and shapes; three-finger designs support sphere and wrap grasps but need fingertip adaptation and added degrees of freedom for asymmetric surfaces.The stated target is a 100% grasping success rate for more than half of each tool type, while transmitting application-dependent torque during use.Source: 2) Grasping versatility · p. 4

3) Task Versatility:

  • Task versatility covers grasping, tool use, breaking, and part holding or fixing, with automated and human-collaborative disassembly as distinct application settings.The stated application goal is to reduce the time needed to unscrew three screws and perform a task during human interaction.Source: 3) Task Versatility · p. 4

IV. DESIGN CONCEPT

  • ARTiS uses seven motors and a separate control box to control its electrical and pneumatic components.The design decisions are intended to provide robustness and versatility.Source: §IV · pp. 4–5

A. Main Design

ARTiS combines articulated fingers, a jamming palm, and two adaptive fingertip designs. Its evaluation examines the gripper’s tool-use capabilities during disassembly of an automotive air-conditioning unit.

A. Main Design

  • ARTiS’s four-bar finger mechanism clears the jamming palm’s working area, while finger orientation and rotation provide one 2-DoF finger and two 3-DoF fingers.The driven link rotates through 0–170 degrees; each finger’s orientation motor spans −90 to 90 degrees, and the two movable fingers rotate symmetrically from 0 to 70 degrees relative to the gripper axis.Source: A. Main Design · p. 5
  • A vacuum-controlled jamming palm fixes tools by changing its density, and its slot allows tools lying on a table to be inserted for power-wrap grasping.The palm is rubber, 100 mm in diameter and 60 mm high, and uses glass-bead jamming elements; the experiments supply 400 kPa compressed air and generate −88 kPa palm pressure.Source: A. Main Design · pp. 5–6

B. Highly Adaptive Fingertips

  • Fingertip 1 uses linked, round Fin-Ray elements to adapt across two planes, while fingertip 2 uses oppositely deployed Fin-Rays to conform to semi-cylindrical tools at different orientations.A rubber ring couples fingertip 1’s released elements, so deformation of one changes the orientation of the others; fingertip 2’s design provides maximum contact area in two folds when finger orientation changes.Source: B. Highly Adaptive Fingertips · p. 6

V. RESULTS

  • The evaluation tests tool use and holding methods during disassembly of an automotive air-conditioning unit.The experiments address adaptability and robustness, grasping and holding versatility, and performance across a range of tasks.Source: §V · p. 6

A. Evaluating Robustness

ARTiS’s combined palm-and-finger hold outperforms the LEAP Hand on screwdriver-tip force in the reported comparisons, while the study also assesses fingertip robustness and slip behavior. The tests cover two screwdriver shapes and multiple holding methods.

1) Fingertip:

  • 0.5 N maximum force dispersion was reported for fingertip 1 across 180 trials; its maximum force fell by an average of 22% as object orientation increased between tested angles.For fingertip 2, force dispersion also remained within 0.5 N when object orientation changed; at 10 mm displacement, even its smallest contact area provided 18 N, and the authors report hundreds of uses without performance loss.Source: 1) Fingertip · p. 7

2) Holding/Force Study:

  • Finger placement changes the limiting slip mode: grasping near the palm favors rotational slip, while grasping near the tool tip makes translation dominant.The analysis compares tool-tip torque with the fingers’ translational and rotational resistance; slipping begins when applied torque meets or exceeds rotational resistance.Source: 2) Holding/Force Study · p. 8
  • 5.4 and 7.8 times greater force at the tool tip is reported for ARTiS than LEAP Hand during two-finger sphere holding; ARTiS also produces 20.4% and 32.8% more force than LEAP’s three-finger wrap hold.ARTiS retains tool-tip control, which the LEAP Hand loses in the reported comparison.Source: Figure 11 · p. 8, 2) Holding/Force Study · p. 9
  • Tests used spherical and conical screwdrivers, repeated each holding method 30 times, and compared four ARTiS palm-and-finger configurations with two LEAP Hand grasps.The force was applied perpendicular to the tool tip; the study identifies two-finger sphere holding as the most unfavorable configuration for slipping.Source: 2) Holding/Force Study · pp. 8–9
  • More than 90% of the average screwdriver-holding contribution comes from the palm, and the combined palm-and-finger method achieves the highest reported robustness.The combination’s force curve has jumps associated with rib-to-rib translation slip; finger-only methods show rotation slip without jumps.Source: 2) Holding/Force Study · p. 9

B. Evaluating Grasping Versatility

ARTiS was evaluated on grasping, releasing, reorienting, and using disassembly tools, screwdriver torque, drilling, and grasping objects beyond the original tool set. Results show strong grasping and torque performance, alongside specific limits from smooth handles and tool vibration.

