ECOLOGICAL PSYCHOLOGY $(I),43-69 Copyright O 19%, Lawrence Erlbaum Associates, Inc. Is Limb Proprioception a Function of the Limbs' Inertial Eigenvectors? Christopher C. Pagano Clemson University, Clemson, SC and Centerfor the Ecological Study of Perception and Action, University of Connecticut, Storrs, CT Steven R. Garrett Center for the Ecological Study of Perception and Action, University of Connecticut, Storrs, CT M. T. Turvey Center for the Ecological Study of Perception and Action, University of Connecticut, Storrs, CT and Haskins Laboratories, New Haven, CT The abilityof humans to match the spatial orientation of occluded contralateral limbs was investigated. We hypothesized that this ability is tied to the inertial eigenvectors of each limb, which correspond to the limb's axes of rotational symmetry. In two experiments, the coincidence between the eigenvectors and the spatial axes of a forearm was broken with the expectation that the matching of occluded forearms would vary as a function of the eigenvectors. Overall, the angles at which the limbs were positioned were affected by the direction in which the eigenvectors were oriented by small appended masses. Discussion focused on the importance of physical invariants for perception, and their relation to hypothesized frames of reference for proprioception and motor concrol. Research has shown that one's impression of elbow position reflects forearm orientation in physical space rather than the elbow joint angle. Soechting (1982), for example, asked participants to point the right arm at a target and then reproduce - -- - Requests for reprints should be sent to Christopher C. Pagano, Department of Psychology, 418 Brackett Hall, Box 34151 1, Clemson, SC 29634-151 1. E-mail: [email protected]. 44 PAGANO, GARRETT, TURVEY with the left arm either the right elbow joint angle or the orientation of the right forearm. The standard deviation was significantly greater for matching joint angle than for matching limb orientation. Similar experiments by Worringham, Stelmach, and Martin (1987) also showed that participants are less accurate at perceiving joint angles than perceiving forearm inclination, and further demonstrated that errors in joint angle perception are biased toward matching forearm inclination. These results provide evidence that limb orientation sense is a function of spatial variables rather than the registration of joint angles, where these spatial variables may be defined relative to an absolute frame of reference anchored either in the body (e.g., the trunk) or the environment (e.g., gravitational or spatial vertical and horizontal axes; Soechting, 1982; Soechting & Ross, 1984). In additional experiments, participants were asked to point to a location in space, then reproduce that location by orienting the arm to the same target endpoint, or by indicating the endpoint with a handheld pointer (Helms Tillery, Flanders, &Soechting, 1991). The error in the pointing task was greater than in the limb-orienting task, indicating that the limb is perceived according to its orientation in space, rather than the location of its endpoint in space (Helms Tillery et al., 1991; see also Flanders, Helms Tillery, & Soechting, 1992).' Taken together, these experiments demonstrated that limb orientation perception is tied to spatially oriented variables independent of the joint angles, and can be quantified with reference to the environment in which the limb is embedded. Additionally, the orientation of the limb's longitudinal axis relative to the vertical has been suggestedas one such variable (Soechting, 1982; Soechting & Ross, 1984). Our research is directed at an alternative possibility first suggested by Pagano and Turvey (1995): that the eigenvectors of the limb's inertia tensor are candidate spatially oriented parameters for proprioception. The eigenvectors of the limb are specific to the orientation of the forearm's mass distribution in space, not the particular joint angles that gave rise to that orientation or the geometric position of the endpoint in space. It is possible that the eigenvectors of the inertia tensor account for the spatially oriented nature of proprioception observed in previous research. A complete exposition of this hypothesis requires a definition of the inertia tensor and an outline of a recent body of work that has established the sensitivity of the haptic system to the inertia tensor. This work has focused primarily upon participants' ability to perceive object properties by dynamic touch. DYNAMIC TOUCH AND THE INERTIA TENSOR When one grasps and wields an occluded object conjointly, there is perception of both aspects of the object and how the body segments are oriented relative to the 'The difference in error cannot be attributed to the use of a handheld pointer. In control conditions, participants pointed with equal accuracy to a virtual target using either the finger or pointer. PROPRIOCEPTION AND INERTIAL EIGENVECTORS 45 object held, and vice versa (e.g., Pagano, Carello, & Turvey, in press). Though different than their visual counterparts (see Burton, Turvey, & Solomon, 1990; Pagano & Tuney, l993), these concurrent haptic perceptions of holding must play a significant role in the control of actions involving the held object. Because of the obvious role of vision in the control and coordination of activity, the role of the haptic system is likely to go unnoticed. Without touch, however, the scaling of muscular forces to the dimensions of the limb segments and handheld objects could not achieve the fluency and autonomy required by these skilled behaviors (e.g., Forget & Lamarre, 1987; Ghez, Gordon, Ghilardi, Christakos, & Cooper, 1990; Ghez, Gordon, Ghilardi, & Sainburg, 1995; Teasdale et al., 1994). The kind of touching characterized by the wielding and hefting of objects is referred to as dynamic touch. It relates primarily to the tensile states of muscles and tendons as these tissues undergo deformation during exploratory and performatory activity (Gibson, 1966). Gibson explained that dynamic touch "is a perceptual system in its own right. More than any others, it is perception blended with performance, for the information comes from muscular effort" (1966, p. 128). Thus, dynamic touch is different from cutaneous (or tactile) touch in that muscle sensitivity plays a greater role in information detection than does skin sensitivity. It is the haptic system based primarily upon the "muscle sense," and is aptly described as kinesthesis employed in the perception of objects and surfaces.*Recent research has identified the inertia tensor Iij as the relevant mechanical quantity to which such perception is tied (e.g., Solomon & Turvey, 1988; see Turvey & Carello, 1995, for a review). Rotational motions about a futed point of the kinds characteristic of movements about a joint follow from where . is the matrix product and x is the vector cross product (Goldstein, 1980). In wielding an object (or moving a limb), the torque Ni, angular velocity 4, and angular acceleration q,vectors are coupled by the inertia tensor Is. Thus I,, is a parameter (a constant) that couples the varying torques and varying rotations. Ii, is represented mathematically by a matrix of numbers. The calculations of these components are done with respect to a rectangular coordinate system Oxyz. Patently, there are indefinitely many sets of three perpendicular axes xyz that can be anchored at a point of rotation 0 (located at the wrist, elbow, or shoulder). For each choice of Oxyz, the components of 4 differ, but the nature in which the tensor specifies properties of the object does not change. This is a basic property of tensors (Lovett, l989), inertia measured about one set of axes can be transformed to inertia measured about a different set of axes. In general, a tensor is a hypernumber-a 2 ~ ePagano, e Carello, and Turvey (in press) for a discussion of how dynamic touch relates to other types of touch (e.g., haptic touch and kinesthesis);see Fitzpatrick, Carello, and Turvey (1994), Pagano and Turvey (1995), and Turvey (1994) for discussions of the muscle sense and its role in dynamic touch. 46 PAGANO, GARRETT, TURVEY matrix of numbers that taken together express a physical state of affairs and transform in a particularly simple way (Moon & Spencer, 1986). Different coordinate systems Oxyz result in different tensorial components (numbers in the matrix change), but the manner in which the tensor transforms is such that the tensor as a whole (with all the components considered together) continues to d e h e the object property it quantifies. As a time-independent and coordinate-independent quantity, L, is an invariant rendering of the persistent material distribution of the limb and wielded object. With a given point of rotation, If does not change: It is constant property of the rigid object. It can, therefore, be used to quantify the information for perceiving the object's unchanging dimensions (e.g., Solomon, 1988; Solomon & Turvey, 1988; Turvey, 1994). Even when an object's motions occur about several joints, such as the wrist, elbow, and shoulder taken singly or in combination, an invariant rendering of I,, can be found that maps onto perceived object properties (Pagano, Fitzpatrick, & Turvey, 1993). The implication is that dynamic touch is tuned to the invariant parameters of the object's dynamics, rather than to the varying states (displacements,velocities) and torques (see Amazeen & Turvey, 1996). Represented mathematically by a symmetric 3 x 3 matrix (Goldstein, 1980), I,,'s diagonal terms (I,, I,, I,,)-referred to as moments of inertia-uantify the object's rotational inertia with respect to the three orthogonal axes of rotation. 