Recognising emotions in biomechanical gait patterns with neural nets

$118

Journal o f Biomechanics 2006, Vol. 39 (Suppl 1)

7224 Fr, 11:00-11:15 (P51) 3D motions of trunk and pelvis during transfemoral amputee gait H. Goujon 1, E. Sapin 1, P. Fod62, F. Lavaste 1. 1Laboratoire de Biom6canique,

ENSAM-CNRS, Paris, France, 2Centre d'Etudes et de Recherche sur I'Appareillage des Handicap~s, Woippy, France Introduction: The conclusions of anterior studies concerning upper body movements of above knee amputee were limited because they involved very few patients [1] or because they analyzed only partially the upper body motion and not the coordination of trunk and pelvis [2]. Aim: The aim of the present study was to identify characteristics of upper body kinematics and torque transmission to the ground during locomotion in transfemoral amputee compared with healthy subjects. The influence of clinical factors as age, length of the stump, activity was also investigated. Method: A protocol was developed to record kinematics of trunk and pelvis by means of an optoelectronic Vicon ® system simultaneously with forces and moments measured by two force plates AMTI ®. 27 femoral amputees and a group of 33 asymptomatic subjects participated in this study. Angular ranges of motion of scapular and pelvic girdles were calculated in the three planes of space. The relative phase of the motion of these girdles in the horizontal plane is also evaluated. Statistical tests (t-test) were used to compare the amputated and the sound population and to determine the influential of clinical factors. Results and Discussion: Angular ranges of motion of upper body were globally increased for the amputees while walking velocity decreased. Specific patterns for above-knee amputees were found. The coordination between pelvic and scapular girdles was altered. Some correlations showed the links between parameters, particularly, the coordination was found to be dependant on walking speed. Clinical factors influenced this coordination and the mobility of upper part of the body. References [1] Cappozzo A., et al., Angular displacements in the upper body of AK amputees during level walking. Prosthetics and Orthotics International, 1982; 6: 131-138. [2] SjSdahl C., et al., Pelvic motion in trans-femoral amputees in the frontal and transverse plane before and after special gait re-education. Prosthetics and Orthotics International, 2003; 27: 227-237. 4990 Functional description of ground-reaction forces

Fr, 11:15-11:30 (P51)

T. Ertelt 1, R. Blickhan 1, H.-J. Ertelt 2. 1Science efMetien, Institute ef

Sportssciences, Friedrich-Schiller-University Jena, Gemany, 21SD Luft- und Raumfahrtkonstruktionen Stuttgart, Germany It is well known that the pattern of the ground-reaction force can be used for distinguishing between different activities such as walking, running, jumping or hopping. For a seemingly similar activity, such as hopping on the spot the pattern differs in dependence of the hopping frequency. Major dependencies can be explained by global lumped parameter models. However, the investigation of the relationship between muscular activation, segment oscillation and ground reaction force requires a more detailed characterisation of the force pattern. Impulse-tests with a sliding carriage on an inclined plane have shown five typical reaction-force patterns, also detected for hopping on the spot: symmetric single-modal (type I), positive mono-modal (type II), negative-mono-modal (type III), symmetric wavy (type IV) and multi-modal associated with plateau (type V). An unbiased and fairly accurate mathematical description of these patterns is possible by using a superposition of Gauss-functions. In our example it was possible to approximate the five typical reaction-force Ioadings by a superposition of only five Gauss processes. The 5x3 coefficients of this approximation had been determinate by a least-squares-fit. For practical reasons it is of advantage to reduce these 15 coefficients to a single one only, which can describe the shape of the ground-reaction force as exactly as possible. Therefore the 15 coefficients of each of the different reaction-force patterns are subjected to a discriminant-analysis in order to obtain the individual function coefficient and the group centroid of the discriminant function for each function pattern. This specific function value, the "Force-Quantification-Coefficient" (FQC), provides nondimensional information about the shape of the reaction force function. The normally used "standard method" for analysing a reaction-force function, provides information about the magnitude and the position of the extremevalue. In combination with the so called FQC-value, it is possible to get information about the shape of the force function as well. The FQC-value is therefore very informative and allows to classify the ground-reaction force pattern to one of the introduced pattern-types.

