This research examined how infants in early stages of walking determine whether a hill is safe or risky for locomotion. A psychophysical staircase procedure provided estimates of infants' physical ability to walk up and down slopes (2-degrees to 36-degrees), and a 'go ratio' indexed the accuracy of their perceptual judgments. On average, perceptual judgments were scaled to walking ability on slopes. Children walked on safe slopes and balked on risky ones. For ascent, perceptual judgments were related to length of walking experience and walking skill on flat ground. Better walkers were also better perceivers. For descent, judgments neatly mirrored exploratory activity. Better perceivers explored hills more efficiently by hesitating, touching, and testing different positions on hills around the limits of their physical ability.

This article presents a developmental account of changes in the visual guidance of locomotion. In contrast to the impressive efficiency of adult locomotion, locomotor activity is not under prospective control at the onset of human mobility. Infants require extensive crawling and walking experience before responding adaptively to variations in the terrain. At the same time that they are learning to navigate in increasingly varied environments, their bodies and skills are rapidly changing. Learning generalizes from safe, flat ground to novel surfaces but it does not transfer to new methods of locomotion. We account for these patterns of generality and specificity of learning by focusing on the role of exploratory behavior in detecting threats to balance control.

The main purpose of this study was to investigate the development of the head stabilization in space strategy (HSSS) during various locomotor tasks in 3- to 8-year-old children and adults. The contribution of visual factors to the HSSS was also examined by applying peripheral visual restriction, stroboscopic visual motion cue restriction, and darkness. The kinematics of the head and trunk rotations (pitch, yaw, and roll) were analyzed by means of an optical TV-image processor (ELITE system). For each of the three angular components, an appropriate ''head anchoring index'' was defined in order to compare the HSSS with a head stabilization on the trunk strategy. Head-trunk correlation rates were also calculated for each angular component in order to evaluate the head-trunk stiffness. The development of head-trunk coordinations during locomotion under normal vision can be said to involve at least three main periods. The first period occurs from the age of 3 to 6 years, when the HSSS is adopted only while walking on the flat ground. While walking on narrow supports, children in this age-group rather tend to increase the head-trunk stiffness, especially at 6 years of age. The second period includes 7- to 8-year-old children. Children of this age become able to adopt the HSSS while walking on narrow supports. During this period, the HSSS is associated with a large decrease in the head-trunk correlations. Lastly, in adulthood the HSSS is commonly adopted but specifically involves the roll component associated with the lateral body oscillations while walking. Vision was found to have little influence on children's HSSS while walking, whatever their age. Moreover, darkness induces an increase in the efficiency of the HSSS in adults. This confirms that the HSSS is the most appropriate strategy available for dealing with an increase in the level of equilibrium difficulty and may reflect a ''top-down'' organization of the postural control while walking. These results also suggest that the HSSS may be mainly of vestibular origin and presumably serves to facilitate the visual input processing, particularly that of the motion and peripheral visual cues which are involved in the control of body equilibrium during locomotion.

On the basis of a review of the literature including the authors' own experimental studies, a model for the ontogenesis of balance control in children was developed. This qualitative ontogenetic model has to do with the equilibrium strategies built up by children in situations which are difficult but within the scope of their abilities. The model involves two functional principles. First: The stable reference frame on which the balance control is based can be either the support on which the subject is standing or the vertical gravity. When the frame of reference is the support surface, balance control is temporally organized either from the feet to the head (posture) or from the hip to the head (locomotion) (ascending organization). When the frame of reference is the vertical gravity, balance control is temporally organized from the head to the feet (descending organization). Secondly: Children gradually become able to master the various degrees of freedom which have to be controlled simultaneously during movement. For example, the head can be stabilized either on the trunk with the neck structures blocked (the en bloc mode of operation) or in space with the neck structures loose (the articulated model). Four main periods can be said to occur during the human life span. The first extends from birth up to the acquisition of the upright stance. This period is characterized by the development of postural responses along a cephalocaudal gradient. This chronological cephalocaudal progression with age of the ability to control several body segments may correspond to a descending temporal organization of unperturbed postural control, associated with an articulated operation of the head-trunk unit. The second period takes place from the acquisition of the upright stance up to around the age of 6. During this period, our experimental results are consistent with an ascending organization of balance control, from the feet to the head in postural stance and from the hips to the head in locomotion. This ascending organization is associated with an en bloc mode of head-trunk operation, which serves to minimize the degrees of freedom. The third period begins at around the age of 7 and continues up to an upper age-limit which is as yet unknown. It is characterized by the return to an articulated mode of head-trunk operation, whereby the head stabilization necessary for the descending temporal organization of balance control is ensured. Lastly, the fourth period, which is reached during adulthood, combines the main features of the third period with a new skill involving the articulated operation of the head-trunk unit along with a selective control of the degrees of freedom at the neck level.