1,s' off-diagonal terms (I,, G,,I,,, I,,, I,, I,), known as products of inertia, quantify the object's rotational inertia in directions perpendicular to the axial rotations and reflect the asymmetrical mass distribution of the object about the axes. 4 is a symmetric tensor; accordingly, I, = I,, and the nine components reduce to six. The moments and products of inertia comprising I,, are calculated with respect to a rectangular coordinate system Oxyz. Clearly, many triplets of perpendicular axes can be anchored at the point of rotation 0.For each choice of Oxyz, the components of IiJ differ. There is, however, a form of I, that is independent of Oxyz. This invariant form is with respect to the principal axes or eigenvectors of I,,. If the eigenvectors are chosen as the axes, then 4 is diagonalized: The principal moments of inertia, or eigenvalues, are on the diagonal, and all other entries are equal to zero (that is, there are no products of inertia) (Goldstein, 1980). The eigenvectors are the directions, with respect to 0, about which the object's resistances to rotation are distributed evenly; they are the symmetry or body axes with respect to the fixed point. These principal directions are fixed in the object and rotate with it. The eigenvalues (11,12, L) are the object's resistances to rotation about the respective eigenvectors (el, e2, ej), where one eigenvalue is the maximum moment of inertia for the object (II),one eigenvalue is the minimum moment of inertia (IJ, and the remaining eigenvalue is intermediate (12). Thus the eigenvectors are the axes of maximal (el), minimal (e4, and intermediate (ez) resistance to rotational acceleration. For any real object, unique eigenvectors and eigenvalues can be found about any given point of rotation (except for cases of degeneracy, such as when 12 = b ). A geometric representation of I,, is provided by the inertia ellipsoid. If all possible PROPRIOCEP'ITON AND INERTIAL EIGENVECTORS 47 axes p are passed through the point ofrotation 0 , and lengths OA, equal to (1p)-% are laid off on each axis, the locus of points A is an ellipsoid called the inertia ellipsoid. It is a quadratic surface with semi-axes of lengths (11)-95, (12)-%, and (13)-%(e.g., Lovett, 1989). Thus, its radii are inversely proportional to the eigenvalues of the inertia tensor, and correspond to the eigenvectors. In sum, the inertia ellipsoid is a geometric representation of the distribution "in the mean" of the mass of a body with respect to its point of rotation (Goldstein, 1980; Starzhinskii, 1982). In light of recent research, a useful summary is that Iij provides the domains for two sets of functions: One consists of the principal moments of inertia or eigenvalues that map onto perceived object "magnitudes," such as object length (e.g., Fitzpanick, et al., 1994; Solomon & Turvey, 1988), shape (Burton et al., 1990), and weight (Amazeen &Turvey, 1996); the other consists of the principal axes ofinertia or eigenvectors that map onto perceived object "directions," such as an object's orientation in the hand (Pagano & Turvey, 1992; Turvey, Burton, Pagano, Solomon, & Runeson, 1992), and the location of the hand relative to a wielded object (Pagano, Kinsella-Shaw, Cassidy, &Turvey, 1994). In other words, the magnitudes of the inertia ellipsoid, quantified by the eigenvalues, map onto perceived object magnitudes, and the orientations of the inertia ellipsoid, quantified by the eigenvectors, map onto perceived object orientations. EIGENVECTORS OF THE INERTIA TENSOR AS CANDIDATE PARAMETERS FOR PROPRIOCEPTION As demonstrated by Pagano and Turvey (1995), the understanding of dynamic touch summarized in the preceding may apply not only to how one perceives "attachments to the skin" (Gibson, 1966) such as tools and instruments, but also to the very traditional concern of how one perceives the body itself. Because the body, its limbs, and its limb segments are describable through liis, defined about the respective joint rotation "points," we can hypothesize that a person's knowledge about the dimensions and directions of his or her body and its appendages is given continuously by the eigenvalues and eigenvectors of the respective tensors. Simply put, this ability of dynamic touch to use the tissue deformation consequences of lii may have a proprioceptive role. Of particular importance to our hypothesis is the shape of limbs. Bodies of animals are composed of cylindrical parts, approximately round or elliptical in cross-section with a readily identifiable longitudinal axis (Wainwright, 1988). The spatial orientation of a cylindrical object can be defined by the orientation of the longitudinal axis. For example, the geometric orientation of two limb segments can be determined from the angle created by their longitudinal axes; this method is analogous to specifyingthe orientation of limb through joint angles or inclination relative to vertical (e.g., Soechting, 1982). An alternative and specifically dynamic 48 PAGANO, GARRETT,TURVEY definition of a limb's orientation assumes it to be in motion about an end point corresponding to a relevant joint. The spatial orientation of one of the object's eigenvectors about that point is coincident with the geometric orientation of the longitudinal axis. This is because the point of rotation for a limb segment, located at a joint, is typically on the longitudinal axis of that segment. The crux of the matter is that although the shape of a limb or object, and thus its longitudinal axis, is a geometric property, the haptic system is stimulated by mechanical parameters. The inertial eigenvectors are mechanical parameters, and thus may quantify properties of tissue deformation patterns. To reiterate, they are the directions of maximal resistance, minimal resistance, and intermediate resistance, to rotational acceleration. The coincidence of the longitudinal axis orientation and the orientation of the e3 eigenvector may allow one to know about the former by means of the latter through dynamic touch (e.g., motions of the limbs). That is, generalizing from what has been shown to be the case with handheld objects, the hypothesis is that one can know about the spatial orientation of a limb by detecting its eigenvectors. By breaking the coincidence between e3 and the longitudinal axis of the arm, the series of experiments described below is directed at the question: Is the perception of limb orientation a function of a limb's eigenvectors or of its geometric orientation? A remaining issue is that for properties of limbs and limb segments to be perceived according to 4, this argument must be applied to multiple I,+, corresponding to multiple parts of the body concurrently. As discussed, our research investigates the hypothesis that the orientation of an arm is perceived according to its inertial eigenvectors defined at a relevant point of rotation. This hypothesis should be considered part of a more general one: that dispositions of limbs and limb segments (including magnitudes and orientations, as well as their changes and rates of changes) are given by properties of Iij defined at each point of rotation. Given that any movement of the body and its limbs typically involves motions about several joints concurrently, a separate Ii, can be defined for each segment and its corresponding joint. That is, an Iij can be assigned to each point in joint space (the joints of the limbs and limb segments). Accordingly, there (Pagano et al., 1993).This 1,)-fieldhas proven useful is an inertia tensor field or 4-field to understanding the constancy of object properties perceived by wielding about varying degrees of freedom (Pagano et al., 1993). In ~rinciple,the hefield is sufficiently structured to be informative not only about postures, but also about transformations of postures. Like the contrast between the optic array (at a fixed point of observation) and the transforming optic array (at a changing point of observation; see Gibson, 1966, 197911986), the specifying ability of the I,,-field should be enhanced by transformations over time. The hypothesized Ii,-fieldfollows from other theories that propose proprioception is a function of mechanical parameters acting locally at each joint. It has been suggested, for example, that detection of gravitational torque N,acting at each joint may account for the perception of limb orientation (Worringham & Stelmach, 1985; Worringham, Stelmach, & Martin, 1987). The torque produced at a joint PROPRloCEPTlON AND INERTIAL EIGENVECTORS 49 due to gravity is proportional to the limb's angle relative to the gravitational vertical. N, is minimal when the limb's center of mass (CM) is aligned verticallywith respect to the joint, and maximal when CM is aligned horizontally. It is possible that individual joint torques and neighboringjoint torque patterns can contribute to the limb position sense (Worringham & Stelmach, 1985). That is, proprioception may be a function of an N,-field. There are several reasons, however, to expect that proprioception is a function of Iij,rather than N,.The first is that changing the load on a limb also changes N, at every position; thus, N , cannot provide reliable limb position information