Oral Presentations 6712 Fr, 11:30-11:45 (P51) Insights on the neural control o f chaotic gait variability in the elderly M.J. Kurz. University of Houston, Houston, USA It is quite evident that human gait is variable. In fact, no two footfalls are exactly alike. Using a newly developed pseudoperiodic surrogation algorithm, we have rigorously demonstrated that variations in the gait pattern are not noise; rather they have a deterministic pattern. Our scientific explorations of the nature of these variations have demonstrated that the elderly have less certainty for selecting the appropriate neural pathways for stable locomotion. Less certainty appears to be related to a more random chaotic gait pattern in the elderly. However, it is not currently clear what neural control mechanisms alter the structure of the chaotic variability in the elderly. We initially used a passive dynamic walking model that has a chaotic gait pattern to explore this question. Our simulations indicated that hip joint actuations that assist the motion of the swing leg alter the structure of the chaotic gait pattern and that they can be used to transition to stable gaits embedded within the chaotic locomotive attractor. For example, a systematic increase in hip joint assistance promotes a period-4 gait pattern to bifurcate to a period-8, and a period-8 gait to bifurcate to a period-16. We explored if these concepts extend to human chaotic gait patterns with a custom built mechanical hip actuator that assisted the motion of the swing leg during gait. The largest Lyapunov Exponent (LYE) was used to quantify the chaotic structure of the gait pattern. An increase in the magnitude of the LyE would indicate a similar behavior as our model. Our results indicated that swing assistance systematically increased the magnitude of the LyE. Furthermore, we determined that the swing limb dynamics account for approximately 33% of the chaotic structure present in human locomotion. Hence, neural control of the swing limb plays an important role in the structure of the chaotic gait variability. Potentially, the lack of certainty observed in the elderly are due to inability to properly 'tune' the motion of the motion of the swing leg as a function of decreased hip strength. 4603 Fr, 11:45-12:00 (P51) Changes in movement patterns while walking due to sport activities at y o u n g e r age C. Haid 1, S. Fischler 1, A. Koller2. 1University Department ef Orthopedics, Medical University Innsbruck, Innsbruck, Austria, 2Department of Sports Medicine, Medical University Innsbruck, Innsbruck, Austria We were interested in the question whether sport activities lead to persistent changes in movement patterns in everyday activities like walking. Therefore we performed gait analysis with former athletic runners and with people who did no sports. The age of the 40 tested male subjects was between 25 and 49 years. One group consisted of former athletes; the other group did no sports in their youth and almost no sports in their later lives. According to Jacqueline Perry we used a walking speed of 4.8 km/h and we analysed the movement of the legs and the foot while walking. The gait analysis was performed on a treadmill using a Lukotronic ® gait analysis system. We found no significant changes except the behaviour of the ankle joint in the second part of the stand phase. In this part of the walking cycle significant smaller changes in the angle of the ankle joint occurred, while the resulting overall force shifted from the mid of the foot to the forefoot. We found differences in the step length and in the knee angle during the stand phase, but these differences did not reach the level of significance. We conclude, that doing sports in the youth has effects on movement patterns in all day living, may be for the whole life. Nevertheless it is not possible to decide, whether differences in lifestyle in the later years also have an influence on the movement patterns. 6390 Fr, 12:15-12:30 (P51) Recognising emotions in biomechanical gait patterns with neural nets D. Janssen, K. FSIling, W.I. SchSIIhorn. University of MEmster, Germany "How is it going?" Who ever has used this term has probably never thought about the connection between the meaning of the word "going" as a mode of locomotion and as an expression of inner feelings or emotions at the same time. Further evidence for the correlation of feelings and motor actions is provided by the etymology of the word emotion (e: "out of"; movere: "move"). Although some research in psychological or clinical context was done, questions mainly concentrated on subjective recognition of different facial (Ekmann and Friesen 1978) or body expressions (Montepare et al. 1999) of emotions. The utilization of optimizing clinical therapy by means of individual characteristics (SchSIIhorn et al. 2003) is mainly neglected so far. The aim of this study was to recognise emotional states of individuals in biomechanical gait patterns with artificial neural nets (ANN). Kinetic and kinematic gait data was derived from 25Hz Video and from a 1000 Hz force plate of 38 and 16 healthy subjects. The gaits were accompanied either by imagination of four emotional states (normal, happy, sad, and angry) or by listening to different types of music (excitatory, calmative, no music).