This kinematic study investigated the effects of visual factors on the angular oscillations of the head and trunk during various locomotor tasks in 3- to 8-yr.-old children and adults. The oscillations of the head under normal vision were limited and changed little across ages. Oscillations of both head and trunk about the roll axis were the most sensitive to difficulty in maintaining lateral equilibrium. On narrow supports, the lateral oscillations of the trunk increased between the ages of 3 and 6 years, with a maximum amplitude at the latter age and then decreased up to adulthood, suggesting a transition phase around the age of 6 years. Visual restriction had little effect on the control of angular oscillations of the head in children or adults. On a narrow support in darkness, adults increased oscillations of the trunk but reduced oscillations of the head. It can be concluded that, regardless of the age, control of locomotor equilibrium aims at limiting the angular oscillations of the head. Vision seems to contribute little to stabilization of the head.

The authors of the present study tested the hypothesis that toddlers initiate lateral body stabilization first at the hip level in order to better control the center of gravity (CG), minimize the upper body destabilization induced by the movement of the feet, and prevent falls. Intersegmental coordination among the hip, the shoulder, and the head was investigated in toddlers during their Ist year of independent walking. The efficiency of locomotor balance control was examined in the frontal plane. An automatic optical TV image processor (ELITE system) was used in analyzing the kinematics of foot, hip, shoulder, and head rotations. For the hip, the shoulder, and the head, appropriate anchoring indices were defined so that comparisons could be made concerning the stabilization of a given body segment with respect to its external space and to the adjacent supporting anatomical segment. Cross-correlation functions were also used for extracting the temporal patterns of the body segments that occurred during locomotion and for obtaining some information about the coupling of 2 consecutive segments such as the head-shoulder and the shoulder-hip. First, hip stabilization in space appeared from the Ist week of independent walking and clearly preceded those of the shoulder and the head, suggesting an ascending progression, with age, in the ability of new walkers to control lateral balance during locomotion. Second, the hip movements occurred before the shoulder movements and the shoulder movements before the head movements, indicating that locomotor balance control is organized temporally in an ascending fashion, from the hip to the head. Third, the high values of the correlation coefficients, mainly between the head and the shoulder, were consistent with a global en bloc operation of the head-trunk unit.

A set of experimental studies showing how inter-segmental coordination develops during childhood in various locomotor tasks is reviewed. On the basis of these results and two functional principles (stable reference frame and control of the degrees of freedom of the body joints), we recently proposed an ontogenetic model for the sensorimotor organization of balance control in humans (5). In this model, the hypothesis was put forward that the two main modes of equilibrium control (ascending vs descending temporal organization) operate alternatively and are associated with either of two modes of head-trunk linkage ('en bloc' vs articulated) during four successive periods in the course of ontogenesis. The advantage of this model is that it is heuristic and therefore open to further improvements, including the generalization of these balance strategies to most of the posture-kinetic activities, the comparison between unperturbed natural balance and reactions to postural disturbances. Some improvements are suggested, and are illustrated by the studies of intersegmental coordination in new experimental tasks such as hops using one foot or two feet and the initiation of gait. These new results are consistent with the idea that mastery of the degrees of freedom to be controlled simultaneously during the movement improves gradually with age. Moreover, they support the concept of multiple reference frames which operate in a complementary manner or in concert to permit the most appropriate organization of balance control, depending on the environmental requirements.

This thesis presents a study of biped dynamic walking using reinforcement learning. Ahardware biped robot was built. It uses low gear ratio DC motors in order to provide freeleg movements. The Self Scaling Reinforcement learning algorithm was developed inorder to deal with the problem of reinforcement learning in continuous action domains. Anew learning architecture was designed to solve complex control problems. It usesdifferent modules that consist of simple controllers and small neural networks. Thearchitecture allows for easy incorporation of modules that represent new knowledge, ornew requirements for the desired task. Control experiments were carried out using asimulator and the physical biped. The biped learned dynamic walking on flat surfaceswithout any previous knowledge about its dynamic model.

This paper presents some results from a study of biped dynamic walking usingreinforcement learning. During this study a hardware biped robot was built, a newreinforcement learning algorithm as well as a new learning architecture were developed.The biped learned dynamic walking without any previous knowledge about its dynamicmodel. The Self Scaling Reinforcement learning algorithm was developed in order to dealwith the problem of reinforcement learning in continuous action domains. The learningarchitecture was developed in order to solve complex control problems. It uses differentmodules that consist of simple controllers and small neural networks. The architectureallows for easy incorporation of new modules that represent new knowledge, or newrequirements for the desired task.