unless an inference process is used to compare N, at one angle to the N , of the same limb at some other angle (Worringham & Stelmach, 1985). In contrast, the eigenvectors of I,, are not affected by symmetrical loading, as described in the methods of Experiment 1. Secondly, N, acts only with respect to the gravitational vertical and fails, therefore, to address one's ability to perceive limb orientations in planes perpendicular to gravity. Even within the vertical plane, N,is not determinate: A mass oriented 45" above the horizontal, for example, has the same N, as an equal mass 45" below the horizontal. In contrast, Ivis indifferent to planes of motion. Significantly, Ii, is gravity independent. It is the three-space expression of the second moments of an object's mass (kg) distribution about a fixed point; it is not tied to an object's weight (kg m s - ~ ) . The finding that perceived limb orientation varies as a function of N , has not held up in all experiments investigating it. While Worringham and Stelmach (1985) found evidence in favor of N,, Soechting (1982) did not (see also Cohen, 1958). It is possible that the participants in Soechting's experiment detected the particular N, perturbations used, and may have ignored their impressions of torque (Worringham & Stelmach, 1985). If such was the case however, then the question arises as to what parameter governed the successful perception of limb orientation, if it was not N,. Relatedly, Carello, Fitzpatrick, Domaniewicz, Chan, and Turvey (1992) investigated participants' abilities to perceive the extent of occluded rods held steadily at different angles relative to the vertical. They found the first moment of the object's mass distribution to affect perceived extent more than N,, which is variable over angle. Like I,,, the first moment is an invariant mechanical parameter tied to the object's mass distribution. They hypothesized that the influence of angle on perception is less likely the result of the object's N, than of muscle deformation specific to the torsion of the limb's tissues needed to position the object at different inclinations relative to the limb. Following their lead, the present thesis investigates the possibility that 4, an invariant mechanical parameter tied to an object's or limb's mass distribution, provides the basis for one's proprioceptive abilities. In previous experiments directed at the eigenvectors' hypothesis (Pagano & Turvey, 1995), participants were asked to point to a visual target with an occluded arm while holding an object. Small masses were appended to the object, so as to break the coincidence between the arm's eigenvectors and spatial axes. In these experiments, participants were restricted to movements about the shoulder, with 50 PAGANO, GARRETT, TURVEY e3 of the arm being manipulated in a horizontal plane. The outcome was that, relative to participants' pointing in a symmetrical mass condition where e3 remained unaltered, participants pointed farther to the right when the left side was weighted (e3 rotated to the left), and farther to the left when the right side was weighted (e3 rotated to the right). This was the expected outcome from the hypothesis that the perceived orientation of the arm varies with the manipulation of the arm's eigenvectors. When, for example, the arm was pointed 2" to the left of a target, and the participant perceived it to be pointed directly at the target, then the participant perceived the arm in that configuration to be 2" to the right of its actual orientation. The results of these experiments indicated that the perception of limb direction was varied as a function of the limb's e3 eigenvector. In contrast to the Pagano and Turvey (1995) experiments, our experiments employed a matching procedure in which the perceived orientation of a forearm is reproduced with the contralateral forearm. Confirming the eigenvectors' hypothesis in the context of a matching task would be an important generalization. Velay, Roll, and Paillard (1989), for example, have contrasted the results from pointing tasks with those from matching tasks, and have concluded that the two manipulations may involve separate proprioceptive mechanisms.' As discussed earlier, the matching procedure has been employed to demonstrate that position sense at the elbow reflects forearm orientation, rather than elbow joint angle (Soechting, 1982; Worringham et al., 1987). In experiments by Soechting (1982) and Worringham et al. (1987), each forearm's e3 was coincident with the segment's geometric orientation, as specified by the orientation of its longitudinal axis. Thus, possibly their participants were actively matching the orientation of the forearm eigenvectors. In our experiments, masses were attached to a "splint" attached to the forearm such that the inertial eigenvectors were rotated upward, downward, or not at all. It was expected that participantswould orient the contralateral arm oriented above, below, or even with the manipulated target arm, respectively, to match the orientation of the forearm eigenvectors. Alternatively, if participants are sensitive to the geometric angle at the joint or the spatial orientation of the longitudinal axis, then no difference in limb matching should be observed as a function of the differential weighting. Regardless of the particular task in question (e.g., matching instead of pointing), it is expected that the dependence of perceived limb orientation on lowill remain the same. Velay et al. (1989) suggested that in both matching and pointing tasks "... the basic position sense is coded in terms of relative angular position in intrapersonal space" (p. 191). According to our hypothesis, this angular position may be defined in terms of the limb's inertial eigenvectors. '~lthoughimportant differences exist between the pointing task used by Pagano and Tunrey (1995) and that used by Velay et al. (1989),the fundamental similarity is that both required continuous knowledge of limb orientation in extrapersonal space. PROPRIoCEPnONAND INERTIAL EIGENVECTORS 51 EXPERIMENT 1 The eigenvectors of the target arm were manipulated by asking participants to hold an object with small appended weights. The object was cross-shaped, consisting of two wooden dowels attached perpendicularly at their midpoints. The object was held such that one dowel, the stem, extended backwards against the inside of the forearm coincident with the longitudinal axis of the forearm and forward from between the middle and ring fingers. The other dowel, the cross-piece, extended laterally from either side of the closed fist, and was held to remain perpendicular to the ground plane. When held in this manner, the object's stem points in the same direction as the forearm. That is, the longitudinal axis of the cyliindrical stem is parallel to the longitudinal axes of the forearm. Thus, the tasks "point the forearm in a particular direction" and "point the stem in a particular direction" are identical. Masses were added to the cross-piece to alter the eigenvectors of the target forearmrod system. The added masses did not alter the matching forearm's eigenvectors. The handheld objects, along with the three mass conditions used in Experiment 1,are depicted schematicallyin Figure 1. In each example in Figure 1, e3 is indicated by an arrow, the arm by solid lines, the object by dashed lines, and the masses by black squares. The el and ez eigenvectors also extend from 0,both being perpendicular to e3 and to each other, with el parallel and e2 perpendicular to the ground plane. The rotation of e, depicted in Figure 1 also involves an equal rotation of e2, \ ---- FIGURE 1 The e3 eigenvector of the right forearm was manipulated by appending small weights to a handheld object. For the matchingleft forearm, the weights always were in the symmetrical arrangement. --- ;: X I both rotating in the same direction by the same amount. The masses were attached to the object to reorient e3 of the limb and the object configuration upwards, downwards, or not at all (see Figure 1 top, lower left, and lower right, respectively). The rotation of ez and e3 occur at about 0,so that in effect, they rotate about el, which remains unaltered by the mass manipulation. In all cases, the geometric orientation of the forearm (as specified by elbow joint angles) remains unaltered. For one third of the trials in Experiment 1, a mass was attached to the cross-piece above the object's midpoint. The effect was to rotate e3 of the forearm an estimated 6.3" upwards (with a point of rotation in the elbow and the wrist joints fixed, and O" being coincident with the longitudinal axis of the forearm). In another third of the trials, a mass was attached below the object's midpoint. The effect was to rotate e3 of the forearm an estimated 6.3"downwards. In the remaining trials, two smaller masses were attached to the object, one on either side of the midpoint, such that e3 of the arm remained unaltered by the addition of the symmetrically weighted object, while the eigenvalues equaled those of the single mass configurations. Manipulating the eigenvectors in a vertical plane in this manner meant that over each experimental condition, (a) the limb's eigenvalues and overall mass remained invariant, and (b) the gravitational torque varied only slightly, thereby ruling out any possibility of observed effects being due to these parameters. Importantly, the top versus bottom positioning of the mass changed