Track 3. Musculoskeletal systems and Performance - Joint ISB/ESB Track

After digitization and filtering the data was fed to following types of ANNs: supervised: MLP and LVQ; unsupervised: Self-Organized Map (SOM) and two coupled SOMS (2SOM). The results show a clear distinction between individuals in all nets and some partially clear indications of emotion-recognition. Consequences on clinical therapy will be discussed. References

Ekman E, Friesen W.V. (1978). Facial Action Coding System. Palo Alto: CA Consulting Psychologists Press. Montepare J., Koff E., Zaitchik D., Albert M. The use of body movements and gestures as cues to emotions in younger and older adults. Journal of Nonverbal Behavior 1999; 23(2): 133-152. Sch611horn W.I., Nigg B.M., Stefanyshyn D., Liu W. Identification of individual walking patterns using time discrete and time continuous data sets. Gait & Posture 2002; 15(2): 180-186.

3.7. Energetics of Human Locomotion 6639 Fr, 14:00-14:15 (P52) Constrained optimization o f gait: integrating cost and control J.E.A. Bertram 1, A. Gutmann 2. 1University ef Calgary, Calgary, Canada,

2 Comell University, Ithaca, USA Coordination of effective gait requires that an individual or animal select specific step lengths and frequencies to suit the circumstances under which it is moving. How are these parameters selected? In order to investigate the selection process, it is first necessary to determine the goal of movement. Using humans as a model system, we observed that individuals spontaneously adapt their gait to minimize the metabolic cost of locomotion when confronted with a constraint on any of the three main parameters of gait: speed, step length or step frequency. This can produce unique step length and frequency combinations for each constraint that are not used under any other circumstance (constraint), yet the subjects report that the gait does not feel unnatural. We have demonstrated for walking and running that gait parameter selection can be predicted by an optimization scheme that minimizes the cost of locomotion within the limits of the imposed constraints. This optimization scheme is referred to as the Constrained Optimization hypothesis. In this presentation we describe this hypothesis and its experimental verification in human walking and running. We also explore the general applicability of this hypothesis to human motion using data for motions that, while repetitive, stretch the boundaries of "normal" locomotion (locomotion for which the system could have been evolutionarily adapted) - namely hopping, cycling, and moving under simulated changes in apparent gravity. We find that in most cases control of gait is tightly linked to metabolic cost but this control allows for a remarkably plastic suite of expressed movement patterns. Movements outside those that can be considered adaptively normal sometimes defy the control system yielding less than optimal behavior, but these may well offer the best information on how the system coordinates gait and responds to specific functional challenges. 7064 Minimal model o f a Iocomoting bipedal animal

Fr, 14:15-14:30 (P52)

M. Srinivasan 1,2, A. Ruina 2. 1Mechanical and Aerospace Engineering,

Princeton University, Princeton, USA, 2 Theoretical and Applied Mechanics, Comefl University, Ithaca, USA We formulate the simplest mathematical model of a bipedal animal that is capable of an infinite variety of gaits - including the common gaits such as walking, running and the two kinds of skipping. The model bipedal animal has a point-mass upper body and mass-less telescoping legs [1,2]. We perform elaborate and accurate optimization calculations to determine which gaits use the least positive work to move a given distance at a given speed. Remarkably, the optimizations discover that the classic descriptions of walking and running - namely, inverted pendulum walking and impulsive running - are the work-optimal gaits at low and high speeds respectively. A third gait - w e call pendular run - minimizes the positive work per unit distance at intermediate speeds and large step-lengths. The simplicity of the model also permits some analytical work - we prove the optimality of walking and running for a small angle approximation of this minimal model of a bipedal animal. The analysis in [2] explicitly ruled out the gaits with double stance and the possibility of skipping. Here, having allowed such possibilities, we find that (1) skipping requires more work than either walking or running at all steplengths and speeds, (2) inverted pendulum walking minimizes work only in the limit of no double stance, with the impulsive push-off entirely before the impulsive heel-strike. The model in [2] did not account for tendon-elasticity, nor did it account for the cost of force production - two features thought to be important for the understanding of the cost of running [3]. We present a simple model that incorporates these features (a spring in series with the actuator and a cost

