Perceiving Action-Relevant Properties of Tools Through Dynamic Touch:Effects of Mass Distribution, Exploration Style, and Intention
Steven J. Harrison, Alen Hajnal, Stacy Lopresti-Goodman, Robert W. Isenhower, and J. M. Kinsella-ShawUniversity of Connecticut
At issue in the present series of experiments was the ability to prospectively perceive the action-relevantproperties of hand-held tools by means of dynamic touch. In Experiment 1, participants judged objectmove-ability. In Experiment 2, participants judged how difficult an object would be to hold if heldhorizontally, and in Experiments 3 and 4, participants rated how fast objects could be rotated. In eachexperiment, the first and second moments of mass distribution of the objects were systematically varied.Manipulations of wielding speed and orientation during restricted exploration revealed perception to beconstrained by (a) the moments of mass distribution of the hand-tool system, (b) the qualities ofexploratory wielding movements, and (c) the intention to perceive each specific property. The results areconsidered in the context of the ecological theory of dynamic touch. Implications for accounts of theinformational basis of dynamic touch and for the development of a theory of haptically perceiving theaffordance properties of tools are discussed.
When an object is held, hefted, or wielded, it imparts resistiveforces on the body; these forces in turn affect the body’s structuresand tissues. A particular form of haptic perception, termed dy-namic touch (Gibson, 1966; Turvey & Carello, 1995) or kinesthe-sis, is implicated whenever mechanical contact affects the tensilestates of muscles, tendons, and fascia, and in turn patterns theensemble activity of mechanoreceptors (Fonseca & Turvey, 2006).Research into dynamic touch has revealed that action-relevantproperties of everyday objects can be perceived through simplemechanical interactions (Bingham, Schmidt, & Rosenblum, 1989;Solomon & Turvey, 1988; Wagman & Carello, 2001). In contrastto the properties of an object (e.g., length, shape, and heaviness),action-relevant properties denote properties defined over theorganism–tool system specific to the performance of a goal-directed action. For example, when given the opportunity to heft a
ball in one hand, it is possible to get an appreciation of how far thatball could be thrown (Bingham et al., 1989). Motivated by previ-ous investigations, we designed experiments to evaluate specifichypotheses regarding how action-relevant properties, perceived bymeans of dynamic touch, are affected (a) by the physical propertiesof the objects and (b) by the manner in which the objects arewielded.
The Informational Basis of Dynamic Touch
From an ecological perspective (Gibson, 1979; Michaels &Carello, 1981; Turvey, Shaw, Reed, & Mace, 1981), hapticallyperceiving the properties of hand-held tools through dynamictouch is dependent upon the pickup of information carried intissue-deformation arrays (Fitzpatrick, Carello, & Turvey, 1994;Solomon, 1988). When wielding a tool about the shoulder, it isnecessarily true that the mechanics1 of both the wielding arm andthe wielded tool will structure the patterning of stresses and strainsimposed upon the tissues of the body. Consequently, it has beensuggested that the information underlying dynamic touch is thefunctional embodiment of the rotational dynamics that physicallydescribe the body–tool system (Kugler & Turvey, 1987; Solomon,Turvey, & Burton, 1989b; Turvey, 1996). The search for hapticperceptual invariants specifying action-relevant properties of toolshas focused upon identifying the invariants of the mechanicshypothesized to be implicated in dynamic touch.
Kingma, Beek, and van Dieen (2002) introduced a simple phys-ical model of the mechanics operating in a hand–object system.The model relates the basic physical properties of a held object (itsmoments of mass distribution) to the muscular forces required tocontrol that object. In this model it is assumed that point masses m
1 This term refers to both the statics and the dynamics of the tool-handsystem.
This article was published Online First November 15, 2010.Steven J. Harrison, Alen Hajnal, Stacy Lopresti-Goodman, and Robert
W. Isenhower, Center for the Ecological Study of Perception and Action,University of Connecticut; J. M. Kinsella-Shaw, Center for the EcologicalStudy of Perception and Action and Department of Kinesiology, Universityof Connecticut.
Alen Hajnal is presently at the Department of Psychology, University ofSouthern Mississippi. Stacy Lopresti-Goodman is presently at the Depart-ment of Psychology at Marymount University. Robert W. Isenhower ispresently at the Department of Psychology, Rutgers University.
The authors wish to acknowledge the assistance of M. T. Turvey andClaudia Carello for their help in developing the experimental paradigm,and Claire F. Michaels and Robert Shaw for their editorial comments andsuggestions. The research was supported by grants from the Provost’sOffice at the University of Connecticut.
Correspondence concerning this article should be addressed to Steven J.Harrison, Department of Psychology, Center for the Ecological Study ofPerception and Action, U-20, 406 Babbidge Road, University of Connect-icut, Storrs, CT 06269-1020. E-mail: [email protected]
are attached to a mass-less rod at distances d from a mass-lesshand. The neuromuscular torque (Nm) acting at the hand in such asystem can be represented as
Nm � I1� � Mg cos�� �, (1)
where � is the angular acceleration of the wielded object, g is thegravitational acceleration, � is the angle of the rod with respect tothe gravitational vertical, and M and I1 are invariants of themechanics of the hand–rod system.
In the four experiments presented in this article, participantswielded objects that had two attached masses (see Figure 1);applying a simple variant of the model above, we have one mass(m1) located at a distance (d1) above the point of rotation and asecond mass (m2) located at a distance (d2) below the point ofrotation. In this system M is the first moment of mass distributionand is captured by
M � m1d1 � m2d2, (2)
and I1, the principle component of the second moment of massdistribution (I), is captured by
I1 � m1d12 � m2d2
2. (3)
On the basis of Equations 1, 2, and 3, it is possible to capture thetorques acting in this system as
Nm � �m1d12 � m2d2
2�� � �m1d1 � m2d2�g cos�� �, (4)
where the first term on the right side of the equation is inertialtorque (Ni) and the second term is the gravitational torque (Ng).The terms m, M, and I are each mechanical invariants of thissystem and are more commonly referred to as the zeroth, first, andsecond moments of mass distribution, respectively.
Consistent with the model above, research has implicated thezeroth, first, and second moments of mass distribution in hapticperceptual judgments of both geometric (length and shape) andkinetic (heaviness) object properties (Amazeen & Turvey, 1996;Burton, Turvey, & Solomon, 1990; Turvey, Burton, Amazeen,
Butwill, & Carello, 1998; Turvey, Carello, Fitzpatrick, Pagano, &Kadar, 1996). The potential for these mechanical invariants toprovide an informational basis for perceiving object properties canbe illustrated through an analysis of haptically perceiving objectlength. For a homogeneous rod in which an increase in length (L)is linearly related to an increase in mass (m)—that is, L � m—it isapparent from Equation 2 that L � M1/2, and from Equation 3 thatL � I1/3. Consistent with this analysis, studies using rods withattached masses as stimuli (used to dissociate L and m) havesuggested that the second moment is implicated in length percep-tion. Moreover, these studies have shown judgments to be scaledwith I with a power of one third, as Equation 3 predicts (Solomon& Turvey, 1988; Solomon, Turvey, & Burton, 1989a).
A basic assumption of the approach just outlined is that themechanics of the body–tool system underwrite dynamic touch bylawfully constraining the patterning of stresses and strains over theneuromuscular system. In the present series of studies we buildupon this assumption. We used Equation 4 to generate hypothesesregarding the haptic informational basis of perceiving action-relevant properties.
The Role of Exploration in the Pickup of Information
Perception is an active process, accompanied by a rich repertoireof movements that facilitate the pickup of information (Gibson,1962, 1966, 1979). Studies of the manual manipulation of objectsreveal that when given the task to perceive specific object prop-erties (e.g., rigidity, shape, and weight), specific classes of hapticexploratory movements (e.g., pressing, contour following, andhefting) are typically used (Lederman & Klatzky, 1987). In thecase of dynamic touch, exploratory wielding movements are alsoconstrained by the intention to perceive a particular object property(Riley, Wagman, Santana, Carello, & Turvey, 2002) or action-relevant property (Michaels, Weier, & Harrison, 2007). Michaelset al. (2007) observed that wielding patterns varied as a function ofthe intention to perceive a given action-relevant property. Forexample, the intentions to perceive either how far a ball might behit with a bat or how effective a hammer might be yieldedexploratory movements performed at high speed with a largedegree of travel. In contrast, exploration of objects for use aspokers or scoops yielded comparably slower motions with a nar-rower range of exploration. Additionally, Riley et al. (2002) ob-served that the intention to perceive either the length of an objector its width resulted in task-specific wielding dynamics as indexedby changes of temporal (recurrent) structure.
The research above shows that exploratory movement is con-strained by task; nevertheless, wielding kinematics for a particulartask show remarkable variety. Length-perception studies revealthat judgments are unaffected by whether an object is wieldedslowly or quickly (Solomon & Turvey, 1988, Experiment 4) orwhether it is wielded about the shoulder, the elbow, or with the fullrange of motion of the arm (Pagano, Fitzpatrick, & Turvey, 1993).Moreover, judgments appear to be unaffected by the particularmuscular synergies adopted, such as those resulting from using anoverhand or underhand grip to wield the object (Solomon et al.,1989a) or even whether object exploration is achieved by wieldingwith the wrist or the ankle (Hajnal, Fonseca, Harrison, Kinsella-Shaw, & Carello, 2007). These results suggest a remarkable re-dundancy in the exploratory actions capable of revealing the length
Figure 1. Experimental set-up used in Experiments 1–4. (Inset) Screenmarkings used in Experiments 1–4 to constrain wielding of the rods withinspecific ranges.