the arm's eigenvectors without changing the angle of the arm at the elbow or the orientation of the limb's longitudinal axis relative to the vertical. The eigenvalues and geometric orientation of the arm remained unaffected by the manipulation of the eigenvectors, and were similar in all conditions. In consequence, an effect of mass position would be contrary to the hypothesis that the perception of the forearm's direction is based solely upon the arm's geometric orientation. Methods Participants. Seven graduate students and staff members, all right-handed, associated with the University of Connecticut participated in Experiment 1 on a volunteer basis. Four were women and three were men. No participant had any foreknowledge of the specific hypothesis in question, nor of the experimental conditions in use. Materials. Two cross-shaped objects similar to those used in Pagano and Turvey (1995) were constructed out of oak dowels 32.5 cm in length with a radius of .6 cm. The cross-piece extended 16.25 cm to either side of the center of the metal brace, and the stem extended 8 cm forward and 24.5 cm backward from the center of the metal brace. The total mass of each object, excluding added weights, was PROPRIOCEPTION AND INERTIAL EIGENVECTORS 53 TABLE I Moments and Products of Inertia (g . an /1,000)for the Object and Object + Limb Combination About the Point of Rotation in the Elbow for Experiment 1 Mass Condition Object 200gU 100gUand 100gD 200g D Object + limb combination 200 g U 100gUand 100gD 200gD 298.3 298.3 298.3 350.1 350.1 350.1 52.3 52.3 52.3 93.0 0.0 -93.0 895.7 895.7 895.7 945.3 945.3 945.3 64.4 64.4 64.4 93.0 0.0 -93.0 Note. U = up; D = down. "L, = Irr = 0. 114.8 g. Masses were attached to one or both branches of the handeheld object in three configurations: 200 g above the midpoint, 100 g above and 100 g below the midpoint, or 200g below the midpoint. Masses were placed 15cm from the midpoint of the cross. The moments and products of inertia for the objects, as well as the objects + arm configuration, are given in Table 1. The I,, values for the the objects + arm configuration were diagonalized to arrive at the the limb's e3 orientation for each mass condition. Calculations revealed that placement of the masses were such that the target limb's e3 was rotated 6.3" up, 0°, or 6.3" down with respect to the arm's longitudinal axis when the object was held in the hand. The eigenvalues and geometric orientation of the arm were similar in all conditions. Apparatus. The participant was seated in a chair and blindfolded. The elbows were positioned on the chair's horizontal arms with the participant's upper arms oriented roughly 65" with respect to the horizontal. The participant's arms were bare around and below the area of the elbow. The objects were placed one in each hand, such that the stem extended backward against the inside of the forearm coincident with the longitudinal axis of the arm and forward from between the middle and index fingers. The cross-piece extended laterally from either side of the closed fist, and was held to remain parallel to the sagittal plane. The objects were secured to the participant's forearms with tape and Velcro straps. Kinematic data regarding the motions of the participant's arms were collected using a three-dimensional sonic digitizer (Science Accessories Corporation, Shelton, CT) and associated MASS kinematic analysis software (Engineering Solutions, Columbus, OH). To collect motion data on the oscillating limbs, high-frequency sound emitters (3 cm long x .5 cm wide) were attached to the distal tips of the handheld objects. The 54 PAGANO, GARRETT, TURVEY emitted sounds were detected by four microphones aligned in a plane perpendicular to the ground plane and parallel to the frontal plane of the participant. The sonic digitizer calculated the xyz coordinates of an emitter in three-space, and thus the distal tip of an arm object configuration, by detecting the distances from each microphone to the emitter. An emitter's signal was sampled at 91 Hz for the duration of each trial, passed through an A-D converter, and stored on the hard disk of a PC. Each individual trial lasted 30 sec. Given that only final positions were of interest, the data were down-sampled to 15 Hz before analysis. + Procedure. Throughout Experiment 1, the right arm served as the target arm and the left arm as the matching arm. Participants began half the trials with the target arm in the up (U) position (roughly 75' with respect to horizontal) and the matching arm in the down (D) position (roughly-30" with respect to horizontal) and the remaining trials with the target arm in the D position, and the matching arm in the U position. The experimenter began each trial by switching on a metronome that repeatedly sounded a triplet of tones, with 2 sec separating the three tones. This continuous repetitive pattern lasted for the duration of each trial. When the participant was ready, Tone 1 acted as the signal to move the target arm about the elbow to some angle intermediate between the positions of the target arm and matching arm. Tone 2 signaled the participant to move the matching arm to an angle such that the matching arm was parallel to the target (see Figure 2). Tone 3 signaled the participant to return both arms to their starting positions, completing the cycle of three tones with their corresponding movements. This cycle was repeated without interruption for the duration of the trial, with 2-sec intervals between each tone and each cycle. When one complete cycle of three tones was FIGURE 2 The configuration used in Experiment 1. Participants intended to match the left forearm with the right forearm. Participants were blindfolded. PROPIUoCEPTION AND INERTIAL EIGENVECTORS 55 completed successfully, the participants were asked to indicate if they felt that the task was being done correctly; if so, the sonic digitizer was switched on by the experimenter to begin data collection. The participant was instructed to sample randomly with the target the full range of angles between the target and matching arm's starting positions. To ensure that the full range of angles was sampled, the experimenter observed each trial and instructed the participant to use an upper, a middle, or a lower angle, if the corresponding area was being neglected. Two baseline trials were performed without the metronome: One with the target arm started in the U position, and one with the target arm started in the D position. The symmetrical mass condition was used in these trials. During the baseline trials, participants moved the forearms in the same manner as in the mals with the metronome, but did so at any pace they found comfortable. The shoulder and wrist joints remained fixed by the apparatus during every trial. At the beginning of each trial, masses were attached to the cross-piece of the target object in one of the three experimental configurations. Each of six different conditions (3 mass conditions x 2 target starting positions, U or D) was presented to the participant twice, in addition to two baseline conditions (one for each target starting position), for a total of 14 trials per participant. Four participants received the trials for the U target starting position before the trials for the D target starting position, and the three remaining participants received the target starting positions in the reverse order. Each participant received the mass conditions in a different randomized order. At the beginning of the experiment, the participant was given two practice trials using the starting positions for the first seven trials: the first with neither the blindfold nor added weights, and the second with the blindfold and both objects weighted symmetrically. After the first seven trials, the participant performed a third practice trial using the starting positions for the second seven trials, with the blindfold worn and both objects weighted symmetrically. Data were not collected during the practice trials. The participant rested the arms for about two minutes between trials, but was allowed to take longer breaks as needed to avoid fatigue. The experiment lasted about 45 min per participant. Participants received no feedback about their performance during the course of the experiment. Eigenvector calculations. The orientation ofes was computed for each arm + object combination using regression equations and procedures provided by Reynolds (1978; see also Chandler et al., 1975; Clauser et al., 1969) applied to the body dimensions of one representative participant (see Appendix for the regression equations and body dimensions used; for similar use of these equations; see also Barac-Cikoja & Turvey, 1993; Pagano & Turvey, 1995). From these regression equations, the mass, distance of segment CMs from the point of rotation in the elbow, and principal moments of inertia for the limb segments were computed. The parallel axis theorem then was used to transform the principal moments of inertia for each segment about its respective CM to moments and products of inertia about 56 PAGANO, GARRETT, TURVEY the elbow; these quantities were combined with those for the object and attached object masses to get the moments and products of Ii, about 0 for the limb combination of each mass condition (see Table 1). Each I,, was diagonalized to arrive at its eigenvalues and eigenvectors. The eigenvectors, expressed as coordinates in Oxyz, were transformed into