PaperlayerView exact revision on arXiv

Source-linked AI summary

Control Architecture for Safe Grasping of Fragile Objects Using a Coarse Position-Controlled Gripper

Marko Pavlic, Moritz Geier, Timo Markert, Darius Burschka

arXiv:2609.12737v2cs.RO

TL;DR

Robots must grasp objects with unknown properties safely, but existing force-control approaches often rely on known object properties. This paper combines fingertip force/torque sensing, online contact-force estimation, and adaptive admittance control; experiments report stable grasps across object characteristics while limiting contact forces to prevent damage.

  • Problem

    Robotic grasping in unstructured environments is challenging because object geometry, material properties, and contact conditions may be unknown or uncertain.

    Source: §I · p. 1

  • Method

    The approach uses fingertip force/torque sensing to estimate minimal contact forces online and combines this with an admittance controller for a position-controlled gripper.

    Source: §I · pp. 1–2

  • Results

    The controller achieved compliant, stable grasp acquisition across objects with different shapes, stiffnesses, and friction coefficients, and adjusted contact force in response to disturbances.

    Source: §V · p. 8

  • Takeaways & Limitations

    The approach limits contact forces to prevent object damage while maintaining stable grasps under the tested conditions.

    Source: §V · p. 8

Abstract

from arXiv · show

Robots are increasingly used in unstructured environments. The need for them to safely grasp unknown objects without damaging them becomes crucial. Humans achieve this by sensing and quickly responding by adjusting their grasping force. Similarly, effective grasp acquisition in robots requires compliant interaction strategies that can adapt to uncertain object properties and adjust to any instabilities during manipulation. We present a geometry-aware force/torque-based contact estimation method for a coarse position-controlled gripper, combined with an adaptive admittance controller for safe grasp acquisition. The desired contact forces are estimated online to keep stable contact with objects of unknown properties. This enables compliant and stable grasps while avoiding excessive forces. Experiments with objects of different sizes, shapes, stiffnesses, and weights show that the proposed algorithm not only prevents slippage but also applies minimal force to safely grasp an object without causing excessive deformation.

I. INTRODUCTION

Grasping unknown objects is difficult because their geometry, material properties, and contact conditions vary, while many controllers assume known object properties. The paper combines fingertip force/torque sensing, online contact-force estimation, and admittance control to acquire and maintain safe grasps.

I. INTRODUCTION

  • Unknown object properties and contact conditions challenge robotic grasping, while many existing force-control approaches are designed for specific, well-defined use cases.Such approaches often assume prior knowledge of object properties, limiting generalization to unstructured environments.Source: §I · p. 1
  • Pure position tracking during contact can produce unstable grasps or excessive forces that damage objects, and current control cannot provide high-fidelity force control.Compliant strategies layered over the internal position controller are therefore needed for safe interaction with unknown objects.Source: §I · p. 1
  • The approach estimates minimal contact forces online from fingertip force/torque sensing and uses an admittance controller to map force commands to motion commands.This avoids relying on prior knowledge of object properties when regulating grasp force.Source: §I · pp. 1–2
  • Tangential force components indicate incipient slip during acquisition, while changing contact conditions make grasp maintenance a continuing challenge after contact is established.The paper motivates continuously adapting the desired grasp force to those interaction signals.Source: §I · p. 1

II. RELATED WORK

Prior compliance and slip-control approaches often depend on object-specific knowledge, tuned parameters, or specialized sensing and hardware. The proposed approach instead uses fingertip force/torque measurements and continuous force adaptation on a coarse position-controlled gripper.

