Human Touch Sensing for Actuators in Haptic Interfaces

Jorge Cham - Last updated 6/5/98

Intro

This document presents a summary of current research that has investigated the quantification of human tactile and kinesthetic sensing capabilities. This information is presented in three layers of detail: a Summary table, a Detailed Table and Notes. The Notes section at the end summarizes the main points from each of the works surveyed. The Detailed Table tabulates the numerical values found in them and references them (a bold reference number indicates that the author is responsible for this value; a non-bold number indicates that the author referenced that from another work). The Summary Table distills these numbers and shows representative values of the parameters we are mainly interested in.

Work on perception is mainly from authors such as M. A. Srinivasan, H. Z. Tan and L. A. Jones. Work by Srinivasan and Tan and others used mainly the "Linear Grasper" (actuated parallel plates which the subject pinches) to test JND's (Just Noticeable Differences) of different mechanical parameters in a pinch grasp. Their testing method generally used the "Forced Choice Discrimination" paradigm where the subject received two stimulus and had to choose between them (for example, "the stiffest one" or "the more viscous one"). Values for the JND's were then derived from the statistical data. Jones and others investigated many of the same parameters but used the forearm and elbow joint as the testing site, stating that it is representative of other joints. In their "Limb-matching" paradigm, the subject's limbs are coupled to two linear motor near their wrists. A stimulus is applied to one limb of the subject. The subject then had to adjust a pedal until it was perceived that the other limb was receiving the same magnitude stimulus. JND values are derived statistically from the error between the applied and adjusted stimulus.

Work on tactile perception is mostly from "Advanced Tactile Sensing for Robotics" (edited by H. R. Nicholls). Other important sources include the reviews by Tan et al. in "Human Factors for the Design of Haptic Interfaces" and C. Hasser's "Force Reflecting Anthropomorphic Handmaster Requirements".

Tasks that remains to be done includes: 1) Get feedback from some of the authors cited and validate or update these values. 2) Search for more information on vibrotactile perception and displacement ranges.

Summary Table

Detailed Table

Notes

• Results – JND’s:
• L: 1mm @ 10mm; 2.4mm @ 80mm
• F: 5-10% @ 2.5-10N; independent of reference, finger-span or displacement
• C (Compliance): 5-15% @ 4mm/N
• "Roving Displacement" (pushing distance varied randomly per trial) had large detrimental effect on F (14%) and C (22%) JND.
• Both cutaneous and proprioceptive active in these trials.
• Methods. A digital caliper for L and the "Linear Grasper" (parallel plates, ones of them displaces and is actuated, fingers pinch the plates) for F and C. Discrimination trials were used here and in most of other papers. Subject was presented with two values and had to choose between them.
• Results show that L discrimination violates Weber’s Law (percent discrimination should be independent of magnitude of stimulus).
• Hypothesis is that subjects discriminate compliance by terminal forces. Data supports this in some way.

• Advances in Robotics ASME 1992 v42, p13-18
• Results: JND for C: 15-99%, which if converted to force yields F: 5.2% which is comparable to previous results
• "Equal Work" Profiles where used where work is the same. Also, the reference forces were increased to minimized the relative terminal forces.
• Comparison with Jones et al (89, J+H 90 and 92) work:
• Results superficially concur F: 6-8% and K: 23%
• Jones ignores "Work hypothesis" i.e. effects of roving displacement.
• How come their stiffness JND was larger than their F JND when the roving displacement affects both F and K?
• Operating Ranges are different

• Perception and Psychophysics 1995 – 57(4) p495-510
• Summary of previous work without roving displacement, with it, and without work cues and minimized terminal force cues.
• References:
• L JND: 1mm @10-20mm; 2.2mm @ 80mm (Durlach 89)
• F JND: 7% @2.5-10N (roughly consistent with other reported figures)
• Jones and Hunter (89,90,92,93)
• B (Viscosity): 34%
• K: 23%
• F: 7% and L: 8%
• Srinivasan and LaMotte studies on softness show that deformable objects are sensed by tactile sensors while rigid objects by kinesthetic sensors.
• Table of Previous Work:

• Advances in Robotics, Mechatronics and Haptic Interfaces, ASME 1993
• Review: F JND: 7% over various conditions, provided kinesthestic sensors are involved. F JND: 10% in distinguishing weights, Jones (89) and Clark + Horch (86).
• Method consisted of subject placing finger on glass plate. A 6-axis force sensor is attached to the plate. Subject had to track a force profile that was shown in the screen (constant, ramp or sine wave).
• Results:
• Constant: Lack of tactile feedback changed error magnitude, lack of visual feedback changed magnitude and variation
• Rates of ramps had little effect on performance.
• Errors in sine waves with same average rate as some ramps still had slightly worse errors than these ramps.
• F: 11 – 15% in general and without visual feedback. This value is actually measuring the ability to control the contact force, not just sense it.