angles about 0.In each case, el remained perpendicular to the longitudinal axis of the arm, and ez and e3 were reoriented by the same amount within a plane perpendicular to e l , where 00 would be coincident with the longitudinal axis of the arm. From these calculations, it was estimated that the mass conditions caused e3 of the arm to be oriented 6.3" above the arm's longitudinal axis, 00, or 6.3" below, or 6.3"' 0°, and -6.3" respectively. As Table 1 shows, the moments of inertia (I,, I,, I,,) remain essentially constant over the different mass conditions, while the products of inertia (I,) change considerably. The sign of I,, indicates in which direction e3 will be rotated to diagonalize the inertia tensor, and the magnitude of I,,, relative to the magnitudes of the moments of inertia, corresponds to the magnitude of this rotation. + Data reduction. The time series for each 30-sec trial consisted of 450 sets of xyz coordinates for the distal tip of each handheld object, one set every 67 msec. Each set of points was converted into an angle within the yz plane, with O" corresponding to the forearm horizontal, and +90° corresponding to the forearm oriented vertically upwards. The yz plane was parallel to the midsagittal plane, and passed through the point of rotation in the elbow. The angle for the right arm was subtracted from the corresponding angle for the left arm to get the relative angle of the two arms at each of the 450 points in time. This relative angle was 0" when the two arms were perfectly parallel, negative when the left arm (matching arm) was positioned lower than the right (target), and positive when the left arm was positioned higher than the right. The relative angles for only those points in time when the participant was matching the two arms were averaged to get each participant's mean relative angle for each condition. The relative angle when both arms were at their starting positions, or when the target was moved to an intermediate position and the matching arm was still in its starting position, was not included in this relative angle because the participant did not intend to match the orientations of the arms at those points in time. Results The relative angles for the two target starting positions and three mass conditions for each of the seven participants in Experiment 1are presented in Table 2. Overall, and " the mean relative angle for the U and D target starting positions were 4.4 -3.2" respectively. The mean relative angle for upwards, zero, and downwards rotations of e3 were -1.7", -3.9"' and -5.8" respectively. A 2 x 3 analysis of variance PROPRIOCEPTION AND INERTIAL EIGENVECTORS 57 (ANOVA) with within-subjects factors of Target Starting Position and Mass Configuration resulted in no main effect for target starting position (F < 1) but a significant main effectfor mass condition, F(2,12) = 11.5,p < .002. The interaction was not significant (F < 1). A Tukey honestly significant difference (HSD) test determined that the mean relative angle for upwards and downwards rotations of e3 were significantly different (P < .01), whereas other comparisons were not (P > .05). The lack of a main effect for target starting position suggests that participants were not biased to matching higher or lower depending on the direction in which the matching arm was moved to reach the orientation of the target arm. Inspection of Table 2, however, suggests that some participants displayed such a bias, although the direction ofbias differed across participants. Participant 1, for example, pointed higher when the target arm started in the upper position compared to when the target arm started in the lower position in all three mass conditions, whereas Participant 2 did just the opposite. The main effect for mass indicates that matching differed as a function of how the object was weighted. As predicted, compared to matching in the symmetrical mass condition, the trend was for the participants to orient the matching arm higher when the upper portion of the target arm was weighted, and lower when the lower portion of the target arm was weighted. Thus, the relative matching directions corresponded to the direction in which the eigenvectors were rotated by the added weight. Figure 3 depicts the relative angle for each of the seven participants as a function of e3. The data in Figure 3 have been normalized by subtracting the mean relative angle for each participant from that participant's relative angle values. In this way, Figure 3 represents the matching bias due to the mass manipulation, unaffected by each participant's individual bias to place the matching arm higher or lower. The mean relative angles were -1.2", 3.9", 0.2", -2.1°, -17.0°, -1.3", and -10.2", respectively,for Participants 1 through 7. Five of the seven participants tended to place the matching forearm in a more flexed (vertical) position than the target arm. This result is similar to those observed in previous experiments (Velay et al., 1989; Worringham and Stelmach, 1985), where participants also tended to place the matching forearm in a more flexed position than the target. A 2 x 2 ANOVA with within-subjects factors of target starting position and metronome condition performed on the baseline (metronome off)and symmetrical mass (metronome on) conditions resulted in no main effect for target starting position (F < l), or metronome condition, F(1,6) = 1.0,p = .35.The interaction was not significant (F < 1). The lack of main effect indicates that relative angles with the metronome on, with participants constrained to moving when signaled by the metronome, did not differ from relative angles with the metronome off, with the participants setting their own pace. To rule out the possibility that participants were matching Ng of the forearms, we investigated whether the relative angle of two limbs in the Above and Below mass conditions varied as a function of the orientation of the target arm. Because cross-shaped objects were used to manipulate mass in a direction perpendicular to TABLE 2 Relative Angle As a Function of the Three Mass Conditions and Two Starting Positions for the Target Arm Used for Each of the Seven Participants in Experiment 1 Target Mass Confimcrahbn Participant Target Starting Positia 1 UP 2 UP 3 UP 4 UP 5 UP 6 UP 7 UP Overall UP Above Symmetric Below Down Down Down Down Down Down Down Down FIGURE 3 Mean relative angle as a -8 -6 -4 -2 0 2 4 6 8 Eigenvector Angle (Deg) function of the eigenvector angles for the seven participants in Experiment 16 = .37x + 0 ) . PROPRIOCEPTION AND INERTIAL EIGENVECTORS 59 the longitudinal axis of the arm, the difference in Ng between the two limbs varied as a function of arm orientation. Specifically, in both the Above and Below conditions, the Ng for the two arms were identical when both were oriented horizontally, and the difference in Ng between the two arms increased as the arms were moved towards the vertical. The difference in Ng between the two arms can be predicted from the orientation of the target arm using simple regression (y = .004x .003, = .98, p c .0001, n = 131). It is thus expected that ifparticipants were matching Ng,the relative angle between the two limbs would vary as a function of the target angle. Of the fourteen comparisons tested with simple regression-7 x 2 (Participant X Condition: Above vs. Below)-four resulted in a significant rZ, but with none matching the expected function. For two of the four significant regressions, the relative angle between the two arms decreased as the arms became more vertical instead of increasing, as would be expected by the Ng hypothesis4 Thus it appears that the participants in Experiment 1 were not matching the Ng of the two forearms, but were biased towards matching the inertial eigenvectors. - EXPERIMENT 2 Experiment 2 was designed to replicate the findings of Experiment 1 with two important differences: First, the left arm served as the target, and the right arm served as the matching arm. As in Experiment 1,e3 of the right arm was manipulated, whereas e3 of the left arm remained unaltered. Thus in Experiment 2, e3 of the matching arm was manipulated, whereas in Experiment I, es of the target arm was manipulated. Second, the overall amount of mass attached to the target arm was varied. In half of the trials, two 100-g masses were attached symmetrically to the object held in the target arm as in Experiment 1, and no mass was attached in the remaining trials. In all cases, ej of the target (left) arm remained unaltered. This second manipulation was designed to further test the possibility that participants are matching Ng.According to the Ng hypothesis, the matching arm should be positioned more vertically relative to the unweighted target arm than when positioned relative to the weighted target arm. Method Participants. Seven graduate students and faculty (two women and five men) associated with the University of Connecticut participated in Experiment 2 he four significant simple regressions were: Participant 1 Below, y = .113x- 3.59 :( = .39;p < .01,n = 20);Participant 4 Above, y = .125x- 5.126:( = .21,p < .05,n = 20);Participant 4 Below, y = -.092x + .003 = .30,p < .05,n = 20);and Participant 7 Below, y = -.105x + 18.236:( = .26,p< .05,n = 20).Polynomial regressions also were tested, but the basic result remained unchanged. (t 60 PAGANO, GARRETT, TURVEY on a volunteer basis. All participants were right-handed, and none had any foreknowledge of