B. Evaluating Grasping Versatility

  • 97% of tool-grasping trials succeeded overall, but screwdriver №5 succeeded in only 57% of jamming-and-lifting trials and 0% of grasping-with-release trials.Its smooth, protrusion-free handle slips in the palm; increasing palm depth could improve contact but would add weight and design complexity.Source: B. Evaluating Grasping Versatility · p. 10
  • 68% success was achieved for reorientation/use and release, and the success rate across all three tool-use stages was 85%.The reported 68% stage rate reflects the difficulty of maintaining tool contact during operation.Source: B. Evaluating Grasping Versatility · p. 10
  • №9 produced the highest tightening torque, reaching 4.79 ± 0.13 Nm maximum and 4.77 ± 0.12 Nm at 10 s; №4 was second in maximum torque at 2.64 ± 0.09 Nm.№9 exceeded №4 by approximately 81.6% in maximum torque and 84.6% in torque at 10 s.Source: B. Evaluating Grasping Versatility · pp. 10–11, Table III · p. 11
  • Across 30 unscrewing trials per screw, reported mean peak torque was 1.71 ± 0.15 Nm for screw 2 and 1.33 ± 0.22 Nm for screw 1, and screw 2 peaked later.Time to peak was 2.36 ± 0.23 s for screw 2 and 1.29 ± 0.20 s for screw 1.Source: B. Evaluating Grasping Versatility · p. 11
  • Screwdriver №2 reached 1.53 ± 0.11 Nm at 9.36 ± 0.7 s, showing that fast torque saturation did not mean greater torque capacity.Across tested designs, №1, 2, 4, and 9 produced the highest torques in the test.Source: B. Evaluating Grasping Versatility · p. 11
  • ARTiS achieved 95.27% success across 43 selected YCB objects; tool and food subsets reached 97.95% and 97.41%, respectively.Kitchen objects reached 89.17%; thin or flat shapes and displaced centers of mass were challenging. These tests covered grasping and lifting, not generalized tool use.Source: Table V · p. 12, B. Evaluating Grasping Versatility · p. 12
  • Power holding produced steadier torque than precise holding while drilling plastic fasteners, which showed significant fluctuations.The authors attribute the difference to greater palm contact and a shorter force lever with power holding.Source: B. Evaluating Grasping Versatility · p. 12

C. Evaluating Task Versatility

ARTiS was adapted to fix a workpiece for collaborative disassembly, with participants teaching and recording gripper configurations. In a three-screw task, robot-assisted fixation reduced total time even after setup time was included, while its advantage depends on the number of fasteners.

C. Evaluating Task Versatility

  • Including the fixed 8 s setup, ARTiS-assisted fixation reduced three-screw task time by 5.89 s on average, or approximately 13.77%, compared with using both hands.The paired t-test found a significant reduction, t(11) = 4.02, p = 0.0020; the Wilcoxon test also found a significant difference, p = 0.00098.Source: Table VI · p. 13, C. Evaluating Task Versatility · p. 13
  • ARTiS teaching mode lets a person position and configure the gripper to fix a part on the table, after which the person can use both hands to unscrew screws.Across 12 subjects, the average fixation time after training was 8 seconds.Source: C. Evaluating Task Versatility · pp. 12–13
  • The fixation strategy becomes advantageous at two or more bolts because per-bolt savings accumulate against the fixed setup cost; the authors estimate a break-even point of about 1.73 bolts.The expected time saving is expressed as ∆T(n) = 4.63n − 8.00.Source: C. Evaluating Task Versatility · p. 13
  • Power-wrap holding also supports impact and levering actions, including using a hammer to strike a screwdriver or levering composite fasteners with a screwdriver.The paper presents these as examples of tool use beyond standard disassembly operations.Source: Figure 18 · p. 13, C. Evaluating Task Versatility · p. 14

VI. CONCLUSION AND FUTURE WORK

ARTiS combines an adaptive jamming palm, pose-specific fingertips, and active finger reconfiguration for grasping, fixing, reorienting, and using tools. Its use-stage performance is limited by tool oscillations, and generalized functional tool use with unseen objects was not evaluated.