194 HARRISON ET AL.
of a wielded object. There is, however, less consensus regardingthe influence of the gravitational field on haptic perception: Someempirical results showed that orientation of wielding does notinfluence length perception (Solomon & Turvey, 1988, Experi-ment 3), whereas follow-up studies by the same group of research-ers (Solomon et al., 1989a, Experiment 2) have shown an effect oforientation.
Building upon the original paradigm of Solomon and Turvey,more recent studies have begun to systematically manipulate thezeroth, first, and second moments of mass distribution of experi-mental stimuli, demonstrating the implication of each momentindependently (Kingma, van de Langenberg, & Beek, 2004; van deLangenberg, Kingma, & Beek, 2006, 2007). In contrast to thestudies that have suggested that haptically perceiving length isindependent of the specifics of exploration (Solomon & Turvey,1988), these investigations have shown that the informational basisfor perceived length is affected by how an object is wielded. Whenobjects are only vertically translated (normal-to-the-ground plane),only the zeroth moment of mass distribution (m) is implicated(Lederman, Ganeshan, & Ellis, 1996; van de Langenberg et al.,2006); whereas, when objects are held still in a horizontal orien-tation (parallel to the ground), perceived length depends on thefirst moment of mass distribution (Burton & Turvey, 1990;Carello, Fitzpatrick, Domaniewicz, Chan, & Turvey, 1992;Kingma et al., 2004; van de Langenberg et al., 2006).
Further dependencies have been uncovered in studies in whichparticipants were asked to actively explore (rotationally accelerate)objects by freely wielding them about the wrist joint. Given suchexploration, the second moment of mass distribution is also im-plicated. When such “free” wielding about the wrist is restricted tosideward wielding (about an initially horizontal orientation), bothM and I in combination influence judgments (Burton & Turvey,1990; Kingma et al., 2004). Comparatively, I, M, and m all appearto play roles in length perception when initially vertically alignedobjects are actively wielded (van de Langenberg et al., 2006).Complicating the interpretation of this literature is the fact that notall studies independently manipulated the moments of mass dis-tribution of the experimental stimuli. It should be noted, therefore,that some of the differences among the aforementioned results maybe attributable to covariation, as opposed to the independent vari-ation, of mechanical invariants (see Kingma et al., 2004).
In each of the preceding examples, it is clear that the implicationof a particular moment of mass distribution depends upon the styleof object exploration brought to bear. Interpreting the model pre-sented in Equation 1, Kingma and colleagues (Kingma et al., 2002,2004) suggested that changes in the style of exploration could beconceived of as changes in mechanical context affecting the sa-lience of particular invariants. Central to this account is the as-sumption that Nm plays an important role in the theory of theinformational basis of dynamic touch. From Equation 1 we see thatwhen a rod is held “still” in a horizontal orientation Nm (andconsequently the patterning of stresses and strains on the body) isdetermined by variation in M; accordingly, M appears to beuniquely implicated in judgments of length. Similarly, as an objectis accelerated away from a vertical orientation, both inertial andstatic torque is generated, and both M and I are co-implicated.Kingma and colleagues suggested that the implication of either Mor I depends upon the respective contributions (signal–noise ra-tios) of Ng and Ni to Nm in a particular mechanical context (i.e.,
specific values for � and �). For example, in the absence ofsubstantial acceleration in exploratory movements, the statictorques contributing to Nm will generally be greater than will theinertial torques. In this example the role of M is said to be more“salient” in the stimulus flow (Kingma et al., 2004). Carefulinspection of Equation 1 further suggests that the salience of Mand I will be influenced by both � and � such that greateracceleration will increase the salience of I and decrease the sa-lience of M, whereas, wielding away from the vertical orientationwill increase the salience of M and decrease the salience of I.
Perceiving Action-Relevant Properties of Hand-HeldTools
In the original paradigm of Solomon and Turvey (1988), theinitial motivation was the development of an account of the infor-mational basis for perceiving the potential of a hand-held imple-ment (a tool) to extend the functional reach of the actor. Simplyput, they were interested in understanding the informational basisof perceiving affordances (Gibson, 1979). In this section we sum-marize research that has led to insights into how an affordance-based perspective on dynamic touch might progress.
Research into heaviness perception has implicated each of thezeroth, first, and second moments of mass distribution (Kingma etal., 2004). The observation that objects feel lighter when theprincipal moments of the second moment of mass distribution aremore nearly equal has led to the proposal that specific combina-tions of the principal moments form the relevant mechanicalinvariant (Amazeen & Turvey, 1996). The principal moments I1,I2, and I3 capture an object’s resistance to angular accelerationabout each of its three axes of rotation.2 Instantiated geometrically,the principle moments form an inertia ellipsoid. Two compoundproperties derived from the inertia ellipsoid are ellipsoid volume(V) and symmetry (S), where V � 4 �/3(I1 I2 I3)–1/2, and S � 2I3/(I1 � I2). Studies using objects varying in I1, I2, and I3 haverevealed that heaviness perception is a function of S and V incombination with m (Shockley, Grocki, Carello, & Turvey, 2001).The experimental demonstration of haptic metamers (see Shock-ley, Carello, & Turvey, 2004; Turvey, Shockley, & Carello, 1999)provides compelling support for this claim.
Carello (2004) suggested intuitive analogues to V and S, pro-posing that V is related to the mean level of torque needed to movean object, and S is related to how that torque should be directed.The implication drawn from such mappings is that heaviness ismore usefully understood as controllability (Carello, 2004; Wag-man & Carello, 2001) or move-ability (Shockley et al., 2004;Turvey, Whitmyer, & Shockley, 2001). In this approach, thequestion of “how heavy” equates perceptually with the action-relevant property of how easily a particular object can be moved ormanipulated.
Following Carello (2004), in the present series of experiments,we did not ask participants to rate properties of the object as suchbut instead asked them to judge action-relevant properties of thebody–tool system of the kind suggested by terms such as control-
2 Note that only the contribution of I1 applies to the present experiments,because the objects in Experiments 1–4 could only be rotated about asingle fixed axis.
195HAPTIC PERCEPTION OF ACTION-RELEVANT PROPERTIES
lability. In Experiment 1, participants judged how difficult eachobject was to move. In Experiment 2, participants judged howdifficult it would be to hold an object if it were to be orientedhorizontally, and in Experiments 3 and 4, participants rated howquickly they would be able to move each object by using aparticular manner of movement.
Previous investigations into the perception of affordancesthrough dynamic touch have focused on the perception of theproperties of hand-held tools that inform the control of action(Carello, Thuot, Anderson, & Turvey, 1999; Hove, Riley, &Shockley, 2006; Michaels et al., 2007; Wagman & Carello, 2001,2003; Wagman & Taylor, 2004). Importantly, such investigationshave revealed that although perception is frequently constrained bysimple geometric properties (such as the length of a body segmentor object), it is also dependent upon both the kinematic and kineticcharacteristics of the tool–body system (Bongers, Michaels, &Smitsman, 2004; Carello, Grosofsky, Reichel, & Solomon, 1989;Kreifeldt & Chuang, 1979).
Probing the Informational Basis for Perceiving Action-Relevant Properties
To recap, the current investigation focuses upon (a) the hapticinformational basis of affordances and (b) the qualities of explor-atory activities that allow for the pickup of such informationthrough dynamic touch. Four experiments are presented. In eachexperiment, participants wielded objects and were asked to giveverbal reports. Wielding was constrained to one of three wieldingorientations, and participants rated each object property on thebasis of their limited exploration in that orientation. The presentedexperiments consequently address a basic functional prerequisiteof adaptive tool use, specifically the ability to prospectively per-ceive an action-relevant property of a tool following limited ex-ploration. In each of our experiments, we expected that perceptionby dynamic touch would be (a) tied to invariant properties of themechanics of the body–tool system and (b) dependent upon themechanical context in which the wielded objects were explored.To probe this latter expectation, we recorded exploratory move-ments in Experiments 1–3 and directly manipulated wieldingspeed in Experiment 4. The movement analysis is reported only forexperiments in which it was used to evaluate specific hypotheses.
On the basis of prior research, we hypothesized that explorationwould vary as a function of the intention to perceive each partic-ular action-relevant property. We also hypothesized that the me-chanical invariants implicated in judging a particular propertywould be systematically constrained by the manner of exploration.
Consistent with the methods of independent variation of me-chanical invariants developed by Kingma and colleagues, partici-pants in each experiment wielded objects taken from two speciallydesigned stimulus sets (Set 1 and Set 2; for object properties seeTable 1). In each set, objects varied only with respect to a singlemoment of mass distribution. The physical characteristics of ob-jects in Set 1 varied only in first moment of mass distribution,whereas the objects in Set 2 varied only in second moment of massdistribution. The zeroth moment of mass distribution was heldconstant in both sets.