the specific hypothesis in question or of the experimental conditions in use. All participants gave their informed consent prior to their inclusion in the study. Materials and apparatus. The cross-shaped objects and experimental arrangement from Experiment 1 were used. Procedure. The procedure from Experiment 1 was used, with the following alterations. The left arm was the target arm and the right arm was the matching arm. The metronome was used during all trials with a 3-sec separation between tones and with the individual trial duration extended to 45 sec. This allowed the participants a more comfortable pace. At the beginning of each trial, masses were attached to the cross-piece of the matching object in one of the three experimental configurations. During half of the trials, two 100-g masses were placed symmetrically on the cross piece of the target object, and no masses were attached to the target arm during the remaining trials. Each of the 12 different conditions (3 Mass Configurations for the Matching Arm x 2 Target Mass Conditions x 2 Target Starting Positions) was presented to the participant once, for a total of 12 trials per participant. Four participants received the trials for the U target starting position before the trials for the D target starting position, and the three remaining participants received the target starting positions in the reverse order. Each participant received the target arm mass conditions and matching arm mass configurations in a different randomized order. The experiment lasted about 45 min per participant. Participants received no feedback about their performance during the course of the experiment. Results As in Experiment 1, the angle for the right arm was subtracted from the corresponding angle for the left arm to arrive at the relative angle between the two forearms. Thus for Experiment 2, this relative angle was 0" when the two forearms were perfectly parallel, negative when the right (matching arm) was positioned higher than the left arm (target), and positive when the right arm was positioned lower than the left. The relative angles for those points in time when the participant was matching the two arms were averaged to get each participant's mean relative angle for each condition. The relative angles for the two target arm mass conditions and three matching arm mass conditions for each of the seven participants in Experiment 2 are presented in Table 3. Overall, the mean relative angle for the U and D target starting positions PROPRIoCEPnoN AND INERTIAL EIGENVECTORS 61 TABLE 3 Relative Angle As a Function of the Two Target Mass Conditions and Three Matching Arm Mass Configurations for the Target Arm Used for Each of the Seven Participants in Experiment 2 Mmhing Arm Mass Confguration Participant 1 2 3 4 5 6 7 Overall Target Mass Condith Above Symmetric Below Weighted Unweighted Weighted Unweighted Weighted Unweighted Weighted Unweighted Weighted Unweighted Weighted Unweighted Weighted Unweighted Weighted Unweighted were -4.0" and -1.3" respectively. The mean relative angle for the weighted and unweighted target arm conditions were -3.4" and -1.9" respectively. The mean relative angle for upward, zero, and downward rotations of e3 were -1.7', -3.9", and -5.a0, respectively. A 2 x 2 x 3 ANOVA with within-subjects factors of Target Starting Position, Target Mass Condition, and Matching Arm Mass Configuration resulted in no main effect for target starting position, F(l, 6) = 1.5, p > .05, a significant main effect for target mass condition, F(1, 6) = 7.0, p < .05, and a significant main effect for matching arm mass configuration, F(2, 12) = 17.9, p < .001. No interactions were significant: Target Mass Condition x Matching Arm Mass Configuration F(2, 12) = 1.2, p > .05 (all other Fs < 1). A Tukey HSD test determined that the relative angles for the downward rotation ofe3 were significantly different from those for the upward rotation of e3 (p < .01) as well as those for the symmetrical weighting (P < .05), whereas the difference between the upwards and symmetric conditions did not reach significance (P > .O5). The lack of a main effect for target starting position indicates that, as in Experiment 1, participants were not biased to matching higher or lower depending on the direction in which the matching arm was moved to reach the target arm orientation. The main effect for target mass condition indicates that target matching differed as a function of whether the target arm was weighted. The overall trend was for participants to place the matching arm more vertically than the target arm, and to do so to a greater 62 PAGANO, GARRETT, TURVEY degree when the matching arm was symmetrically weighted than when the target arm was unweighted. Thus, participants tended to match the limbs more closely when the mass (and thus the N,) of the two limbs differed. This result is contrary to the N, hypothesis, which predicted the opposite result. Note, however, that the eigenvectors hypothesis predicted no difference as a function of this manipulation. Tne main effect for the target mass configuration indicates that matching differed as a function of how the object was weighted. As predicted, compared to matching in the symmetrical mass condition, the trend was for participants to orient the matching arm lower when the upper portion of the matching arm was weighted, and higher when the lower portion of the matching arm was weighted. Thus, as in Experiment 1, the relative matching directions corresponded to the direction in which the eigenvectors were rotated by the added weight. Figure 4 depicts the relative angle for each of the seven participants as a function of e3. The data in Figure 4 have been normalized by subtracting the mean relative angle for each participant from that participant's relative angle values. The mean relative angles were -4.3", -9.0", -2.6", 2.6", -7.1°, -4.1°, and 5.9" respectively for participants 1 through 7. The mean relative angles indicate that five of the seven participants in Experiment 2 tended to place the matching forearm in a less flexed (more horizontal) position than the target arm. This result appears as contrary to that of Experiment 1, where five out of seven participants tended to place the matching forearm in a more flexed position than the left target arm. The results are similar, however, in that in the matching configuration, the right forearm tended to be more flexed than the left for 10 out of the 14 total participants, regardless of whether the right forearm was employed as the target arm (Experiment 1) or the matching arm (Experiment 2). FIGURE 4 Mean relative angle as a -8-6-4-2 0 2 4 6 8 Eigenvector Angle (Deg) hnction of the eigenvector angles for the seven participants in Experiment 2 0, = .28x 0). + PROPRIoCEPTION AND INERTIAL EIGENVECTORS 63 DISCUSSION The two experiments demonstrate that forearm matching is reliably biased by manipulations of the eigenvectors of the limb segment's inertia t e n ~ o rIn . ~ agreement with the research of Pagano and Turvey (1995), our results ~rovidefurther confirmation of the general hypothesis that the inertial properties of the limbs have proprioceptive consequences. This hypothesis also has been evidenced by the findings of Ghez, Gordon, and colleagues (Ghez et al., 1990, 1995; Gordon, Ghilardi, & Ghez, 1994) in respect to the guidance of limb movements, and is implicit in Hogan's "mobility tensor" theory of multijoint posture and movement control (Hogan, 1985; see also Mussa-Ivaldi et al., 1985). In both Experiments 1 and 2, the shift in perceived limb matching due to eigenvector manipulations was consistently less than predicted inboth experiments, as demonstrated by the regression slopes of .37 and .28 obtained in Experiments 1 and 2 respectively (see Figures 3 and 4). These slopes were similar to those observed in Pagano and Turvey's (1995) Experiment 3. As in the present experiments, participants in Pagano and Turvey's Experiment 3 were restricted to motions occurring primarily within a single plane. In experiments allowing free movement of the limb, the observed slope predicting pointing bias from e3 tended to be two to three times steeper (Pagano & Turvey, 1995, Experiments 1 and 2). The low slope observed in the present data may be due, in part, to restrictions placed on participants' exploratory movements during the experiments. A low slope obtained in cases of restricted movements, compared to those allowing participants to move more freely, may underscore the dynamic nature of parameters for proprioception. lois a parameter of rotational dynamics about a fixed point that is revealed during active exploration. Greater accuracy in position sense with active as opposed to passive movements has been demonstrated (e.g., Paillard & Brouchon, 1968, 1974), implicating a dynamic component originating within the muscles over and above signals from joint receptors. In studies by Paillard and Brouchon (1968, 1974), blindfolded participants were asked to move one arm to a position along a vertically fixed rod, then to match its elevation with the other arm. Positioning error was greater when the target arm was passively moved into position by the experimenter compared to when it was actively positioned by the participant. It was concluded that proprioceptive information originating in the discharge of muscle spindle receptors, brought into play by self-induced movement and absent during stabilized position of