VI. CONCLUSION AND FUTURE WORK

  • ARTiS treats robotic tool handling as a combined grasping, fixation, tool-use, and human-teaching problem, not only a grasping problem.The conclusion describes its approach as mechanically simple but functionally versatile for human-designed tools in robotic disassembly.Source: §VI · p. 14
  • YCB experiments demonstrated grasp acquisition and lifting across diverse objects, but did not evaluate generalized functional tool use.The authors identify generalized YCB tool use as future work.Source: §VI · p. 14
  • Tool oscillations can disrupt palm contact and stiffness control; the gripper’s 3D-printing fabrication also constrains its dimensions.The authors suggest stiffer palm materials, greater vacuum, or active side stops for oscillation-related limitations, and different manufacturing and structural optimization for dimensions.Source: §VI · p. 14

A. Control Algorithm and Autonomous Operation

ARTiS supports autonomous operation through a Python/ROS 2 API and a sequence that combines vision, wrist force-torque feedback, palm jamming, and fingertip control. The controller seats and verifies the tool before use, tracks its tip during the task, and releases it in a controlled sequence.

A. Control Algorithm and Autonomous Operation

  • The Python/ROS 2 API exposes high-level control of joints J0–J6 and the jamming-palm valve, including replay of taught grasp sequences.The command vector includes seven joint positions and a binary palm-jamming state.Source: A. Control Algorithm and Autonomous… · p. 16
  • Autonomous tool handling combines vision-based pose and contact-region detection with wrist force-torque feedback to regulate seating and verify stable contact before jamming.The seating criteria use desired contact force Fd and bounds on force error and misalignment torques Mx and My.Source: A. Control Algorithm and Autonomous… · p. 16
  • The autonomous sequence detects and approaches the tool, jams the palm, verifies the grasp, closes the fingertips, tracks and corrects tool-tip pose during use, then places and releases the tool.The paper summarizes these steps in Algorithm 1.Source: A. Control Algorithm and Autonomous… · pp. 16–17

B. Teaching-Based Gripper Programming

Teaching-based programming lets an operator configure and record a tool grasp as a reusable ARTiS sequence, which can be replayed for repeated execution or used as imitation-learning data.

B. Teaching-Based Gripper Programming

  • Operators manually configure a tool grasp and record the ARTiS joint positions, jamming-palm state, timing, and optional sensing data as a reusable sequence.The sequence is saved as a tool-grasping primitive for replay or robot imitation learning.Source: B. Teaching-Based Gripper Programming · p. 17
  • Recorded steps include the gripper configuration and palm state; vision-estimated tool pose and wrist wrench are omitted when those sensors are unavailable.The gripper-state-only recording remains possible without vision or force-torque sensing.Source: B. Teaching-Based Gripper Programming · p. 17
  • During playback, ARTiS adapts the taught approach to the observed tool pose when vision is available, then executes the stored steps in sequence.Available wrist force-torque sensing can regulate palm contact, while detected unsafe contact, slip, or instability stops playback.Source: B. Teaching-Based Gripper Programming · p. 17

APPENDIX B GRIPPER ABLATION USING SCREWDRIVERS

The screwdriver ablation compares combined palm-and-finger performance with isolated contacts and finds translation, rather than rotation, governs slip in both tested configurations. The longer moment-arm geometry shows evidence of beneficial nonlinear palm–finger interaction.