Experiment 1: Perceiving Move-Ability by DynamicTouch
Experiment 1 addressed the informational basis of hapticallyperceiving how difficult it is to move a hand-held object. If, as hasbeen previously suggested (Carello, 2004; Turvey et al., 2001),judgments of move-ability can be equated with judgments ofheaviness, then both the first and second moments of mass distri-bution (both previously implicated in heaviness perception),should constrain judgments. To avoid problems in interpretation ofresults due to covariation among moments of mass distribution,two groups of participants wielded objects that varied in either thefirst (Set 1) or second moment (Set 2) of mass distribution (seeTable 1) and rated how difficult to move they perceived eachobject to be. The stimuli were well suited to test whether M and I(a) form the informational basis for move-ability judgments and(b) can support perception in the absence of covariation with theother.
To probe the hypothesized connection between exploratory con-text and the roles of I and M, participants actively wielded hand-held objects in one of three orientations (vertically oriented, VO;diagonally oriented, DO; or horizontally oriented, HO). Followingprior claims that the relative implication of either M or I dependsupon the respective contributions (signal–noise ratios) of Ng andNi to Nm (Kingma et al., 2004), we expected that the implication
Table 1Mechanical Properties of Rods Used in Experiments 1–4
Object SetLength
(m)
Ring (top) Ring (bottom) Resulting object properties
of both M and I would be affected by the experimental manipu-lation of orientation. On the basis of an examination of the con-sequences of changing � in Equation 4, we hypothesized that (a)the ability to detect changes in move-ability arising because ofvariation in M (Set 1) would be compromised in the vertical (leastsalient) orientation in which Ng makes a minimal contribution toNm and (b) the ability to detect changes in move-ability arisingbecause of variation in I (Set 2) would be compromised in thehorizontal orientation because the relative contribution to Nm ofthe term with I is minimal.
Method
Participants. Twelve University of Connecticut studentswere recruited for the experiment. Six participants wielded objectsfrom Set 1, and 6 others wielded objects from Set 2. The partici-pants received experimental credits as part of a requirement for anIntroductory Psychology class. Participants were naıve to the pur-pose of the experiment. All participants gave their consent inaccordance with the University of Connecticut’s Internal ReviewBoard’s regulations for studies with human participants. The mo-tion recording of 1 participant in the first moment variation groupwas lost because of a file becoming corrupted and was omittedfrom the reported motion analysis.
Apparatus and stimuli. Figure 1 shows the experimentalset-up. The apparatus was designed to constrain the act of wield-ing, restricting the manner in which each object could be explored.Once attached to the apparatus, each object could be grasped andfreely rotated about a lubricated (low friction) axle. The apparatushad three notable properties: (a) all exploratory wielding motionswere rotations; (b) the weight of the objects was supported at alltimes such that the stresses and strains imparted at the point ofrotation were solely a function of changes in net torque at the axisof rotation; and (c) I was effectively simplified to I1 (the object’sresistance to rotation about its principle axis). Given these con-straints, Equation 1 provides a good approximation of the mechan-ical constraints operating in the experimental system.
Participants were seated in a manner that allowed them tocomfortably wield the objects. A writing chair modified to have anarmrest attached to its right side was positioned in front of anopaque screen. The screen occluded the participant’s view of boththe experimental objects and the experimenter. A 12-cm-diameterhole in the screen allowed participants to take hold of the objectcurrently attached to the axle. Behind the screen a simple woodenstructure provided a base, supporting the axle that held the exper-imental stimuli. A second cut- out in the screen allowed the end ofa lightweight dowel, attached to each object, to be visible to theparticipant. Using the dowel, participants could visually controlthe orientation of the occluded object during a trial. Three 0.7-m-long strips of red electrical tape were placed on the screen (thickblack lines in inset panel of Figure 1). These strips identifiedspecific starting orientations used to manipulate the orientation ofwielding during the experiment. Four 0.6-m- long strips of yellowelectrical tape (thin white lines in the inset panel of Figure 1)identified a range of 22.5° around each starting orientation.
Twelve objects in total were used in the experiment. The base ofeach object was a wooden dowel, 1.22 m in length, 1.28 cm indiameter, and with a mass of 0.1 kg. A lightweight hollowT-shaped PVC piping joint was used to connect the rod to the
lubricated axle. It also functioned as a handle for the participantsto grasp. A second PVC joint, attached 14 cm above the axis ofrotation, held the dowel used to identify object orientation toparticipants. Nylon weights were used to add 0.4 kg of additionalweight at different locations along the shaft of each rod. Theparticular placement of these nylon weights was used to manipu-late each objects mass distribution. The resulting objects could allpotentially have been freely wielded. Two sets of objects wereconstructed. The design of each object is summarized in Table 1.Set 1 was made up of Objects 1–5 (I constant and M varying), andSet 2 was made up of Objects 6–10 (M constant and I varying).Objects 3 and 8 were identically constructed so that one could beused as a standard object and the other as a test object. Thisavoided the possibility that these objects could be “recognized” asthe standard. Once attached, all objects were bottom heavy andnaturally tended toward a vertical orientation.
Wielding motions were recorded (at a sampling rate of 60 Hz)with a FASTRAK motion-capture system (Polhemus Corporation,Colchester, Vermont) and were processed with 6-D ResearchSystem software (Skill Technologies Inc., Phoenix, Arizona). Twomotion sensors were used to record the rotation of the object aboutthe fixed axis. The recordings were used to calculate time series oforientation of exploration, exploratory accelerations, and gravita-tional torque (Ng). For each trial, mean values of these time serieswere calculated. In the case of exploratory accelerations the meanabsolute values of the time series were also computed.
Maximum grip strength was also recorded for each participantas a simple way to obtain a measure of individual differences inaction capabilities. Three consecutive measures were taken using aJamar hand-held dynamometer (Preston Corporation, Ontario,Canada) at both the start and end of the experiment.
Procedure. On each trial, participants were asked to verballyrate how difficult to move they perceived each test object to be. Testobjects were rated by comparison with a standard object, after par-ticipants had wielded both consecutively. A rating scale was used,with the standard object assigned to be “100.” All judgments of testobjects were made in relation to this baseline. For example, a testobject that was perceived to be half as difficult to move as was thestandard object was rated as “50,” and a test object that wasperceived to be twice as difficult to move as was the standard wasrated as “200.” On a given trial, participants explored the standardobject followed by one of the test objects.
The starting orientation and angular range of exploration werecarefully controlled for both the standard and test objects. Beforethe start of each trial the experimenter supported the object in aparticular starting orientation (0°, 45°, and 90°). For the standardobject this was always 45°. Participants then grasped the handleand took hold of the object by rotating it free of the experimenter’ssupport. Aided by the sight of the dowel protruding from thescreen and the markings on the screen, participants attempted tokeep their exploratory movements within 22.5° from the startingorientation. Participants were free to explore the object in anymanner that they deemed appropriate, as long as the orientation ofthe object remained within the designated range. In Figure 1, theparticipant is engaged in a vertically oriented trial, wielding theobject in an initially 90° orientation, in the range 67.5°–112.5°.When participants had a “feel” for how difficult to move an objectwas, they returned the object to the starting orientation where theexperimenter took it from them. In all trials, participants wielded
197HAPTIC PERCEPTION OF ACTION-RELEVANT PROPERTIES
the standard object in a diagonal orientation (45° 22.5°). Ex-perimental objects were explored in horizontal (0° 22.5°),diagonal (45° 22.5°), and vertical (90° 22.5°) orientations.
Exploratory movements were recorded for both the standardobject and test objects. Recording started when participants ini-tially grasped the object and concluded either when they reportedhaving a feel for the object or when they made a judgment.
Each individual participated in a total of 45 trials, receiving eachof the five test objects three times in each orientation. Presentationof trials was fully randomized. The experiment lasted approxi-mately 45 min.
Results and Discussion
Ratings of how difficult to move were analyzed through sepa-rate repeated measures analyses of variance (ANOVA). The sep-arate analyses were performed on the ratings of the two groups ofparticipants assigned to wield objects from either Set 1 or Set 2. A3 (orientation) � 5 (object) ANOVA for the Set 1 group showedthat both variation in M over the objects in the set, F(4, 28) �7.62, p � .001, p
2 � .52, and the orientation of exploration, F(2,14) � 7.32, p � .01, p
2 � .51, had significant effects. The effectsof orientation and object interacted, F(8, 56) � 3.77, p � .01, p
2 � .35, such that ratings increased as a function of both increas-ing M and increase in tendency of objects to be wielded horizon-tally (see Figure 2, left panel). An identical ANOVA that wasperformed for the Set 2 group showed no effect of variation of Iover objects. The analysis again revealed an effect of orientation,F(2, 10) � 107.83, p � .001, p
2 � .96, and no interaction wasobserved (see Figure 2, right panel). For both the analysis of theSet 1 and Set 2 groups, post hoc pairwise comparisons withBonferroni corrections revealed that ratings in all orientationsdiffered from one another significantly ( p � .05). These resultsindicate that I was not implicated in participants’ ratings and thatthe implication of M on participant’s ratings explicitly dependedupon the orientation of exploration.
On the basis of these results, we reject the hypothesis thatspecific mechanical invariants (namely, I and M) would be impli-cated in the task of rating move-ability. We concluded that hapticjudgments of move-ability were not based on a property invariantover orientation of exploration but were instead based upon a
variant property of the hand–object system that depended uponorientation of exploration.