the limb, may contribute to the perception of a limb's spatial position (Paillard & Brouchon, 1974). The major distinction between the active and passive case is that the former is a case of exploratory behavior; the latter is not. In the experiments by Pagano and Turvey (1995) for - - ' ~ o t ethat this result means experimenters should be aware that any apparatus applied directly to the limb of their participanc (e.g., manipulandum, potentiometers, splints, etc.) may bias matching or pointing performance if the device alters the eigenvectors of the limb. 64 PAGANO, GARRETT, TURVEY example, the active exploration (i.e., the dynamics) occurred when the participant freely moved the limb in three-space before pointing, or when the participant moved the limb in a restricted manner within a single plane. The pointing behavior was found to be more strongly determined by the dynamic parameter I,, when less restricted exploration was allowed: the condition that better reflected natural movements. The less restricted wielding of Pagano and Turvey's Experiments 1 and 2 allowed participants to more fully explore the dynamics of the limb; consequently, Ii, was better revealed in that case. Thus, the dependency of limb proprioception on I,, observed in the present experiments already has been found to be stronger in tasks allowing a fuller range of movement. This finding may reveal a major shortcoming of the often used matching procedure: The limited range of motion allowed in the matching task is not adequate to reveal the full power of the dynamic parameters used in proprioception. Much work has been directed at identifying the coordinate system in which limbs are controlled (e.g., Soechting, Helms Tillery, &Flanders, 1990).A proposed model of the sensorimotor transformations involved in targeted arm movements assumes that the preferred coordinate system is shoulder centered (Flanders et al., 1992). The possible arbitrariness of the origin of this coordinate system has been commented upon, with different locations both within and outside the body advanced as competing candidates (e.g., Alexander, 1992; Blouin et al., 1992). In our view (see also Hayward, 1992), the paramount issue should not be the particular coordinate system in which parameters for proprioception are rendered (e.g., head centered vs. shoulder centered), but rather the nature of the parameters themselves (e.g., Iijvs.joint angles or arm inclination; or more generally, dynamic vs. geometric). In light of the transformation law of tensors discussed in our introduction (see also Lovett, 1989; Moon &Spencer, 1986), the origin for Ii, may be translated from the actual point of rotation to some other location. In doing so, no argument need be made with respect to the preferred origin used by participants in the perception and control of limbs and objects, because each origin provides a rendering of the same physical facts. With such transformations, Is computed about this new point will remain specific to the aspects of a limb or object's mass distribution of relevance to dynamic touch (Pagano & Turvey, 1992; Pagano et al., 1993). The origin of the coordinate system may be chosen by the scientist on the basis of which origin offers the best description of the phenomenon in question. The numbers from one coordinate system, for example, may make it easier to obtain quantities that enter a regression with, and thus better predict, a relevant dependent variable. The use of one coordinate system over another does not change the physical facts at issue; thus, it does not automatically imply a preferred coordinate system used by the nervous system (only the scientist's preferred coordinate system). In research by Pagano and Turvey (1992), for example, participants were required to perceive the orientation of an object in the hand by wielding. I,,about the actual point of rotation was transformed to I,, about the proximal end of the object. This transformation allowed for a rendering of 4 as an angular position in a coordinate system centered PROPRIOCEPTIONAND INERTIAL EIGENVECTORS 65 around the portion of the handheld object, permitting the experimenter to use lii to predict perceived object orientation. This does not require the conclusion that the same coordinate system and origin are used by the participants. Rather, it allows for the conclusion that participants are sensitive to tissue deformations specific to the eigenvectors of 4 when perceiving the orientation of handheld objects by wielding. These tissue deformations, like all physical facts, remain invariant to the particular coordinates used to measure them. Participants capitalize upon such physical facts during the act of perception; they need not heed the articular coordinates required by the experimenter for their discovery and quantification. Similarly, Pagano et al. (1993) found that when the same object is wielded about one of several joints, taken singly or in combination, the perceived magnitude of that object remains invariant. To best predict the performance of the participants, the liis computed about the actual points of rotation were translated to an 0 common to all wielding conditions. This, again, does not suggest a coordinate system within which the participants act, but rather an invariant property of the wielding dynamics to which the perceptual constancy is tied. In sum, the use of a particular coordinate system to describe behavior does not necessarily imply the use of that coordinate system by the participant to produce the behavior. To the contrary, the participant's perception may be specific to coordinate-free quantities. The types of parameters found to be relevant to perception (e.g., tensors) may set aside questions regarding coordinate systems. The eigenvectors of I,, provide an intrinsic frame of reference for any object or limb in rotation about a point of rotation. Although based upon the distribution of mass and revealed through the patterning of forces, this reference frame is spatial and thus, is commensurate with spatial frames used concurrently by vision and the sensorimotor system (Pagano & Bingham, 1995; for a discussion of transformations between such reference frames, see also Soechting & Flanders, 1992). The inertial eigenvectors are the object's axes of rotational symmetry, the axes about which all bearing forces are zero (i.e., there are no products of inertia). For example, e3 is the axis of minimal resistance to rotational acceleration for a given object. Thus, rotations about e3 require the least work. Soechting, Buneo, Herrmann, and Flanders (1995) demonstrated that in pointing tasks requiring multijoint limb movements, participants tend to use limb movements involving rotation about axes chosen to minimize work. They identified the longitudinal axis ofeach limb segment as candidate axes for the minimization of inertia-thus, the minimization of work. However, a complete analysis may be in terms of e3 for the entire multisegmented limb taken as a whole. Although e3 for a single limb segment typically is coincident with the longitudinal axis of that segment, this may not be the case for a multisegmented limb. With the forearm and upper arm taken together, e3 for the entire arm rotated about the shoulder is coincident with the longitudinal axis of the upper arm and forearm only when the elbow is at full extension (i.e. when the angle between the forearm and upper arm is 180"). As the forearm is rotated about the elbow towards a position of 90" relative to the upper arm, the e3 for the entire arm rotates 66 PAGANO, GARRETT, TURVEY away from the longitudinal axis of upper arm, and extends from the shoulder through a point along the forearm (see Turvey et al., 1992, for a related description of the eigenvectors for L-shaped rods). Future research should be directed at uncovering the extent to which the eigenvectors of multisegmented limbs may be used to perceive limb positions and to constrain the movement trajectories of both limbs and handheld objects. ACKNOWLEDGMENTS Experiment 1 formed part of a doctoral dissertation presented by Christopher C. Pagano to the University of Connecticut. This work was supported in part by National Science Foundation Grant SBR 93-0937 1, awarded to M. T. Turvey and Claudia Carello; a University of Connecticut Doctoral Dissertation Fellowship; an American Psychological Association Doctoral Dissertation Research Award; and U.S. Public Health Service Individual National Research Service Award lFS32NS09575-01, awarded to Christopher C. Pagano. We thank Claudia Carello and Endre Kadar for their contributions to this research, as well as Geoffrey Bingham, Michael Stassen, Dragana Barac-Cikoja, and an anonymous reviewer for comments made on an earlier version of this article. Figure 2 was prepared by Claudia Carello. REFERENCES Alexander, G. E. (1992). For effective sensorimotor processing must there be explicit representations and reconciliation of differing frames of reference?Behavioral Brain Sciences, 15,321-322. Amazeen, E. L., & Turvey, M. T. (1996). Weight perception and the haptic "size-weight illusion" are functions of the inertia tensor.Joumd of E x p k m t a l Psychologv: Human Perception and PerfOTrnaTlCe, 22,2 13-222. Barac-Cikqa, D., &Turvey, M. T. (1993).Hapticallyperceiving size at a distance.Joudof Experimental Psychology: General, 122,347-370. Blouin, J., Teasdale, N., Bard, C., Fleury, M. (1992) The mapping of visual space is a