II. RELATED WORK

  • Conventional active-compliance frameworks commonly rely on manually specified target forces, object-property knowledge, or tuning for specific object classes.The cited approaches include force sensing layered on position control, fuzzy-logic admittance tuning, and tuned operational-space compliance.Source: §II · p. 2
  • Slip-regulating prior work used custom two-finger hardware with a single-finger force sensor and dedicated optical slip sensing, while reported scope and limitations varied.Ding et al. demonstrated reduced deformation on delicate objects; Al-Mohammed et al. excluded highly deformable objects, and O’Toole et al. reported slow observer convergence and model–plant mismatch sensitivity.Source: §II · p. 2
  • The proposed method continuously updates desired grasp force from tangential force feedback, allowing adjustment before slippage rather than only after a physical slip event.It uses distributed force/torque sensing at every contact point and does not require the rigidity assumption made by Al-Mohammed et al. [8].Source: §II · p. 2
  • The approach combines measured fingertip force/torque, contact-state estimation, geometry-aware force distribution, and adaptive control without manual parameter tuning.It runs on a standard coarse-position-controlled industrial gripper; direct experimental comparison with related work is infeasible because hardware and sensing differ.Source: §II · p. 2

III. FORCE/TORQUE-BASED CONTACT ESTIMATION AND CONTROL

  • The framework pipeline combines signal filtering, contact-geometry estimation, geometry-aware force distribution, and adaptive admittance control to regulate contact forces.The goal is minimum force to prevent slipping while maintaining the deformation constraint, using fingertip-mounted force/torque sensors.Source: §III · p. 2

II. RELATED WORK

  • Table I summarizes structural differences among force-regulated grasping approaches; it is not a direct experimental performance comparison.The proposed approach seeks to infer control behavior from contact measurements rather than object-specific parameterization.Source: §II · p. 2, Table I · p. 3

A. Control Framework

The control framework converts fingertip force/torque measurements into contact-state and desired-force estimates, then adjusts gripper position through admittance control. Its sensor signals are low-pass filtered to suppress noise while preserving contact dynamics.

A. Control Framework

  • Fingertip force/torque measurements feed a contact estimator and grasp-force estimator, whose desired-force error drives admittance-based position compensation.The compensation is added to the gripper reference position; the initial reference is fully closed and updated upon contact.Source: A. Control Framework · p. 3

B. Signal Filtering

  • The measured wrench comprises contact force F and torque τ about the sensor origin, and these measurements are subject to high-frequency noise.The framework filters signals to support reliable contact detection.Source: B. Signal Filtering · p. 3
  • At a 1000 Hz sensor sampling rate, a 50 Hz cutoff was selected to suppress noise while preserving the relatively slow grasp-force dynamics.The same low-pass filtering treatment is assumed for measurements throughout the remainder of the paper.Source: B. Signal Filtering · p. 3

C. Geometry Constrained Contact Point Estimation

Contact estimation combines fingertip force/torque measurements with a hemispherical geometry model, without assuming known object mass or friction. The estimated interaction is represented through contact location and normal and tangential force components.

C. Geometry Constrained Contact Point Estimation

  • The method adapts grasping force directly from measured contact interactions without assuming known object mass or friction coefficient.After contact is established, wrench measurements are used to estimate the fingertip–object contact point.Source: C. Geometry Constrained Contact Point… · p. 3
  • Contact is detected above a force threshold; at initial low-force contact, the contact point is approximated using a force-only estimate of the normal projected onto the hemispherical fingertip surface.Once force exceeds a second switch threshold, torque measurements become informative enough to refine contact localization.Source: C. Geometry Constrained Contact Point… · p. 3
  • With torque-informed localization, the force line intersects the hemispherical surface at two candidate points, from which the physically valid contact point is selected.The measured force is constrained to act inward, and the contact normal is the outward surface normal at the estimated contact point.Source: C. Geometry Constrained Contact Point… · p. 4
  • The estimator decomposes contact force into normal and tangential components using a hard-finger point-contact model.Although the rubber cover forms a finite patch, its approximately 1 mm thickness is small relative to the 12 mm fingertip radius, so its torsional moment is described as negligible compared with resolved forces.Source: C. Geometry Constrained Contact Point… · p. 4, Figure 3 · p. 4

D. Slip–Driven Normal Force Adaptation

The controller adapts a baseline normal force from tangential-force changes, then adds a term proportional to tangential load to set the desired grasp force. This estimates slip-preventing force online without assuming object mass or friction.