Orientation-dependent effects have been previously reported:Investigations of object-length perception by static holding haveimplicated the mechanical invariant M (invariant over orientation)in combination with the property Ng (dependent upon orientation)when orientation has been manipulated (Carello et al., 1992;Lederman et al., 1996). On the basis of the strong and systematiceffect of orientation and M in the present experiment, we inves-tigated whether changes in Ng over the experimental conditionswould explain the observed results. To explore this possibility, weperformed a regression analysis between perceived move-abilityand M, I, and Ng, respectively. A regression analysis including allobjects from both object sets revealed that very little variance inmove-ability ratings was explained by either I (r2 � .003, p �.514), or M (r2 � .190, p � .001). In contrast, Ng accounted formost of the variance in judgments (r2 � .824, p � .001).
The results indicate that judgments of move-ability dependedupon Ng. Although Ng is not a mechanical invariant, it is impli-cated in Equation 4 and is therefore consistent with the generalexpectation that haptic perception of the action-relevant propertiesof tools by dynamic touch is constrained by the mechanics of thebody–tool system. According to Equation 1, not only were partic-ipants’ judgments dependent upon Ng, but on the basis of theabsence of an effect of I, they were quite notably independent ofNi. This suggests that the haptic perceptual system of dynamictouch can, in effect, parse out the independent contributions of Ni
and Ng out of Nm.In Experiment 1 we assumed (on the basis of previous findings)
that perceived heaviness (an invariant property of the object) couldbe equated with perceived move-ability. The current results, im-plicating Ng, suggest one of two possibilities: either (a) partici-pants were unable to detect M or I (a possibility, given the tightconstraints placed upon exploratory movements) and instead basedtheir judgments on the “best” alternative, namely, Ng; or (b) theproperty they were in fact judging was not functionally equivalentto object heaviness (where heaviness is equated with objectweight, a physical property of the object that is invariant overorientation). With regard to this latter possibility, we hypothesizedthat participants could have been judging how difficult to moveeach object was in the local context of exploration. The localcontext of exploration is operationally defined here as explorationconfined to a specific initial orientation 22.5°. More generally, inrelation to developing an account of dynamic touch, we wouldsuggest that this term refers to the context of restricted possibilitiesfor exploration facilitating the pick-up of invariant information.Because the studies that had previously equated move-ability withheaviness did not constrain exploration, in Experiment 2 we in-vestigated the possibility that the intention was underspecified inthe instructions given to participants in Experiment 1.
Experiment 2: Perceiving How Difficult to Hold
From Experiment 1 we concluded that the haptic perception ofhow difficult to move was not perceived invariantly over orienta-tion of exploration. This result raised the issue of whether it ispossible to perceive action-relevant properties of objects indepen-dently of orientation of exploration, when objects are wielded as isexperimental apparatus used in Experiment 1. To explicitly test
Figure 2. Results from Experiment 1. Ratings of how difficult to moveafter wielding stimulus objects in vertical orientations (VO), diagonalorientations (DO), and horizontal orientations (HO). (Left) Ratings madefor five objects (Objects 1–5) designed to be of equal I, and to vary in M,in relation to a standard (Object 3). (Right) Mean ratings made for fiveobjects (Objects 6–10) designed to be of equal M, and to vary in I, inrelation to a standard (Object 8).
198 HARRISON ET AL.
that scenario, participants in Experiment 2 again explored exper-imental objects in three orientations, but did so while attempting toperceive how difficult each object would be to hold if it wereoriented horizontally. In other words, participants were explicitlyasked to judge a basic functional property of an object that wasindependent of the orientation of exploration, namely, if one wereto orient an object horizontally, how difficult would it be to hold.
From Equation 4 we can see that for an object held still (� � 0)and horizontally (� � 90°), the resultant Nm is determined by M.We consequently hypothesized that the informational basis for thistask is constrained by variation in M. Repeating our initial hy-pothesis from Experiment 1, we predicted that if breaks in thehypothesized invariance were to occur, it would be most likely inthe vertical orientation. To directly assess this hypothesis, we hada single group of participants wield objects that varied in firstmoment (Set 1) of mass distribution and rated how difficult theyperceived it would be to hold each object in a horizontal orienta-tion.
Method
Participants. Eight University of Connecticut students wererecruited for Experiment 2. All participants wielded objects fromSet 1. Participant recruitment mirrored that of Experiment 1.
Apparatus and stimuli. Only object Set 1 was used; other-wise, the apparatus and experimental set-up was identical to that ofExperiment 1.
Procedure. With the exception of a change in task, all proce-dures followed the protocol outlined for Experiment 1. Participantsagain wielded standard objects in a diagonal orientation, and testobjects were explored in either horizontal orientation (HO), diag-onal orientation (DO), or vertical (VO) orientation.
Results and Discussion
A repeated-measures ANOVA was performed on ratings of howdifficult to hold horizontally. A 3 (orientation) � 5 (object)ANOVA showed significant effects of variation in object M, F(4,28) � 45.94, p � .001, p
2 � .87, and orientation of exploration,F(2, 14) � 35.54, p � .001, p
2 � .84; and an interaction effect oforientation and object, F(8, 56) � 15.63, p � .001, p
2 � .69. Apost hoc pairwise comparison with a Bonferroni correction formultiple comparisons revealed that ratings for the VO differedfrom those for the DO and HO ( p � .05), whereas DO and HOwere not significantly different (see Figure 3). A simple effect ofobject was found in all three orientations. As might be expectedfrom inspection of Figure 3, the simple effect of object was morepronounced in the horizontal, F(4, 20) � 38.98, p � .001, anddiagonal, F(4, 20) � 53.85, p � .001, orientations than in thevertical orientation, F(4, 20) � 13.47, p � .001.
Similar to that found in Experiment 1, the reliance of judgmentson M was found to be influenced by orientation of exploration. Incontrast to Experiment 1, ratings were not found to explicitlydepend on orientation; instead a conditional dependency (namely,whether or not the object was wielded vertically) was suggested.Consistent with our analysis of the mechanics, these results appearto implicate M, but did so consistently only under conditions inwhich the objects are wielded away from a vertical orientation.One interpretation is that in this orientation, M has low salience as
a result of the low signal–noise ratio of Ng (Kingma et al., 2004),and as such it fails to “adequately” constrain (structure) the hapticstimulus array in vertically oriented wielding. This interpretation isreinforced by the observation that ratings in the vertical orientationwere (a) consistently lower, suggesting that the contribution of Mwas less intense, or (b) less influenced by changes in M, suggest-ing that M is more difficult to detect around the vertical, leading,in turn, to poorer differentiation among objects. It is possible thatI (which was the same for all objects in this set), in combinationwith M, was the basis of these judgments.3 In either case theobserved conditional dependency is consistent with prior observa-tions of a decrease in the salience of M when objects are wieldedvertically.
Last, these results appear to support our prediction that theinstructions provided to participants in Experiment 1 led them torate move-ability in terms of how difficult it was to move theobject in the local context of exploration.
Experiment 3: Perceiving How Fast an Object Can BeMoved
In Experiment 3, a third task was considered, in which wepredicted that both M and I would be implicated. The basic taskpresented to participants was that of rating how fast an objectcould be rotated. To avoid a repeat of Experiment 1, we elaboratedon this basic task so that an invariant object property was specifiedin the instructions. Participants were asked to rate how fast anobject could be rotated past the horizontal orientation if they wereto take hold of the object in a vertical orientation and were to rotateit as hard and fast as possible. Here, participants again attempt toprospectively perceive an action-relevant property through limitedexploration.
By relating these instructions to the mechanics, we can see howboth M and I are implicated. From Equation 4, when a fixed
3 To preempt, this possibility is addressed and ruled out in the movementanalysis presented later.
Figure 3. Results from Experiment 2. Ratings of how difficult partici-pants perceived it would be to hold particular objects, varying in M, in ahorizontal orientation. Ratings are shown as a function of orientation ofexploration.
199HAPTIC PERCEPTION OF ACTION-RELEVANT PROPERTIES
muscular torque is applied (comparable to the situation of a par-ticipant rotating the object as hard as possible at the limit of his orher action capabilities), the acceleration of the object (and there-fore the speed when the object reaches the horizontal) is dependentupon both M and I. Accordingly, we hypothesized that both M andI would be implicated in combination. On the basis of the resultsof Experiment 2 we predicted that reliance on M would again beconditional upon exploring the objects in a nonvertical orientation.
On the basis of consideration of the consequences of changing� in Equation 4, we originally hypothesized (see Experiment 1)that the implication of I would be affected by orientation ofexploration. Specifically, given the change in relative contributionof I and M to Nm over changes in orientation, we originallypredicted that I would be most salient in the vertical orientation.Following the apparent independence of Ni and Ng observed inExperiment 1, we alternatively hypothesized that Ni and Ng wouldmake independent contributions to Nm, and as such the implicationof I (part of Ni) may in fact be found to be independent of � (partof Ng).
Method
Participants. Twelve University of Connecticut studentswere recruited for Experiment 3. A group of 6 participants wereassigned to wield objects from Set 1; the other participantswielded objects from Set 2. Participant recruitment again mir-rored that of Experiment 1. Average grip strength was 50.1newtons (SD � 8.3).
Apparatus and stimuli. See Experiment 1.Procedure. Participants were instructed to rate how fast an
object could be rotated past the horizontal orientation if they wereto take hold of the object in a vertical orientation and were to rotateit as hard and fast as possible in a single movement. All proceduresfollowed the protocol outlined for Experiments 1 and 2. To recapthe paradigm, we did not allow participants to physically experi-ence the full range of motion between the vertical and horizontalorientations. More specifically, they were asked to estimate howfast the object could be moved past the horizontal on the basis ofthe exploration of the objects in a limited range around one of threeorientations (VO, DO, and HO).
