Source-linked AI summary
Maintaining Grasps within Slipping Bound by Monitoring Incipient Slip
Siyuan Dong, Daolin Ma, Elliott Donlon, Alberto Rodriguez
TL;DR
Robotic gripping-force control has been limited by the lack of tactile feedback for detecting incipient slip. This paper compares GelSlim marker motion with a rigid-body estimate to detect slip, reporting 86.25% detection accuracy across 240 trials with 10 daily objects. Closed-loop bottle-cap tasks demonstrate how the feedback can inform gripping-force adjustments and stopping.
Problem
Robotic systems have limited gripping-force control, largely because tactile sensors that provide incipient-slip feedback are lacking.
Source: §I · p. 1Method
The method detects incipient slip by comparing GelSlim's measured marker displacement field with a 2D rigid-body estimate of the contact patch motion.
Source: §I · p. 2, §III · p. 2Results
86.25% detection accuracy was achieved in 240 slip experimental trials with 10 daily objects.
Source: §VI · p. 6Takeaways & Limitations
In closed-loop bottle-cap screwing and unscrewing, slip feedback informs the robot when to increase gripping force and when to stop.
Source: §I · p. 2For flat, smooth objects, larger gripping force is needed to generate a contact signal strong enough for the method to work reliably.
Source: A. Incipient Slip Detection Accuracy · p. 5
Abstract
from arXiv · showhide
In this paper, we propose an approach to detect incipient slip, i.e. predict slip, by using a high-resolution vision-based tactile sensor, GelSlim. The sensor dynamically captures the tactile imprints of the contact object and their changes with a soft gel pad. The method assumes the object is mostly rigid and treats the motion of object's imprint on sensor surface as a 2D rigid-body motion. We use the deviation of the true motion field from that of a 2D planar rigid transformation as a measure of slip. The output is a dense slip field which we use to detect when small areas of the contact patch start to slip (incipient slip). The method can detect both translational and rotational incipient slip without any prior knowledge of the object at 24 Hz. We test the method on 10 objects 240 times and achieve 86.25% detection accuracy. We further show how the slip feedback can be used to monitor the gripping force to avoid slip with a closed-loop bottle-cap screwing and unscrewing experiment with incipient slip detection feedback. The method was demonstrated to be useful for the robot to apply proper gripping force and stop screwing at the right point before breaking objects. The method can be applied to many manipulation tasks in both structured and unstructured environments.
I. INTRODUCTION
Incipient-slip feedback could help robots regulate grip force, but robotic tactile systems have been limited in their ability to provide it. The paper proposes a GelSlim-based detector and illustrates its use for rotational and translational slip during object handling.
I. INTRODUCTION
- Robotic systems' limited gripping-force control is linked largely to the lack of tactile sensors that provide incipient-slip feedback.The introduction describes potential uses in avoiding a slipping bottle or dropped tool and preventing object breakage from excessive grip force.Source: §I · p. 1
- The examples show rotational slip when a screwdriver collides with the ground and translational slip when a bottle cap reaches the end of its threads.The figure sequences progress from no shear or slip, through shear without slip, to full slip.Source: Figure 1 · p. 1, §I · p. 1
- The GelSlim sensor captures deformation of a gel-based skin to measure contact position and local object texture.A fabric layer protects the gel and enhances the contact signal with its own texture.Source: §I · p. 1
- The method compares real marker motion with a no-slip 2D rigid-body estimate to detect incipient slip without object-specific friction, mass, or shape knowledge.It produces a dense slip field and runs at around 24 Hz.Source: §I · p. 2
- The detector is tested on 10 daily objects and achieves an 86.25% success rate.Closed-loop bottle-cap screwing and unscrewing experiments use the feedback to inform the robot when to increase gripping force and when to stop.Source: §I · p. 2
II. RELATED WORK
The related work groups tactile slip detection into vibration-based and relative-motion-based approaches. The paper positions its method among vision-based relative-motion techniques, emphasizing simpler computation and faster detection.