The aim of this paper was to study, from a developmental perspective, the transient phase of gait during the period between the standing posture and the achievement of steady gait, using temporal and biomechanical parameters. Eight children who had been walking autonomously for 64 to 200 days were observed. A total of 64 sequences of steps were analyzed. A sequence of steps began with the child standing still and was executed on a large force plate. From the determination of the instantaneous velocity of center of gravity results establish that, unlike adults, progression velocity in children is not reached at the end of the first step, but after two to four steps. The gait initiation process does not depend on the steady state velocity, but results from an initial fall. The duration of the movement up to the end of the first step is independent of the progression velocity but depends only upon the body mass and moment of inertia of the children.

By using inverse dynamics and forceplate recordings, this study established the principle of oscillating systems and the influence of gravity and body parameters on the programming of the gait parameters, step frequency and length. Calculation of the ratio of the amplitude of the center of mass (CM) and the center of foot pressure (CP) oscillations yielded an equation and established a biomechanical constant, the natural body frequency (NBF). NBF appears to be an absolute invariant parameter, specific to human standing posture and gait in terrestrial gravity, which influences the relative positions of CM and CP and whose value separates the frequency bands of standing posture from those for gait. This equation was tested by using the experimental paradigm of stepping in place and then used in calculating the magnitude of CM oscillations during gait. The biomechanical analysis of the experimental observations allows one to establish the relationships between body parameters and gravity and the central programming of locomotor parameters.

After a brief presentation of the development of free walking interpreted as learning dynamical equilibrium, the problem of sensory integration in the process of walking development is discussed. A critical review of the role of vision in the development of posture-locomotor task is presented, along with recent test results on the development of the vestibular system. A final section presents the development of head stabilization and coordination as a necessary means to assist sensory integration. It is suggested that if sensory information is necessary to enhance posture-locomotor skills, a good mastery of walking is in turn necessary to increase the efficiency of sensory integration.

This article describes developmental changes in gait velocity and relates these changes to gait parameters that index postural stability (step width and lateral acceleration) and two components of velocity (cadence and step length).... From a developmental point of view, the data lead us to interpret early walking (the first 5 months) as a process of integration of postural constraints into the dynamic necessities of gait movement. A second phase, beginning after 4 to 5 months of independent walking, is considered to be a tuning phase characterized by a more precise adjustment of the gait parameters.

We investigated a pattern of involuntary lower extremity stepping-like movements which recently appeared in a subject with a 17-year history of neurologically incomplete injury to the cervical spinal cord. The movements were rhythmic, alternating and forceful, involved all muscles of the lower extremities and could be reliably evoked by lying the subject down (supine) and extending his hips. Once in this position, the movements continued spontaneously in the absence of external sensory perturbations, with a step-cycle duration of similar to 3.5 s (0.3 Hz). This rate could be either increased or temporarily halted by specific sensory inputs. Anaesthetizing the subject's tight hip joint, in which we found evidence of pathology, led to a marked attenuation of the stepping movements for a period of similar to 15 min. We believe that a combination of (i) preserved but limited supraspinal tonic facilitation, and (ii) abnormal, perhaps noxious afferent in flow from the subject's right hip to the spinal cord may underlie the appearance of this highly unusual and involuntary movement pattern. The striking similarity between the movement and EMG patterns in this subject and those described in many reports using the surgically reduced cat model suggests that we were witnessing the first well-defined example of a central rhythm generator for stepping in the adult human.

A new variable structure control law based on the Lyapunov's second method that can be used in trajectory planning problems of robotic systems is developed. A modified approach to the formulation of the sliding domain equations in terms of tracking errors has been presented. This approach possesses three distinct advantages: (i) it eliminates the reaching phase, (ii) it provides means to predict the entire motion and directly control the evolution of tracking errors, (iii) it facilitates the trajectory planning process in the joint and/or cartesian spaces. A planar, five-link bipedal locomotion model has been developed. Five constraint relations that cast the motion of the biped in terms of four parameters are developed. The new control method is applied to regulate the locomotion of the system according to the five constraint relations. Numerical simulation performed to verify the ability of the controller to achieve steady gait by applying the proposed control scheme. Bifurcation diagrams of the periodic motions of the biped are used to demonstrate the improvements in controller performance that arise from the application of the proposed method.

Dynamic biped walking is a difficult control problem. The design involves that of the controller as well as the gait. A typical design procedure involves tedious analysis, careful planning, and testing. The procedure is time consuming and the analysis is often based on some linearized model. Selection of control parameters and nominal trajectory determines the quality of control and in typical designs, some or all of the parameters are selected intuitively. The result is often not the best. If some special goal (such as to walk as fast as possible) is desirable, the design may become even harder. While the analytical approach is not easy, one possible alternative is to obtain the optimal or near-optimal design through parameter search. This study explores this approach. The design of the biped controller and gait is formulated as a parameter search problem, and a genetic algorithm is applied to help obtain the optimal design. Designs to achieve different goals, such as being able to walk on an inclined surface, walk at a high speed, or walk with a specified step size have been evolved with the use of the genetic algorithm. Simulation results show that the genetic algorithm (GA) is capable of finding good solutions.