Results and Discussion
Ratings of how fast an object can be rotated past the hori-zontal. Two repeated-measures ANOVAs were performed onthe ratings from the two groups assigned to wield objects fromeither Set 1 or Set 2. A 3 (orientation) � 5 (object) ANOVA forthe Set 1 group showed that variation in M, F(4, 20) � 8.65, p �.001, p
2 � .63, and the orientation of exploration, F(2, 10) �10.78, p � .001, p
2 � .68, had significant effects. The effects oforientation and object interacted, F(8, 40) � 5.89, p � .001, p
2 �.54. A post hoc pairwise test with a Bonferroni correction formultiple comparisons revealed that ratings for the VO differedfrom those for the HO ( p � .05), whereas DO was not signifi-cantly different from either HO or VO (see Figure 4, left panel). Aswas expected, this pattern of interaction approximates that ofExperiment 2, supporting the hypothesis of a shared informationalbasis for the tasks of rating of how difficult to hold and how fastcan you move. Consistent with our earlier interpretation that M is
more difficult to detect around the vertical, a simple effect ofobject was observed in the horizontal orientation, F(4, 20) �10.28, p � .001, and the diagonal orientation, F(4, 20) � 8.95, p �.001, but was not observed in the vertical orientation, F(4, 20) �1.99, p � .13. These results also further reinforced the conclusionthat the implication of M in dynamic touch–based judgmentsdepends on exploring objects in nonvertical orientations.
Although the strong similarities in the patterns of interactioninvolving variation in M over Experiments 2 and 3 suggest acommon informational basis, a visual comparison of Figure 3 andthe left panel of Figure 4 reveals distinct functions, indicating aclear effect of the intention to perceive a particular action-relevantproperty.
A second 3 (orientation) � 5 (object) ANOVA for the Set 2group showed that variation in I, F(4, 20) � 13.62, p � .001, p
2 � .73, and the orientation of exploration, F(2, 10) � 18.73, p �.001, p
2 � .79, had significant effects. This supported our hypoth-esis that M and I would be co-implicated in ratings. A post hocpairwise comparison with a Bonferroni correction for multiplecomparisons revealed that although ratings in each of the orienta-tions of exploration differed from one another ( p � .05; seeFigure 4, right panel), the effects of orientation and object inter-acted, F(8, 40) � 4.03, p � .001, p
2 � .45. Contrary to thehypothesis that the implication of I would be independent oforientation of exploration, a strong and systematic effect wasobserved. Contrasting the results for the Set 1 group, a simpleeffect of object was observed in the VO condition, F(4, 20) �4.37, p � .05, and the DO condition, F(4, 20) � 9.76, p � .001,but was not observed in the HO condition, F(4, 20) � 2.52, p �.073.
Part of the effect of orientation in the Set 2 group may beaccounted for if we assume (quite reasonably) the co-implicationof both M and I in ratings. Given this assumption, the observedmain effect of orientation for Set 2 may be due to the fixedcontribution of M over the object set (see Table 1), and theestablished orientation specific dependence of M. This conjectureis reinforced by the sameness of results apparent in the effect oforientation for Set 1 and Set 2. Whereas this may explain the maineffect of orientation, it does not account for the observed interac-tion effect. Consequently, it would seem to be the case that theimplication of I in this experiment depends upon orientation aswas originally hypothesized in Experiment 1 (and motivated on thebasis of interpretation of Equation 4).
Figure 4. Results from Experiment 3. Ratings of how fast participantsperceived they would be able to rotate objects, varying in M (left panel) orI (right panel) past the horizontal. Ratings are shown as a function oforientation of exploration. VO � vertical orientations; DO � diagonalorientations; HO � horizontal orientations.
200 HARRISON ET AL.
Exploratory accelerations. To investigate the possibility ofother contributing factors, an analysis of the recorded exploratorymovements was performed. In reference to Equation 1, the mostobvious candidate for investigation was taken to be angular accel-eration (a). To evaluate the degree to which objects were acceler-ated in each exploratory episode, we recorded accelerationsthroughout the trial and averaged them to provide a value of meanabsolute acceleration (|a|) for each trial. A 3 (orientation) � 5(object) ANOVA on recorded |a| values for the test object explo-ration in the Set 2 group showed that variation in I, F(4, 20) �11.35, p � .001, p
2 � .69, and the orientation of exploration, F(2,10) � 14.58, p � .001, p
2 � .75, had significant effects (seeFigure 5). A post hoc pairwise test with a Bonferroni correction formultiple comparisons revealed that accelerations in the VO andHO significantly differed ( p � .05). All other comparisons werenot significant.
We hypothesized that the effect of orientation on accelerationmay have been due to the increased physical demands (workrequired) in the experiment when objects were explored away froma vertical orientation. To investigate this possibility, we looked tosee whether individual differences in the capacity to do workaffected exploration. A simple regression revealed that changes inwielding acceleration could be accounted for by differences in therecorded grip strength of each participant (r2 � .58, p � .05). Thissuggested that exploratory wielding accelerations were influencedby individual differences in the capacity for generating neuromus-cular torque, and that the change in physical demands of wieldingwith orientation is likely to have affected some participants.
Motivated by this finding, that participant ratings were affectedby a requirement for “appropriately” accelerating the objects, inExperiment 4 we set out to directly manipulate exploratory accel-erations. Given Equation 1, angular acceleration should affect boththe contribution of Ni to Nm and the contribution of Ng to Nm.
Experiment 4: Does the Manner of Wielding Affectthe Haptic Perception of How Fast an Object Can Be
Moved?
To explicitly test the effect of magnitude of exploratory accel-erations upon ratings of how fast an object can be moved, we
directly manipulated wielding speed in Experiment 4. Althoughprevious research has suggested that the implication of I is unaf-fected by wielding speed (Solomon & Turvey, 1988), interpreta-tion of this result is complicated by the covariation of M and I inthe stimulus sets used. The independent contribution of M and Iwas considered in the present experiment. We hypothesized thatthe changes in exploratory accelerations incurred by the manipu-lation of wielding speed would affect judgments of how fast thewielded objects could be moved. Specifically, we predicted thatincrease in wielding acceleration would (a) increase the signal–noise ratio of Ni to Nm and therefore affect the implication of I injudgments and (b) decrease the signal–noise ratio of Ng to Nm andtherefore affect the implication of M in judgments. Because wield-ing speed was controlled, movement recording was not performed.
Method
Participants. Ten participants were recruited for Experiment4. A group of 5 participants was assigned to wield objects from Set1, and the remaining 5 participants wielded objects from Set 2.Each participant was selected on the basis of physical strength asassessed through a measure of grip strength obtained by a hand-held dynamometer. Physically strong graduate students and facultywere recruited to minimize possible fatigue effects. Average gripstrength was 55 newtons (SD � 7.3).
Apparatus and stimuli. See Experiment 1.Procedure. As in Experiment 3, participants were instructed
to rate how fast an object could be rotated past the horizontalorientation if they were to take hold of the object in a verticalorientation and were to rotate it as hard and fast as possible. As inExperiments 1–3, participants were again restricted by instructionto explore in horizontal, diagonal, or vertical orientations. Explor-atory accelerations were manipulated through the use of a digitalmetronome-pacing exploration. Participants were instructed tomove the object to the upper and then the lower boundary of aparticular orientation on successive beats. In the slow wieldingspeed condition, exploration was paced at 60 beats per minute(bpm; oscillation frequency of 0.5 Hz). In the fast wielding con-dition, movement was paced at 120 bpm (oscillation frequency of1 Hz).
Results and Discussion
Ratings of how fast an object can be rotated past the hori-zontal. Two repeated-measures ANOVAs were performed onthe ratings from the two groups assigned to wield objects fromeither Set 1 or Set 2.
A 2 (speed) � 3 (orientation) � 5 (object) ANOVA for the Set1 group again revealed that variation in M, F(4, 16) � 22.80, p �.001, p
2 � .86, and orientation of exploration, F(2, 8) � 81.75,p � .001, p
2 � .95, had significant effects. A post hoc pairwisetest with a Bonferroni correction for multiple comparisons re-vealed that ratings for the VO differed from those for the HO andDO ( p � .05), whereas DO and HO were not significantly differ-ent. The effects of orientation and object interacted, F(8, 32) �9.95, p � .001, p
2 � .71. Wielding speed systematically affectedratings (see Figure 6). The main effect of speed was significant,F(1, 4) � 7.88, p � .05, p
2 � .66. Faster wielding speeds werefound to be associated with lower ratings of how fast each object
Figure 5. Absolute acceleration of exploratory movements as a functionof I and orientation of exploration for Set 2 objects in Experiment 3. VO �vertical orientations; DO � diagonal orientations; HO � horizontal orien-tations.
201HAPTIC PERCEPTION OF ACTION-RELEVANT PROPERTIES
could be moved (fast � 114.6; slow � 129.5). The effects of speedand object interacted, F(4, 16) � 8.88, p � .001, p
2 � .69. Thisinteraction suggests that as wielding speed increased, the degree towhich objects were differentiated on the basis of variation in Mdecreased. This interaction reveals that the implication of M wasaffected by the manipulation of speed and that it was affected in away that was independent of the affect of the manipulation ofspeed upon the implication of I. A significant three-way interac-tion among speed, orientation, and object was also observed, F(8,32) � 2.44, p � .05, p
2 � .37, suggesting that as wielding speeddecreased, ratings made for vertically oriented objects were af-fected by changes in M to a greater degree (see Figure 6).