II. RELATED WORK
- Prior tactile slip detectors are grouped into vibration-based and relative-motion-based approaches.The latter are mostly used with vision-based tactile sensors.Source: §II · p. 2
- Vibration-based approaches use high-frequency signals from relative motion, including piezo-resistive sensing and accelerometers,,,,.The cited work includes grasp-force control with slip feedback, while one sensor configuration is described as hard to use directly for grasping objects.Source: §II · p. 2
- Vision-based relative-motion methods use contact or marker motion, but reported constraints include texture dependence, sensing area, computational cost, or reduced manipulation speed,,,.The paper says its simpler algorithm increases detection frame rate while achieving similar or better accuracy, and that GelSlim's larger sensing region provides more time to sense and react.Source: §II · p. 2
III. METHOD
The method detects incipient slip by comparing measured marker displacement with the displacement predicted by rigid-body motion. It uses the still-adhered inner contact region to estimate that motion, then flags sufficiently large residuals.
III. METHOD
- Incipient slip occurs when a small part of the contact area loses contact, generally at the periphery, where marker displacement becomes smaller than at the center.The peripheral displacement difference supplies the trigger signal.Source: §III · p. 2
- The method assumes the no-slip contact patch moves as a planar rigid body and uses its inner region as a reference to estimate marker displacement.When several real marker displacements differ sufficiently from their estimates, the detector triggers incipient slip.Source: §III · pp. 2–3
- The algorithm detects the contact region, tracks marker displacement, estimates rigid motion from inner-region markers, and computes a slip field from the residuals.Incipient slip is indicated when multiple markers in the slip field exceed a threshold.Source: §III · p. 3
- The illustrated procedure detects both translational and rotational slip, with the rotational example showing upper-right and lower-left regions losing contact first.That pattern is described for clockwise object rotation with the instantaneous center of rotation roughly at the contact-patch center.Source: Figure 2 · p. 3, §III · p. 3
- The method detects directional slip when the contact patch spans a marker cluster, requires no prior object or contact knowledge, and runs at around 24 Hz.Marker localization is the largest computational cost; smaller contact patches can run faster.Source: §III · p. 4
A. Experiment Setup
The experiments use a robot-arm grasping setup with GelSlim sensors mounted as gripper fingers. One sensor supplies the slip measurements.
A. Experiment Setup
- The setup uses a 6-DOF ABB-1600 arm with a WSG-50 parallel gripper and two GelSlim sensors mounted as fingers.The system is based on the grasping setup developed in,.Source: A. Experiment Setup · p. 4
- Each GelSlim sensor captures contact imprints at 640*480 spatial resolution and 30 Hz across a 3×4 cm2 sensing surface.Source: A. Experiment Setup · p. 4
- Only one of the two mounted sensors is used to detect incipient slip in the experiment.The passage gives computational simplicity as the reason.Source: A. Experiment Setup · p. 4
B. Incipient Slip Detection Test
The slip-detection test used ten varied daily objects and manual pushes, pulls, or rotations to produce slip and non-slip cases. Trials included different applied forces, and data were relabeled after testing based on human observation.
B. Incipient Slip Detection Test
- Ten daily objects varied in shape, mass, softness, friction, and surface properties; several were soft, flat, or smooth.The water bottle was full of water, while the yellow duck, brain foam, and scrub sponges were soft and deformable.Source: B. Incipient Slip Detection Test · p. 4
- Objects were manually pushed, pulled, or rotated to test translational and rotational slip, with non-slip cases attempted by moving objects without making them slip.Translational tests used four evenly spaced directions parallel to the finger phalanges; rotational tests used clockwise and anticlockwise turns.Source: B. Incipient Slip Detection Test · p. 4
- In open-loop experiments, 240 experiments covered 10 objects, each tested 24 times (18 slip, 6 non-slip), across three forces (5 to 30N for most objects); data were relabeled afterward by human observation.Source: B. Incipient Slip Detection Test · p. 4
C. Bottle Cap Screwing and unscrewing with Incipient-Slip Feedback
The bottle-cap experiments used slip feedback to adjust gripping force during screwing and unscrewing. Screwing began at 10N and increased force in steps after slip warnings, while unscrewing began with a tightly secured cap and did not require the maximum force.