D. Slip–Driven Normal Force Adaptation

  • After contact, the controller adapts grasp force directly from measured interactions rather than assuming object mass or friction.The force adaptation is used during lifting, after initial normal and tangential force estimates are available.Source: D. Slip–Driven Normal Force Adaptation · p. 4
  • An increase in tangential-force magnitude indicates growing shear that typically precedes slip, prompting the slowly varying baseline normal force to increase.The baseline remains constant under steady-contact conditions; its adaptation rate is set by a fixed gain independent of object mechanics.Source: D. Slip–Driven Normal Force Adaptation · p. 4
  • The desired normal force combines an adaptive baseline with a tangential-load term, coupling grasp force to object loading without estimating mass or friction.The desired force is bounded above to avoid overloading the F/T sensor and gripper actuators.Source: D. Slip–Driven Normal Force Adaptation · pp. 4–5

E. Adaptive Admittance Control

An outer admittance loop regulates finger closure by converting the difference between desired and measured normal force into incremental position commands for the gripper.

E. Adaptive Admittance Control

  • The admittance dynamics map desired-minus-measured normal force into an incremental finger-displacement command.The command is applied to the current finger position to produce the commanded position reference.Source: E. Adaptive Admittance Control · p. 5
  • The admittance loop generates a position reference for the gripper’s internal position controller, whose discrete command range is 0–255.Positive increments close the gripper and negative increments open it; Δp ranges from −255 to 255.Source: E. Adaptive Admittance Control · p. 5

F. Stability Analysis

The stability analysis models contact as an elastic spring and projects gripper motion onto the contact normal. Under the stated positivity and timescale conditions, the admittance equilibrium is stable and its applied force does not exceed the desired force.

F. Stability Analysis

  • Contact is modeled as an elastic spring, with environment mass and damping neglected; gripper displacement produces indentation only through its projection onto the contact normal.The projected indentation is δ = nz Δp, where nz = cos θ ∈ (0, 1].Source: F. Stability Analysis · p. 5
  • With the force reference treated as a slow, constant set-point, the inner closed loop has a unique equilibrium at Δp⋆ = fn,d/(K + Ke nz).The controller objective is convergence to the rendered displacement equilibrium, not exact force tracking, because K > 0 leaves a steady-state offset.Source: F. Stability Analysis · p. 5
  • The Lyapunov energy derivative is nonpositive, and the analysis concludes that displacement error and its rate converge to zero.The argument uses positive definiteness and radial unboundedness of V together with the standard Barbalat/LaSalle result.Source: F. Stability Analysis · p. 5
  • For any positive contact stiffness, the analysis bounds the settled normal force by the desired force, providing an intrinsic force offset.Positive normal alignment follows from unilateral contact on the leading hemisphere of the spherical fingertip.Source: F. Stability Analysis · p. 5
  • The stiffest object sets the binding timescale condition: the admittance bandwidth is kept well below the internal position-controller bandwidth.The mass M is sized so that ωadm(Ke,max) ≪ ωpos.Source: F. Stability Analysis · p. 5

IV. EXPERIMENTS AND RESULTS

Experiments use a bimanual platform with a position-controlled gripper instrumented at the fingertips, testing deformable household items and stiff objects. Controller parameters are tuned once and held constant across objects.