A second 2 (speed) � 3 (orientation) � 5 (object) ANOVA forthe Set 2 group showed that variation in I, F(4, 16) � 12.72, p �.001, p
2 � .76, and the orientation of exploration, F(2, 8) � 10.72,p � .001, p
2 � .73, had significant effects. Again, the effects oforientation and object interacted, F(8, 32) � 2.37, p � .05, p
2 �.37. No main effect of speed was observed. The effects of speedand I interacted, F(4, 16) � 5.51, p � .01, p
2 � .58. Contrary tothe results for Set 1, as wielding speed increased, the degree towhich objects were differentiated on the basis of changes in Iincreased rather than decreased (see Figure 6). The effects oforientation and speed interacted, F(2, 8) � 12.70, p � .01, p
2 �.76, and a marginal three-way interaction of speed, orientation ofexploration, and object I, F(8, 32) � 2.15, p � .06, p
2 � .35, wasobserved.
Consistent with the results of Experiment 3, these results sug-gest that both I and M are co-implicated in ratings of how fast anobject can be rotated. As was hypothesized, the implication of Iand M were found to be mediated by the exploratory accelerationsof the participants. This effect is most dramatic for judgmentsmade in the vertical orientation. In the vertical orientation, ratingswere constrained by I when objects were being moved quickly,and constrained by M when moved slowly. This suggests that
when objects are accelerated, I is more salient, and M is lesssalient.
Cross analysis: Are exploratory wielding motions specific tointention? In Experiments 2–4 the exploratory movements ofparticipants appeared to be constrained by the informational basisof the action-relevant property that they intended to perceive. Tothe extent that exploratory movements serve the purpose of mak-ing the appropriate invariant or combination of invariants “acces-sible,” we may be in a position to better understand the relationshipbetween the kinematics of exploration and the intention to perceivea particular action-relevant property. To this end, we comparedexploratory movements of the objects from Set 1 (variation in M)under the intention to rate either how hold-able an object would beif held horizontally (Experiment 2) or how fast an object could berotated past the horizontal (Experiment 3).
Are exploratory accelerations dependent upon intention toperceive particular action-relevant properties? A 2 (inten-tion) � 3 (orientation) � 5 (object) mixed-design ANOVA wasperformed with intention (“how hold-able?” or “how fast?”) as abetween-groups factor and orientation of exploration and object Mas within-group factors. Absolute acceleration was the dependentvariable. As might be intuitively predicted, exploratory accelera-tions were found to be greater for the intention to perceive how fastan object could be rotated, F(1, 12) � 21.86, p � .001. Orientationof exploration, F(2, 24) � 20.31, p � .001, p
2 � .63, was alsofound to have a significant effect. Most interestingly, a stronginteraction between orientation and intention was observed, F(2,24) � 12.55, p � .001, p
2 � .51 (see Figure 7). We claim that thisinteraction supports the conclusion that exploratory movementsare specific to a particular intention. Given the intention to per-ceive how fast an object could be rotated, participants were foundto increase object acceleration (suggesting exploration specific toI) in the vertical orientation. Comparably, no corresponding in-crease in exploratory accelerations was observed, given the inten-tion to perceive how hold-able.
Is the orientation of exploratory wielding dependent uponintention? A second 2 (intention) � 3 (orientation) � 5 (object)mixed- design ANOVA was performed with intention (“how hold-able?” or “how fast?”) as a between-groups factor and orientationof exploration and object M as within-group factors. Mean re-corded angle (M�) normalized to starting orientation (e.g., 0, 45,
Figure 7. Absolute acceleration of exploratory movements for objectswielded from Set 1 as a function of intention to rate how difficult to holdor how fast can object be moved. VO � vertical orientations; DO �diagonal orientations; HO � horizontal orientations.
Figure 6. Ratings of how fast an object can be moved as a function of Mand I and wielding speed. VO � vertical orientations; DO � diagonalorientations; HO � horizontal orientations.
202 HARRISON ET AL.
and 90) for test object exploration was the dependent variable. Nomain effect of intention was observed, and orientation, F(2, 24) �73.18, p � .001, p
2 � .86, again had a significant effect. Again,the effects of orientation and intention interacted, F(2, 24) � 8.21,p � .01, p
2 � .41. Examination of Figure 8 reveals the source ofthis interaction. Looking at the vertical orientation, the explorationof the test objects differs as a function of intention. Given theintention to rate how hold-able, participants biased their explora-tion away from the vertical in both the VO condition (M� �2.5°) and the DO condition (M� � 3.7°). This is consistentwith the assumption that M is most useful for assessing hold-ability, but only if accessed through exploration around a nonver-tical orientation. In contrast, for the task of “how fast,” the prob-ability distribution of exploration is closer to symmetrical aroundthe vertical, in the VO condition (M� � 0.75°) and the DOcondition (M� � 0.68°). These differences may be due to thepossibility that less severe orientation-dependent detection prob-lems are associated with I in vertically oriented exploration.
General Discussion
Mechanical Constraints on the Informational Basis ofDynamic Touch
For each of the three tasks investigated over the four experi-ments conducted, we evaluated the general claim that haptic per-ception of action-relevant properties of tools is constrained by themechanics of the body–tool system. Consistent with this claim,ratings of how difficult to move were found to depend on thephysical property of gravitational torque (Ng); ratings of howdifficult to hold were concluded to be constrained by the firstmoment of mass distribution (M); and judgments of how fastobjects could be rotated were found to depend upon the first andsecond moments of mass distribution (M and I, respectively). Ineach case, judgments were constrained by quantities of a simplephysical model of the investigated hand–object system outlined in
Equation 4. These findings are consistent with Kingma et al.’s(2002) conclusion that when the moments of mass distribution areindependently controlled in experimental stimuli, then the mechan-ical invariant M plays an important role in dynamic touch.
The results of Experiments 2–4 were found to support addi-tional hypotheses motivated by Kingma and colleagues’ account ofthe mechanical constraints on dynamic touch (Kingma et al., 2002,2004; van de Langenberg et al., 2006). This account predicts thatthe implication of either M or I depends upon the respectivecontributions (signal–noise ratios) of Ng and Ni to Nm in a partic-ular mechanical context (i.e., specific values for � and �). Con-sistent with this account, Experiments 2–4 revealed that the im-plication of M and I depended on � and �, and did so in a mannerpredictable on the basis of a simple model of the mechanics givenin Equation 4. These conclusions are consistent with an earlyconjecture by Solomon and Turvey (1988), with which they sug-gested that attunement to mechanical invariants of the body–toolsystem over major physical transformations may be the definingproperty of the haptic perceptual subsystem of dynamic touch.
For Gibson (1979), information is construed as a formlessinvariant under the transformation of a stimulus array. Informationin the case of the haptic perceptual subsystem of dynamic touchhas consequently been taken to be an invariant structure in thehaptic array (Kugler & Turvey, 1987; Turvey, Solomon, & Burton,1989), a structure made manifest over transforming patterns ofstresses and strains. It is consequently only in the presence ofappropriate transformations that a particular haptic invariant canbe picked up. Presently, we have found that both invariants (M andI) and major physical transformations (� and �) of the mechanicsare implicated in the search for the informational basis of dynamictouch. We believe that this suggests that both the informationalbasis of haptic perceptual invariants and the transformations of thehaptic array over which such invariants are made manifest are eachlawfully constrained by the mechanics of the system.
Although the above framework is consistent with both theresults and conclusions drawn from Experiments 2–4, it is unclear
Figure 8. Probability plots capturing the orientation of exploratory movements for both the standard and testobjects taken from object Set 1. Each bar in a distribution captures the proportion of time spent exploring in aparticular orientation over all trials in a given condition. Distributions are shown for exploratory movementsgiven the intention to rate either how difficult to hold each object would be if held horizontally (left) or how fasteach object could be moved (right). VO � vertical orientations; DO � diagonal orientations; HO � horizontalorientations.
203HAPTIC PERCEPTION OF ACTION-RELEVANT PROPERTIES
how it could accommodate the apparent implication of Ng ob-served in Experiment 1. As Lederman et al. (1996) pointed out, itis not apparent what the physical transformation would be underwhich Ng is invariant.4 Given the apparent independence of Ng andNi in participants’ judgments, it is uncertain how participants’responses in Experiment 1 could be predicted by Equation 4. Atpresent the potential implications of these results for future modelswould appear to be either that (a) an alternative formulation of themechanics constraining dynamic touch needs to be found, or that(b) the mechanics depend upon the intention to perceive a partic-ular property. Although the theoretical framing for this result isambiguous, it is nevertheless apparent that the observed indepen-dence of Ng from Ni may be pointing the way to a basic organizingprinciple of the haptic array.