• Dynamic Systems and Control, v1, p369-375
• Results – JND’s of Torque:
• Discrimination: 12.7% @ 60mNm
• Control: 12.5% of mean angular velocity
• References to Tan et al (92), Jones + Hunter (92), Srinivasan + Chen (93) and Johansson + Cole (92) + Westling (84)
• Torque discrimination done in trials. Set reference at 60mNm because it is about ½ of the fatigue limitations (source not specified).
• In Control, the angular speed was left to the subject, and the errors were statistically processed.
• JND found (12.7%) is higher than the F JND previously found at 6-8%.
• Both activities are similar. When discriminating, subjects were trying to keep a constant velocity

• Proc. ASME Dynamic Systems and Control Div. ASME 1995, p657-662
• Results – JND’s:
• B (viscosity): 13.6 +-3% @ 120Ns/m
• M: 21.0 +-5% @ 12 Kg
• Review of past work.

• Advances in Robotics, v42 ASME 1992, p49-53
• Summary of previous results and present work on tracking performance (study of the effects of the mechanical properties of a manipulandum that an operator controls on their tracking performance).
• Introduction: Muscle receptor are the primary source of proprioceptive sensing; joint and cutaneous also contributes. The hand is unique with respect to proprioception. Proprioceptive system can be divided and studied as three: position, movement and Force.
• References: Fingers: Pang et al (92), Durlach (89); forearm: Jones (89), Erickson (74), Jones and Hunter (90a)
• Results – JND’s:

• Dynamic Systems and Control Div. ASME 1994, v1 p353-359
• Framework of issues in Human capabilities is presented and related to design issues of haptic interfaces. "...the human haptic system has limitations that can be exploited."
• Perceptual Issues:
• Force Sensing
• Slow-varying field: 7% (Jones 89, Pang Tan and Durlach 91) regardless of test conditions, reference force or body site.
• Vibration: the detection threshold for vibrotactile stimulation is roughly 28 dB, or 1 micron, below 30Hz and decreases at -12dB/oct from 30-300Hz. After that, it rises again.
• Pressure Perception: this is particularly an issue in ungrounded interfaces. The reaction force needed at the fingertip that is applied at the forearm might go unnoticed if it is below the threshold value. Pressure JND tests were performed and it was found to vary with contact area. But, if calculated as "weight per perimeter JND" it stays constant at 0.06-0.09N/cm. This can be explained that humans are sensitive to pressure gradients.
• Position Sensing Resolution. Using hinged plates with a protactor, subject's joint was moved by the experimenter and asked to discriminate between them. Finger JND are from (RLE Progress Report, 1992, MIT). JND Results are:

• Proc. ASME Dynamic Systems and Control Div. ASME 1995, v57-2 p663-674
• Force and Torque References:
• An et al (86) measured forces normal to midpoint of finger joints in a cylindrical max-strength power grasp
• Sutter et al (89) measured max force at the tip of the finger when the fingers are fully extended.
• Tan et al (94) measured fingertip forces while selectively immobilizing joints. His results are very close to the ones above which are compiled below.
• Using these force measurements along with representational finger lengths and angles at the appropriate measurement configuration, joint torques are estimated (in N-cm):

• Chapter on "Lessons From the Study of Biological Touch for Robotic Tactile Sensing" by S. Lederman and D. Pawluk
• Pressure Amplitude: 14% JND static pressure and 20% JND vibrations at 160Hz. Typical range is 20-25% over 20-300Hz (Lederman, 91)
• Displacement: 10 microns static for a .45mm probe. Pre-indentation, lateral movement and contactor size greatly affect this threshold (Johansson and Valbo, 79).
• Spatial Sensitivity (minimum discriminable spatial separation between elements): 0.9mm (Johnson + Phillips, 81) and 0.7mm with relative motion (Phillips et al 83). Mechano-receptor spacing is typically not less than 1mm (Johansson + Valbo 83). Discrimination was found at 0.17mm (Loomis, 74, 79, 81).
• Temporal (Vibrotactile) Sensitivity: Different "channels" and types of sensors have different vibrotactile thresholds over different bandwidths. So at a given frequency, the channel with the lowest threshold determines the over-all threshold (Bolanowski et al, 88).
• Texture discrimination: Morley (83) showed humans can discriminate gratings with spatial period variations of 5% while Lamb (83) found 2%.

• A single 2 micron raised element can be detected when stroking. A texture with elements 0.06 microns can be detected, but only through stroking.
• Tactile spatial resolution of gratings corresponds to afferent spacing of 1mm. "..the SAI afferents set the limits of tactile spatial resolution for spatial forms indented into the fingerpad."

• Stiffness and viscosity JND: 13-30% (Beauregard et al 95 and Jones et al 97). Limb movement, position and force: 6-8% (Jones et al 92 and Tan et al 95).