IV. EXPERIMENTS AND RESULTS

  • The experiments use a bimanual platform with a position-controlled Robotiq 3-Finger Gripper and fingertip six-axis F/T sensors.The gripper can set rough force limits from 15 N to 60 N.Source: §IV · p. 5
  • An aluminium-sphere fingertip with a thin rubber coating provides the test setup’s human-finger-like contact surface.The fingertip integration is inspired by.Source: §IV · p. 5
  • The test objects include delicate, deformable household items of varying weights as well as stiff objects.The object suite is described as demonstrating the approach’s versatility.Source: §IV · p. 5
  • Controller parameters were tuned once and held constant across all experiments and object types.Source: §IV · p. 5

A. Safe Grasp Acquisition with Unknown Objects

The proposed controller combines fingertip force/torque sensing with adaptive control to grasp delicate objects without visible deformation. On the same actuation platform, it regulates force below the built-in controller’s minimum threshold, which damages delicate objects.

A. Safe Grasp Acquisition with Unknown Objects

  • Trials on a bimanual system used a Robotiq 3-Finger Gripper with fingertip-integrated miniature 6-axis F/T sensors, comparing its internal grasping control with the proposed framework on the same actuation platform.Source: A. Safe Grasp Acquisition with Unknown… · p. 5, Figure 4 · p. 6
  • Standard Control visibly deforms delicate objects, whereas Proposed Control follows their shape without noticeable deformation.Fig. 5 compares the gripper’s internal control with the proposed algorithm.Source: A. Safe Grasp Acquisition with Unknown… · p. 6, Figure 5 · p. 6
  • Proposed Control regulates grasp force well below the 15 N minimum threshold, while standard current-based control damages delicate objects.The standard setting settles slightly above 15 N; the proposed control uses fingertip F/T feedback on the same actuation platform.Source: A. Safe Grasp Acquisition with Unknown… · p. 6

B. Force Adaptation During Lift

During lifting, the adaptive controller monitors contact forces and increases grasp force when added weight or slippage requires it. The same controller settings worked across deformable and stiff objects, with minimal deformation reported for the tested deformable objects.

B. Force Adaptation During Lift

  • For the plastic bottle, paper cup, and sponge, the proposed control applied forces an order of magnitude smaller than standard control.The comparison concerns the grasping forces shown in Fig. 6.Source: Figure 6 · p. 7
  • The controller lifted supported objects from contact using a 0.6 N minimum force, adapting to slippage or increased tangential force during lifting.The controller continuously observes F/T measurements and adjusts force as needed.Source: B. Force Adaptation During Lift · pp. 6–7
  • Empty deformable objects were lifted with little deviation from initial contact force, while added weight prompted greater force.The lift began at approximately 5.0 s after contact was established at approximately 0.5 s.Source: B. Force Adaptation During Lift · p. 7
  • The paper cup and plastic bottle were only minimally deformed during lifting, which the authors considered acceptable given the small finger-contact areas.The bottle was left open to make deformation easier and the task more challenging.Source: B. Force Adaptation During Lift · p. 7
  • The same controller parameters lifted the stiff objects; the steel mug slipped slightly, then the controller increased force and recovered it without dropping it.The tested objects included a coffee container with added weight and a steel mug filled with water.Source: B. Force Adaptation During Lift · p. 7, Figure 8 · p. 7
  • Coarse gripper position resolution produces stepwise force profiles, with larger force steps for stiffer objects and settling transients from the rubber fingertip cover.The cover’s viscoelastic response adds compliance and damping at contact.Source: B. Force Adaptation During Lift · p. 7

C. Force Adaptation During Disturbance

After establishing a safe grasp, the controller adapted contact force during disturbances and prevented tested objects from dropping. The paper concludes that geometry-aware F/T contact estimation and adaptive control support stable grasps across varying loads and object properties.

C. Force Adaptation During Disturbance

V. CONCLUSION

  • Contact-point position on the finger, together with normal and tangential forces, supports desired-force computation without prior knowledge of object properties.The approach combines geometry-aware F/T contact estimation with an admittance controller.Source: §V · p. 8
  • The experiments showed compliant, stable grasps across different shapes, stiffnesses, and friction coefficients, with force adjustment supporting grasp maintenance under disturbances.The system analyzes contact relations through its finger geometry design to maintain stable grasps under varying loads.Source: §V · p. 8
Loading 2609.12737v2…