Exploratory Movements and the Informational Basisof Dynamic Touch
Investigations of haptic exploration of hand-held objects haveled to the identification of distinct classes of exploratory proce-dures used by participants in order to perceive different classes ofobject properties (Klatzky & Lederman, 1992; Lederman &Klatzky, 1993). In the case of dynamic touch, analyses of mini-mally constrained wielding motions have revealed that the com-plex exploratory patterns are constrained by the intention to per-ceive a particular action-relevant property (Michaels et al., 2007).Consistent with this finding, in the present experiments the inten-tion to perceive a particular action-relevant property was found tosystematically constrain exploratory action. Given the intention toperceive how difficult to hold an object would be if held horizon-tally, rotational acceleration of exploratory movements was con-sistently low and the distributions of exploration angles weresystematically biased so as to be away from the vertical orientation(see Figures 7 and 8). In contrast, for participants given theintention to perceive how fast an object could be rotated, rotationalaccelerations were comparatively high, and the distributions ofexploration angles were comparably unbiased by orientation ofexploration. These results compliment the findings of similar in-vestigations (Michaels et al., 2007; Riley et al., 2002) in support-ing the hypothesis that the information specifying a particularaction-relevant or object property is exploited by distinct types ofexploratory movements (Turvey, 1988, 2007; Turvey, Carello, &Kim, 1990), or put more simply, that exploration is specific tointention.
A more detailed comparison of wielding movements withineach of the three considered wielding orientations led us to furtherconclusions: namely, that exploration is also constrained by themechanical invariants implicated in a given task. For the task ofperceiving how fast an object could be rotated (in which both Mand I were implicated), exploratory accelerations (the transforma-tion that increases the salience of I) were exaggerated in theorientation of exploration in which the salience of M was lowest.In contrast, no such patterning of exploratory accelerations wasevident for the task of perceiving how difficult to hold (a task inwhich M alone is implicated). This suggests that the dynamics ofthe exploratory procedures assembled in the service of dynamictouch (i.e., exploratory actions) are grounded in the informationalbasis of the to-be-perceived property.
Worthy of note at this point is that a simple measure of thehands’ capacity for grasping was found to predict how partici-pant’s exploratory accelerations were biased by the orientation ofexploration. Situated in an account of haptic perceptual invariantsand transformations, it is evident that individual differences inwork capacity (corresponding to some dimension of wielders’“effectivities”) may have direct implications for the potentialtransformations that can be performed, and as a result the potentialinformational support for dynamic touch.
An Affordance-Based Theory of Dynamic Touch
The emphasis of the present series of experiments was uponuncovering the informational basis of the action-relevant proper-ties of tools. From an affordance-based perspective, properties ofthe organism–environment system relevant to the control of action(affordance properties) are hypothesized to be specified by infor-mation (Gibson, 1979; Michaels & Carello, 1981; Turvey et al.,1981). Attempts to develop an affordance-based theory of dynamictouch have motivated a methodological commitment in which it isassumed that the mapping between affordance properties andspecifying information is a single valued function (Carello, 2004;Carello & Turvey, 2000).
One of the more surprising results obtained from this investi-gation was that, contrary to our predictions, judgments of objectmove-ability were not found to be perceived as an invariantproperty, as was indicated by previous research (Shockley et al.,2004). In contrast, ratings of move-ability were found to be morereadily equated with the task of perceiving how difficult to movean object would be in the local context of exploration. WhereasShockley et al. (2004) found move-ability to be specified by aspecific combination of mechanical invariants (m, S, and V),move-ability was presently found to be specified by the mechan-ical property that varied as a function of orientation with respect tothe gravitational vertical (Ng).
For terrestrial organisms, the potential for performing an actiondepends upon the body’s orientation with respect to the gravita-tional vertical. Moreover, the ability to take advantage of theforces for free generated by the body moving through a gravita-tional force field has been taken to be a basic principle of motorlearning (Bernstein, 1996). It should come as no surprise thathumans and animals would have a capacity for perceiving thosepossibilities for action that depend upon gravity and, moreover,that orientation-dependent affordances might be expected to play apart in an affordance-based theory of dynamic touch. The task inExperiment 1 of rating “how locally difficult to move an object is”is a basic example of this. In this experiment, participants’ ratingsof the move-ability explicitly depended upon orientation withrespect to the potential field of gravity.
Of consequence to the development of the current affordance-based account of dynamic touch is the apparent loss of the hy-pothesized 1-to-1 mapping (single-valued function) between affor-
4 This raises the possibility that not all transformations of the hapticarray are based in the mechanics of the body–tool system. Given thispossibility, future investigations may need to consider attempting to modelthe biomechanical constraints upon the haptic array, such as those imposedby the neuromuscular system when assembled so as to explore an objectwith the intention of perceiving a particular property.
204 HARRISON ET AL.
dance properties and specifying information. Instead, there is a1-to-many mapping between move-ability (the candidate affor-dance property) and the information specifying that property. Wesuggest that the term move-ability is not a “legitimate” affordanceproperty, in as far as it fails to unambiguously specify an action-relevant property. Experiments 2–4 were motivated by the need toreduce the apparent ambiguity of the term move-ability. In theseexperiments, move-ability was recast in Experiment 2 as “howdifficult to hold horizontally” and in Experiment 3 as “how fastcan the object be rotated past the horizontal using a specified styleof movement.”5 Each of these instructions continues to refer to anobjects “move-ability,” but does so in a way that more adequatelyspecifies a performable action.
The modification of instructions in Experiments 2–4 is consis-tent with ecological theorizing regarding the characterization of anintentional act (Shaw, Kadar, & Kinella-Shaw, 1994; Shaw &Kinsella-Shaw, 1988; Shaw, Flascher, & Kadar, 1995). For Shawand colleagues an affordance is a property of the environment thatsupports a goal-directed act. They suggest that goal-directed ac-tions performed at the ecological terrestrial scale are minimallycaptured through the specification of target and manner parame-ters. The target parameter captures an environmental target (inExperiments 3 and 4, the horizontal orientation); whereas, themanner parameter defines the kinetics of the action (holding still inExperiment 3 and moving as hard and fast as possible in Experi-ment 4). From this perspective, the intention to perform a goal-directed act selects target and manner parameters that constrain theinformation specifying the control of action. In turn, this specificinformation selects an affordance. The implications for anaffordance-based theory of the perception of the action-relevantproperties of tools are clear: This approach requires not onlyuncovering the informational basis of dynamic touch (althoughquite evidently, this is an enterprise unto itself), but also requirescorrectly identifying and instantiating in the perceiver the intentionto perceive a property of the body–tool system relevant to thecontrol of action (an affordance property).
Investigations into the informational basis of heaviness haveresulted in apparent contradictions. Whereas some studies haveimplicated the second moment of mass distribution in judgmentsof heaviness (e.g., Turvey et al., 1999), more recent studies haveimplicated the first moment (e.g., Kingma et al., 2004). Therewould appear to be two potential issues that might shed light onthese findings. First, the inconsistent application of the method ofindependent variation of all mechanical invariants across thesestudies makes it difficult to draw reliable conclusions. Second,resolving the apparent disparity may only be possible after appro-priately characterizing the findings in the context of an affordance-based theory of haptic perception. With regard to the second point,if one allows the assumption that the task of perceiving heavinessinstantiates an underspecified intention to perceive a particularaction-relevant property, then it becomes clear how these previ-ously contradictory results might be reconciled in a single account.In the case of Kingma et al. (2004), heaviness could be standingproxy for the action-relevant property of how difficult to hold (i.e.,Experiment 2), whereas in the case of Turvey et al. (1999),context-specific experimental design differences could have re-sulted in heaviness standing proxy for an action-relevant propertyspecified by qualities of the second moment of mass distribution(as in Experiment 3). Future investigations of dynamic touch will
benefit from both a careful consideration of the affordances inves-tigated and from adoption of methods of independently varying themechanical properties of the stimuli.
5 A similar conceptual move was made by Shockley et al. (2004) insuggesting that heaviness as a to-be-perceived property should be recon-ceptualized as move-ability.
References
Amazeen, E. L., & Turvey, M. T. (1996). Weight perception and the hapticsize–weight illusion are functions of the inertia tensor. Journal of Ex-perimental Psychology: Human Perception and Performance, 22, 213–232.
Bernstein, N. A. (1996). Levels of construction of movement. In M. L.Latash & M. T. Turvey (Eds.), Dexterity and its development. Mahwah,NJ: Erlbaum.
Bingham, G. P., Schmidt, R. C., & Rosenblum, L. D. (1989). Hefting fora maximum distance throw: A smart perceptual mechanism. Journal ofExperimental Psychology: Human Perception and Performance, 15,507–528.
Bongers, R. M., Michaels, C. F., & Smitsman, A. W. (2004). Variations oftool and task characteristics reveal that tool-use postures are anticipated.Journal of Motor Behavior, 36(3), 305–315.
Burton, G., & Turvey, M. T. (1990). Perceiving the lengths of rods that areheld but not wielded. Ecological Psychology, 2(4), 295–324.
Burton, G., Turvey, M. T., & Solomon, H. Y. (1990). Can shape beperceived by dynamic touch? Perception & Psychophysics, 48(5), 477–487.
Carello, C. (2004). Perceiving affordances by dynamic touch: Hints fromthe control of movement. Ecological Psychology, 16(1), 31–36.
Carello, C., Fitzpatrick, P., Domaniewicz, I., Chan, T. C., & Turvey, M. T.(1992). Effortful touch with minimal movement. Journal of Experimen-tal Psychology: Human Perception and Performance, 18, 290–302.
Carello, C., Grosofsky, A., Reichel, F. D., & Solomon, H. Y. (1989).Visually perceiving what is reachable. Ecological Psychology, 1(1),27–54.