C. Bottle Cap Screwing and unscrewing with Incipient-Slip Feedback
- During screwing, the robot started at 10N and increased gripping force by 10N after each slip warning, stopping at 60N.The robot paused when slip was detected, then resumed the low-speed screwing motion.Source: C. Bottle Cap Screwing and unscrewing… · p. 4
- During unscrewing, the cap started tightly screwed; slip occurred at the beginning, so gripping force did not need to reach 60N.The process reversed the screwing procedure, which used maximum gripping force to firmly torque the cap on.Source: C. Bottle Cap Screwing and unscrewing… · p. 4
- The supplied caption identifies Fig. 5 as showing the experiment setup and steps for the unscrewing process.Source: Figure 5 · p. 5
A. Incipient Slip Detection Accuracy
The detector achieved 86.25% average accuracy, with most objects near 100% but more false negatives for the crayon box and scrub sponges. In bottle-cap screwing, slip feedback guided force increases and repeated detections near the thread end accompanied continuous slip.
A. Incipient Slip Detection Accuracy
- 86.25% was the method's average detection accuracy, with nearly 100% accuracy for most objects except the crayon box and scrub sponges.False positives were 0.83%, while false negatives were 12.90%.Source: A. Incipient Slip Detection Accuracy · p. 5
- Most false negatives came from the crayon box and scrub sponges, whose weak contact signals required larger gripping force for reliable detection.At 50N, success rate increased to 100.0% for the crayon box and 66.67% for the scrub sponges; the contact-patch quality and size could assess reliability in real time.Source: A. Incipient Slip Detection Accuracy · p. 5, Figure 6 · p. 5
B. Bottle Cap Screwing
- During screwing, the detector indicated when to increase gripping force and when the cap was tightly screwed.The robot tightly screwed the cap in all three trials.Source: B. Bottle Cap Screwing · p. 5
- The first slip signal occurred midway through screwing; after force increased to 30N, the contact patch expanded and later detections accumulated near the thread end.The initial loose cap barely stretched the gel, and successive slip detections informed force adjustment.Source: B. Bottle Cap Screwing · p. 5, Figure 7 · p. 6
- Large marker displacements persisted after displacement resets, indicating continuous slip; unchanged cap orientation in external images validated that interpretation.The direction of the slip field could potentially distinguish vertical cap motion from horizontal loss of grasp.Source: B. Bottle Cap Screwing · pp. 5–6
C. Bottle Cap Unscrewing
In the unscrewing trials, slip began immediately and the cap initially did not move with the sensor. After gripping force rose to 30N, the cap followed the gripper, with no further slip events reported.
C. Bottle Cap Unscrewing
- Slip occurred right after unscrewing began, and the first slip events indicated that the cap had not yet moved with the sensor.The initial slip arrows pointed opposite to those in screwing because the tasks used opposite force directions.Source: C. Bottle Cap Unscrewing · p. 6
- After gripping force increased to 30N, the cap began following the gripper and no further slip events occurred.Marker displacements then remained small.Source: C. Bottle Cap Unscrewing · p. 6
- Slip markers were located in the peripheral contact region, and the bottle cap was successfully opened in all three trials.Source: C. Bottle Cap Unscrewing · p. 6
VI. CONCLUSIONS
The conclusion presents a displacement-field slip detector with 86.25% accuracy and real-time feedback, while noting conditions that make detection more reliable.
VI. CONCLUSIONS
- 86.25% detection accuracy across 240 slip trials on 10 daily objects was comparable to, while running at 24 Hz, at least 8 times faster.Source: §VI · p. 6
- The algorithm compares contact-patch displacement with perfectly rigid-body displacement and detects directional slip without prior knowledge of the grasped object.Source: §VI · p. 6
- Detection works best with large contact patches and large tangential forces; flat, smooth objects require large gripping force for reliability.Source: §VI · p. 6
- Real-time incipient-slip feedback proved useful in bottle-cap screwing and unscrewing, with potential applications in grasp-force control, in-hand manipulation, and tool use.Source: §VI · p. 6