3.7. Energetics of Human Locomotion

$119

for the integral of force), and describe its predictions for the metabolic cost of running. References

[1] Alexander R.McN. J. Zool. Lond. 1980; 192: 97-117. [2] Srinivasan M., Ruina A. Nature 2006; 439: 72-75. [3] Alexander R.McN., Minetti A. J. Theor. Biol. 1997; 186: 467-476. 7563 Fr, 14:30-14:45 (P52) A particle collision model for calculating the energetic cost o f the step-to-step transition in human walking A. Ruina 1, M. Srinivasan 1,2. 1 Theoretical and Applied Mechanics, Cemell

University, Ithaca, NY, USA, 2Mechanical and Aerospace Engineering, Princeton University, Princeton, N J, USA Terrestrial legged locomotion requires repeated ground-reaction forces to redirect the body's motion from generally down to generally up. Examples of such down-to-up redirection include the push-off and heel-strike in human walking and the stance phase in human running. Here we approximate an animal as a point mass and the redirection as being accomplished by impulsive leg forces [1]. For a forward-moving animal these redirection impulses cause small-angle glancing "collisions". The word "collision" usually brings up imagery of two objects passively hitting each other, often losing some of the total kinetic energy in the process. However, we use the word in a more general sense, to describe the action of the leg-forces on the body - the leg-forces do both negative work and positive work. So mechanical energy can be gained or lost across a leg-impulse collision. Using freshman-physics-style calculations, we calculate the work absorbed and generated in these collisions. If we then approximate the metabolic cost as being proportional to positive muscle work, we can calculate the food energy needed for legged locomotion to a good first approximation. This talk will emphasize the application of this collisional model to the energetic cost of the step-to-step transition in human walking. Both push-off and heelstrike are modeled as collisions. The model has no extended double stance, and both collisions happen within an instant. Nevertheless, there are a number of scenarios: push-off happening entirely before heel-strike, heel-strike happening entirely before push-off, or the push-off and the heel-strike having various degrees of overlap. The energetics of these almost-simultaneous collisions are subtle. The energetic costs vary by a factor of 8 depending on the details of what happens within that "instant". References

[1] A. Ruina, Bertram and Srinivasan. J. Theor. Biol. 2005; 237(2): 170-192. 7580 Fr, 14:45-15:00 (P52) The energetics of controlling balance during human walking A.D. Kuo 1, J.M. Donelan 2. 1Dept. of Mechanical Engineering, University of

Michigan, USA, 2Dept. of Kinesiology, Simon Fraser University, Vancouver, Canada The control or stabilization of walking appears to exact a metabolic cost, and we consider here the potential mechanical or physiological attributes. The overall metabolic cost of walking may be due to physiological costs of using muscles to perform work or to produce force (Kuo, 2001). Feedback control may be necessary to stabilize lateral balance, which appears to be mechanically unstable during walking (Bauby & Kuo, 2000). This stabilization appears to be performed largely through adjustments to lateral foot placement. These adjustments, while of small amplitude, appear to cost energy, for two potential reasons related to the production of work or force. First, deviations in step width will entail higher step-to-step transition costs than a less variable gait with the same mean width. Second, the forced leg motions that produce those adjustments require muscle force, and therefore metabolic energy. Two experiments demonstrate these costs. In one study, we provided subjects with external lateral stabilization and found that their metabolic cost decreased (while controlling for mean step width and other variables; Donelan et al., 2004). In another study, we added lightweight obstacles to the medial side of subjects' ankles, forcing them to swing the legs laterally during each step to avoid each other (Shipman et al., 2002). Controlling for step width, we found metabolic cost to increase as a function of the additional induced lateral motion. These two costs of controlling balance may explain the preferred step width during human walking. References

Bauby C.E., Kuo A.D (2000). J Biomech 33: 1433-1440. Donelan J.M., et al. (2004). J Biomech 37: 827435. Kuo A.D. (2001). J Biomech Eng 123: 264-269. Shipman D.W., et al. Metabolic cost of lateral leg swing in human walking. 4th WCB, Calgary 2002, #1050.