Carello, C., Thuot, S., Anderson, K. L., & Turvey, M. T. (1999). Perceiv-ing the sweet spot. Perception, 28(3), 307–320.
Carello, C., & Turvey, M. T. (2000). Rotational dynamics and dynamictouch. In M. Heller (Ed.), Touch, representation and blindness (pp.27–66). Oxford, England: Oxford University Press.
Fitzpatrick, P., Carello, C., & Turvey, M. T. (1994). Eigenvalues of theinertia tensor and exteroception by the “muscular sense.” Neuroscience,60(2), 551–568.
Fonseca, S., & Turvey, M. T. (2006, June 22–24). Biotensegrity perceptualhypothesis: A medium of haptic perception. Paper presented at the NorthAmerica Meeting of the International Society for Ecological Psychol-ogy, Cincinnati, Ohio.
Gibson, J. J. (1962). Observations on active touch. Psychological Review,69, 477–491.
Gibson, J. J. (1966). The senses considered as perceptual systems. OxfordEngland: Houghton Mifflin.
Gibson, J. J. (1979). The ecological approach to visual perception. Boston,MA: Houghton, Mifflin.
Hajnal, A., Fonseca, S., Harrison, S. J., Kinsella-Shaw, J., & Carello, C.(2007). Comparison of dynamic (effortful) touch by hand and foot.Journal of Motor Behavior, 39(2), 82–88.
Hove, P., Riley, M. A., & Shockley, K. (2006). Perceiving affordances ofhockey sticks by dynamic touch. Ecological Psychology, 18(3), 163–189.
Kingma, I., Beek, P. J., & van Dieen, J. H. (2002). The inertia tensor versusstatic moment and mass in perceiving length and heaviness of hand-
205HAPTIC PERCEPTION OF ACTION-RELEVANT PROPERTIES
wielded rods. Journal of Experimental Psychology: Human Perceptionand Performance, 28, 180–191.
Kingma, I., van de Langenberg, R., & Beek, P. J. (2004). Which mechan-ical invariants are associated with the perception of length and heavinessof a nonvisible handheld rod? Testing the inertia tensor hypothesis.Journal of Experimental Psychology: Human Perception and Perfor-mance, 30, 346–354.
Klatzky, R. L., & Lederman, S. J. (1992). Stages of manual exploration inhaptic object identification. Perception & Psychophysics, 52, 661–670.
Kreifeldt, J. G., & Chuang, M. C. (1979). Moment of inertia: Psychophys-ical study of an overlooked sensation. Science, 206, 588–590.
Kugler, P. N., & Turvey, M. T. (1987). Information, natural law, and theself-assembly of rhythmic movement. Hillsdale, NJ: Erlbaum.
Lederman, S. J., Ganeshan, S. R., & Ellis, R. E. (1996). Effortful touchwith minimum movement: Revisited. Journal of Experimental Psychol-ogy: Human Perception and Performance, 22, 851–868.
Lederman, S. J., & Klatzky, R. L. (1987). Hand movements: A windowinto haptic object recognition. Cognitive Psychology, 19, 342–368.
Lederman, S. J., & Klatzky, R. L. (1993). Extracting object properties byhaptic exploration. Acta Psychologica, 84, 29–40.
Michaels, C. F., & Carello, C. (1981). Direct perception. EnglewoodCliffs, N J: Prentice-Hall.
Michaels, C. F., Weier, Z., & Harrison, S. J. (2007). Using vision anddynamic touch to perceive the affordances of tools. Perception, 36(5),750–772.
Pagano, C. C., Fitzpatrick, P., & Turvey, M. T. (1993). Tensorial basis tothe constancy of perceived object extent over variations of dynamictouch. Perception & Psychophysics, 54(1), 43–54.
Riley, M. A., Wagman, J. B., Santana, M.-V., Carello, C., & Turvey, M. T.(2002). Perceptual behavior: Recurrence analysis of a haptic exploratoryprocedure. Perception, 31(4), 481–510.
Shaw, R. E., Flascher, O. M., & Kadar, E. E. (1995). Dimensionlessinvariants for intentional systems: Measuring the fit of vehicular activ-ities to environmental layout. In J. M. Flach, P. A. Hancock, J. Caird, &K. J. Vicente (Eds.), Global perspectives on the ecology of human–machine systems, Vol. 1 (pp. 293–357). Hillsdale, NJ: Erlbaum.
Shaw, R. E., Kadar, E., & Kinsella-Shaw, J. (1994). Modelling systemswith intentional dynamics: A lesson from quantum mechanics. In K.Pribram (Ed.), Origins: Brain & self-organization (pp. 51–101). Hills-dale, NJ: Erlbaum.
Shaw, R. E., & Kinsella-Shaw, J. (1988). Ecological mechanics: A physicalgeometry for intentional constraints. Human Movement Science, 7(2),155–200.
Shockley, K., Carello, C., & Turvey, M. T. (2004). Metamers in the hapticperception of heaviness and moveableness. Perception & Psychophys-ics, 66(5), 731–742.
Shockley, K., Grocki, M., Carello, C., & Turvey, M. T. (2001). Somato-sensory attunement to the rigid body laws. Experimental Brain Re-search, 136(1), 133–137.
Solomon, H. Y. (1988). Movement-produced invariants in haptic explora-tions: An example of a self-organizing, information-driven, intentionalsystem. Human Movement Science, 7(2-4), 201–223.
Solomon, H. Y., & Turvey, M. T. (1988). Haptically perceiving thedistances reachable with hand-held objects. Journal of ExperimentalPsychology: Human Perception and Performance, 14, 404–427.
Solomon, H. Y., Turvey, M. T., & Burton, G. (1989a). Gravitational andmuscular variables in perceiving rod extent by wielding. EcologicalPsychology, 1(3), 265–300.
Solomon, H. Y., Turvey, M. T., & Burton, G. (1989b). Perceiving extentsof rods by wielding: Haptic diagonalization and decomposition of theinertia tensor. Journal of Experimental Psychology: Human Perceptionand Performance, 15, 58–68.
Turvey, M. T. (1988). Simplicity from complexity: Archetypal actionregimes and smart perceptual instruments as execution-driven phenom-ena. In J. A. S. Kelso, A. J. Mandel, & M. F. Shlesinger (Eds.), Dynamicpatterns in complex systems. Singapore: World Scientific.
Turvey, M. T. (1996). Dynamic touch. American Psychologist, 51, 1134–1152.
Turvey, M. T. (2007). Action and perception at the level of synergies.Human Movement Science, 26(4), 657–697.
Turvey, M. T., Burton, G., Amazeen, E. L., Butwill, M., & Carello, C.(1998). Perceiving the width and height of a hand-held object bydynamic touch. Journal of Experimental Psychology: Human Percep-tion and Performance, 24, 35–48.
Turvey, M. T., & Carello, C. (1995). Dynamic touch. In W. Epstein & S. J.Rogers (Eds.), Perception of space and motion (pp. 401–490). SanDiego, CA: Academic Press.
Turvey, M. T., Carello, C., Fitzpatrick, P., Pagano, C., & Kadar, E. (1996).Spinors and selective dynamic touch. Journal of Experimental Psychol-ogy: Human Perception and Performance, 22, 1113–1126.
Turvey, M. T., Carello, C., & Kim, N.-G. (1990). Links between activeperception and the control of action. In H. Haken & M. Stadler (Eds.),Synergetics of cognition (pp. 269–295). Berlin, Germany: Springer-Verlag.
Turvey, M. T., Shaw, R. E., Reed, E. S., & Mace, W. M. (1981). Ecologicallaws of perceiving and acting: In reply to Fodor and Pylyshyn (1981).Cognition, 9(3), 237–304.
Turvey, M. T., Shockley, K., & Carello, C. (1999). Affordance, properfunction and the physical basis of perceived heaviness. Cognition, 73(2),B17–B26.
Turvey, M. T., Solomon, H. Y., & Burton, G. (1989). An ecologicalanalysis of knowing by wielding. Journal of the Experimental Analysisof Behavior, 52(3), 387–407.
Turvey, M. T., Whitmyer, V., & Shockley, K. (2001). Explaining metam-ers: Right degrees of freedom, not subjectivism. Consciousness andCognition: An International Journal, 10(1), 105–116.
van de Langenberg, R., Kingma, I., & Beek, P. J. (2006). Mechanicalinvariants are implicated in dynamic touch as a function of their saliencein the stimulus flow. Journal of Experimental Psychology: HumanPerception and Performance, 32, 1093–1106.
van de Langenberg, R., Kingma, I., & Beek, P. J. (2007). Perception oflimb orientation in the vertical plane depends on center of mass ratherthan inertial eigenvectors. Experimental Brain Research, 180(4), 595–607.
Wagman, J. B., & Carello, C. (2001). Affordances and inertial constraintson tool use. Ecological Psychology, 13(3), 173–195.
Wagman, J. B., & Carello, C. (2003). Haptically creating affordances: Theuser-tool interface. Journal of Experimental Psychology: Applied, 9,175–186.
Wagman, J. B., & Taylor, K. R. (2004). Chosen striking location and theuser–tool–environment system. Journal of Experimental Psychology:Applied, 10, 267–280.
Received March 23, 2009Revision received September 16, 2009
![Final touch[1]](https://static.documents.pub/doc/80x56/631b55eca906b217b906699d/final-touch1.jpg)