function of the structure of the visual field. Behavioral Brain Sciences, 15,326327. Burton, G., Turvey, M. T., & Solomon, H. Y. (1990) Can shape be perceived by dynamic touch? Perception Et Psydwphysics, 48,477-487 Carello, C., Fitzpatrick, P., Domaniewicz, I., Chan, T. C., &Turvey, M. T. (1992). Effortful touch with minimal movement. Journal of Expenmental Psychdogv: Human Perception and Petformance, 18, 290-302. Chandler, R. F., Clauser, C. E., McContville,J. P., Reynolds, H. M., &Young, J. W. (1975). Inuestigation of inertial pmpertaes of the human body, Find report, Apr 1, 1972 - Dec. 1974 (AMRL-TR-74-137). Dayton, OH: Aerospace Medical Research Laboratories,Wright-Patterson Air Force Base. Clauser, C. E., McContville, J. P., &Young, J, W. (1969). Weight, volume and center of macs of segments of the human body (AMRL-TR-69-70). NASA CR-11262. Dayton, OH: Aerospace Medical Research Laboratories, Wright-Patterson Air Force Base. PROPRIoCEPTION AND INERTIAL EIGENVECTORS 67 Cohen, L. A. (1958). Contributions of tactile, musculo-tendinous and joint mechanisms to position sense in human shoulder. Jacmal of Neurophysldogy,21,563-568. Fitzpatrick, P., Carello, C., &Turvey, M.T. (1994). Eigenvalues of the inertia tensor and exreroception by the "muscular sense." Neuroscience, 60,551-568. Flanders, M., Helms Tillery, S. I., & Soechting, J. F. (1992). Early stages in a sensorimotor transformation. Behavioral Brain Sciences, 15,309-362 Forget, R, & Lamarre, Y. (1987). Rapid elbow flexion in the absence of proprioceptive and cutaneous feedback. Human Neumbtdogy, 6,27-37 Ghez, C., Gordon, J., Ghilardi, M. F., Christakos, C. N., &Cooper, S. E. (1990). Roles ofproprioceptive input in the programming of arm trajectories. Cold Spring Harbor Symposia on Quantitative Bldogy, LV, 837-847 Ghez, C., Gordon, J., Ghilardi, M. F., &Sainburg, R. (1995).Contributionsof vision and proprioception to accuracy in limb movements. In M. S. Gazzaniga (Ed.), The cognitive neurosciences. Cambridge, MA: MIT Press. Gibson, J. J. (1966). The senses considered US perceptual systems. Boston: Houghton Miftlin Gibson, J. 1. (1986). The e c o w apprwh to visual perception. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. (Original work published 1979) Goldstein, H. (1980). Clussical mechanics. Reading, MA: Addison-Wesley Gordon, I., Ghilardi, M., & Ghez, C. (1994). Accuracy of planar reaching movements 11. Systematic extent errors resulting from inertial anisotropy. Experimental Brain Research, 99, 112-130. Hayward, V. (1992). Physical modeling applies to physiology, too. Behauioral Brain Sciences, 15, 342-343. Helms Tillery, S. I., Flanders, M., & Soechting, J. F. (1991). A coordinate system for the synthesis of visual and kinesthetic information. TheJournal of Neuroscience, 11, 770-778, Hogan, N. (1985). The mechanics of multi-joint posture and movement control. Bidogrcal Cybernetics, 52,315-331. Lovett, D. R. (1989). Tenrorpr@ertier of q s t n l s . Philadelphia: Adam Hilger. Moon, P., &Spencer, D. E. (1986). Theory of holm. Cambridge, England: Cambridge University Press. Mussa-Ivaldi, F. A., Hogan, N., Bizzi, E. (1985) Neural, mechanical, and geometric factors subserving arm posture. TheJoumal of Neuroscience, 5,2732-2743. Pagano, C. C., & Bingham, G. P. (1995). Spatial frames for motor control would be commensurate with spatial frames for vision and proprioception, but what of the control of energyflows? Behavioral and Brain Sciences, 18,773. Pagano, C. C., Carello,C. & Turvey, M. T. (in press). Exteroception and exproprioception by dynamic touch are different functions of the inertia tensor. Perceptwn 19Psychophysics. Pagano, C. C., Fitzpatrick, P., & Turvey, M. T. (1993). Tensorial basis to the constancy of perceived object extent over variations in dynamic touch. Perception E ) Psychophysics, 54,43-54. Pagano, C. C., Kinsella-Shaw, J. M., Cassidy, P. E., &Turvey, M. T. (1994). Role of the inertia tensor in haptically perceiving where an object is grasped. Joumal of Exprmentnl Psychology: Human Perception and Performance, 20, 276-285. Pagano, C. C., &Turvey, M. T. (1992). Eigenvectors of the inertia tensor and perceiving the orientation of a hand-held object by dynamic touch. Perception E ) Psychophysics,52,617424. Pagano, C. C., & Turvey, M. T. (1993). Perceiving by dynamic touch the distances reachable with irregular objects. Ecdogrcal Psychdogy, 5, 125-15 1. Pagano, C. C., & Turvey, M. T. (1995). The inertia tensor as a basis for the perception of limb orientation.Joumal of E q e m e d Psychology: Human Perception and PedDrmrrnCe, 21, 1070-1087. Paillard,J., &Brouchon, M. (1968). Active and passive movements in the calibration of position sense. In S. J. Freedman (Ed.), The neur@sychology of spatially oriented behavior (pp. 37-55). Homewood, IL: Dorsey. Paillard, J., & Brouchon, M. (1974). A proprioceptive contribution to the spatial encoding of position cues for ballistic movements. Brain Research, 71, 273-284. 68 PAGANO, GARRETT, TURVEY Reynolds, H. M. (1978). The inertial properties of the body and its segments. In E. Churchill, L. L. Laubach,]. T. McContville, &I. Tebbetts (Eds.),Anthropometric source bmk, Volume I: Anthropomrtry fw designers. (NASA Reference Publication 1024, pp. 1V-1-N-76). Washington, DC: NASA Scientific and Technical Information Office. Soechting, 1. F. (1982). Does position sense at the elbow reflect a sense of elbow joint angle or one of limb orientation? Brain Research, 248,392-395. Soechting, J. F., Buneo, C. A., Herrmann, U., & Flanders, M. (1995). Moving effortlessly in three dimensions: Does Donder's law apply to arm movements? The Journal of Neuroscience, 15, 6271-6280. Soechting, J. F., & Flanders, M. (1992). Moving in three-dimensional space: Frames of reference, vectors, and coordinate systems. Annud Review of Neumscience, 15, 167-191. Soechting, J. F., Helms Tillery, S. I., & Flanders, M. (1990). Transformation from head- to shouldercentered representation of target direction in arm movements. Journal of Cognitive Neumscience, 2, 32-43. Soechting,J.F., &Ross, B. (1984). Psychophysicaldetermination ofcoordinate representationof human arm orientation. Neuroscience, 13, 595404. Solomon, H. Y. (1988). Movement-produced invariants in haptic explorations: An example of a self-organizing, information-driven, intentional system. Human Movement Science, 7,201-224. Solomon, H. Y., &Turvey, M. T. (1988). Haptically perceiving the distances reachable with hand-held objects. Journal ofEx@imental Psychology: Human Perceptiun and Perjomrance, 14, 404-427. Starzhinskii, V. M. (1982) An advanced course oftheoretical m e c h s . Moscow: MIR. Teasdale, N., Bard, C., Fleury, M., Paillard, J., Forget, R., &Lamarre, Y. (1994). Bimanual interference in a deafferented patient and control subjects. In S. P. Swinnen, H. Heur, J. Massion, & P. Casaer (Eds.), Interlimb cmrdination: Neural, dynnmical, and cognitive constraints. San hego, CA: Academic. Turvey, M. T. (1994). From Borelli (1680) and Bell (1826) to the dynamics of action and perception. Journal of Spot and Exercise Psychology, 16, S128S157. Turvey, M. T., Burton, G., Pagano, C. C., Solomon, H. Y., & Runeson, S. (1992). Role of the inertia tensor in perceiving object orientation by dynamic touch.Journal of Etperimental Psychology: Human Perception and Performance, 18, 714-727. Turvey, M. T., & Carello, C. (1995). Eynamic touch. In Epstein W, Rogers S (Eds.), Handbook of perception and cognition, Vol. 5. Perception of space and motion (pp 401-490). New York: Academic. Velay, 1. L., Roll, R., Paillard, J. (1989). Elbow pasition sense in man: Contrasting results in matching and pointing. Human Movement Science, 8, 177-193 Wainwright, S. A. (1988).AXISand circumference: The cylindrical shape ofplants and animals. Cambridge, MA: Harvard University Press. Womngham, C. I., &Stelmach, G. E. (1985). The contributionof gravitational torques to limb position sense. Experimental Brain Research, 61,3842. Worringham, C. J., Stelmach, G. E., & Martin, Z. E. (1987). Limb segment inclination sense in proprioception. Experimental Brain Research, 66,653-658. APPENDIX: CALCULATING 1 j j FOR LIMB SEGMENTS Body dimensions used in the calculation of mass, center of mass, and moment of inertia of the limb segments (for a precise definition of all dimensions, see Clauser et al., 1969): Total body weight: 70 kg Elbow breadth: 6.5 cm PROPRIOCEPTION AND INERTIAL EIGENVECTORS 69 - Forearm length (radiale stylion): 25 cm Forearm circumference: 23 cm Wrist breadth: 6.0 cm Wrist circumference: 18 cm Fist diameter: 7.0 cm Regression equations used to calculate (a) the mass (kg) of the limb segments, (b) the distance (cm) of the limb segment and object centers of mass from the point of rotation in the elbow, and (c) the principal moments of inertia (g cm2)of the limb segments about CM (using total body weight in g; Chandler et al., 1975; Clauser et al., 1969; Reynolds, 1978): a. Forearm: .081 (wrist circumference) + .052 (forearm circumference) 1.650 = 1.004 Hand: .029 (wrist circumference) + .075(wristbreadth) + .031 (fist diameter) - .746 = .443 b. Forearm: .440 (forearm length) + .761 (wrist breadth) - 5.645 + upper arm length = 9.9 Hand: Forearm length + 6.0 = 3 1.0 c. Forearm:l, = 1.O84 (total body weight) - 4,8 12 = 7 1,068 I,, = 1.062 (total body weight) - 5,444 = 68,896 I,, = ,271 (total body weight) - 9,020 = 9,950 Hand& = I, = I,, = 215 (mass of hand) x [(fist diameter)/212= 2,17 1
© Copyright 2026 Paperzz