APPENDIX B GRIPPER ABLATION USING SCREWDRIVERS

  • The ablation compares combined Palm+Finger performance with Palm-only and Finger-only configurations across slip-mode contributions and safety factors.Table VII is described as covering slip mechanics, mode contributions, and safety factors.Source: Appendix B · p. 18, Table VII · p. 18
  • For screwdriver 2, the combined configuration slips at 0.603 Nm, equivalent to 5.022 N translational friction; screwdriver 4 slips at 0.705 Nm, equivalent to 5.879 N.The tested tools differ in length, while finger position, palm radius, and tool radius are held constant.Source: Appendix B · pp. 17–18
  • Both screwdrivers fail by translation: screwdriver 2 requires 5.022 N against a rotational scale of about 20.7 N, and screwdriver 4 requires about 5.88 N against about 23.3 N.These comparisons identify translational slip as the governing mode in each tested configuration.Source: Appendix B · p. 18
  • ηglobal = 0.111 is reported as the conservative global value, while screwdriver 4 has ηabl T ≈1, indicating that linear addition predicts its translational capacity well.The longer moment-arm geometry shows beneficial nonlinear interactions that are less pronounced for the shorter screwdriver.Source: Appendix B · p. 18

APPENDIX C GENERALIZATION EVALUATION USING YCB OBJECTS

ARTiS was tested on YCB objects to assess grasping beyond the studied disassembly tools, using grasp poses chosen according to object geometry. Results show high success across many object types, with lower performance on difficult kitchen geometries.

APPENDIX C GENERALIZATION EVALUATION USING YCB OBJECTS

  • Objects with handles or protrusions were difficult to grasp at an ideal position and orientation, while boxes and cans were grasped by an edge to permit an encompassing hold.The palm could not wrap around boxes or cans when grasping at their middle.Source: Appendix C · p. 19
  • Sharp or puncturing YCB objects, including knives and needles, were excluded to prevent damage to the jamming palm.Per-object success rates and 95% Wilson score confidence intervals are reported in Table VIII.Source: Appendix C · p. 19, Table VIII · p. 19
  • 33 of 43 tested YCB objects were grasped with 100% success over 30 trials, and 36 objects reached at least 90% success.Tool and food subsets had the highest success rates, 97.95% and 97.41%, while the kitchen subset reached 89.17%.Source: Appendix C · p. 19
  • ARTiS maintained reliable grasping across varied object types, while remaining failures mainly involved limited contact area or unfavorable load distribution on the jamming palm.The reported object types include boxes, cans, fruit-like objects, cups, bowls, utensils, hand tools, and small tool-related components.Source: Appendix C · p. 19

APPENDIX TORQUE-BASED PERFORMANCE EVALUATION

The torque-based evaluation characterizes screwdriver–palm interaction through strength, stability, controllability, and slip-related measures. Results show trade-offs: strong torque transfer can accompany reduced controllability, while weaker interactions may be more predictable.

APPENDIX TORQUE-BASED PERFORMANCE EVALUATION

  • Screwdriver 9 reaches the highest maximum torque at 4.80 N·m, while screwdrivers 7 and 8 produce the lowest torque at approximately 0.41 N·m.The passage attributes the lower torque to weak geometric and frictional interaction.Source: APPENDIX TORQUE-BASED PERFORMANCE… · p. 20
  • Screwdriver 3 reaches maximum torque fastest at approximately 5 s, whereas screwdrivers 7 and 8 require more than 9 s.The passage associates shorter slip time with more rapid torque loss and longer times with gradual torque accumulation.Source: APPENDIX TORQUE-BASED PERFORMANCE… · p. 20
  • Screwdrivers 3–6 have the lowest angular error and smoother, more predictable torque growth, while screwdrivers 7–9 have the highest error.The passage links the latter pattern to insufficient or strong grip.Source: APPENDIX TORQUE-BASED PERFORMANCE… · p. 20
  • Screwdriver 4 has the highest torque-ramp slope, S = 10.20, indicating strong grip with rapid, less controllable torque increase.Screwdrivers 3 and 5 have moderate slopes of 5–6, balancing engagement and user control.Source: APPENDIX TORQUE-BASED PERFORMANCE… · p. 20
  • Screwdriver 9 provides the strongest torque transfer but has the largest slip and reduced controllability; screwdrivers 7 and 8 have weak interaction but predictable behavior.Screwdriver 4 combines excellent friction with a small controllability margin.Source: APPENDIX TORQUE-BASED PERFORMANCE… · p. 20
Loading 2609.03362v2…