This study was performed to test the hypothesis that the motion of the lower extremities when stepping over obstacles is governed by the criterion of minimum mechanical energy. The trajectories of the swing ankle during level walking and when stepping over obstacles of 51, 102, 153, and 204 mm heights were predicted and measured for eight healthy young adults. The predictions were made with a planar, seven-link linkage model based on the criterion of minimum mechanical energy using the method of dynamic programming. When stepping over obstacles, the predicted trajectories of the swing ankle were just high enough for the swing toe to clear the obstacles. The clearances measured between the obstacle and toe were significantly larger than those predicted. When stepping over obstacles the levels of work required to generate the measured trajectories were significantly larger (p less than or equal to 0.002) than those required to produce the predicted trajectories. The amount of work necessary to generate the measured or predicted trajectories increased linearly (significant at p less than or equal to 0.022) with obstacle height and, except when predicting the trajectory for the lowest obstacle, was significantly greater than that required when walking on level ground (p < 0.02). Thus, conservation of energy was found to become a less dominant criterion for governing the motion of the body when crossing obstacles than when walking on level ground.

We built a simple two-leg toy that can walk stably with no control system. It walks downhill powered only by gravity. It seems to be the first McGeer-like passive-dynamic walker that is statically unstable in all standing positions, yet is stable in motion. It is one of a few known mechanical devices that are stable near a statically unstable configuration but do not depend on spinning parts. Its design is loosely based on simulations which do not predict its observed stability. Its motion highlights the possible role of uncontrolled nonholonomic mechanics in balance.

Animal locomotion is generated and controlled, in part, by a central pattern generator (CPG), which is an intraspinal network of neurons capable of producing rhythmic output. In the present work, it is demonstrated that a hard-wired CPG model, made up of four coupled nonlinear oscillators, can produce multiple phase-locked oscillation patterns that correspond to three common quadrupedal gaits - the walk, trot, and bound. Transitions between the different gaits are generated by varying the network's driving signal and/or by altering internal oscillator parameters. The above in numero results are obtained without changing the relative strengths or the polarities of the system's synaptic interconnections, i.e., the network maintains an invariant coupling architecture. It is also shown that the ability of the hard-wired CPG network to produce and switch between multiple gait patterns is a model-independent phenomenon, i.e., it does not depend upon the detailed dynamics of the component oscillators and/or the nature of the inter-oscillator coupling. Three different neuronal oscillator models - the Stein neuronal model, the Van der Pol oscillator, and the FitzHugh-Nagumo model - and two different coupling schemes are incorporated into the network without impeding its ability to produce the three quadrupedal gaits and the aforementioned gait transitions.

The model of a legged locomotory system is optimized to ensure stable motion, reliable with respect to different initial conditions. The cost function suggested is based on the frequency of the model's loss of stability evaluated for randomly chosen initial leg positions. The optimized model can start from the majority of allowed leg configurations, demonstrating stable walking at low and moderate speeds. Furthermore, an acceleration procedure is designed to permit the model to pick up practically every speed and then walk successfully.

The muscle force sharing problem was solved for the swing phase of gait using a dynamic optimization algorithm. For comparison purposes the problem was also solved using a typical static optimizaton algorithm. The objective function for the dynamic optimization algorithm was a combination of the tracking error and the metabolic energy consumption....

Diedrich and Warren (1995a) proposed that gait transitions behave like bifurcations between attractors, with the relative phase of the leg segments as an order parameter and stride frequency and stride length as control parameters. In the present experiments, the authors tested the prediction that manipulation of the attractor layout, either through the addition of load to the ankles or through an increase in the grade of the treadmill, induces corresponding changes in the walk-run transition. As predicted, the load manipulation shifted the most stable walk and the transition to lower stride frequencies. In contrast, the grade manipulation shifted the most stable walk and the transition to shorter stride lengths. Other features of the dynamic theory were also replicated, including enhanced fluctuations of phase and systematic changes in stride length and frequency at the transition. Overall, in these experiments a shift of the attractors in control parameter space yielded a corresponding shift of the transition.

Non-patterned electrical stimulation of the posterior structures of the lumbar spinal cord in subjects with complete, long-standing spinal cord Injury, can induce patterned, locomotor-like activity, We show that epidural spinal cord stimulation can elicit step-like EMG activity and locomotor synergies in paraplegic subjects. An electrical train of stimuli applied over the second lumbar segment with a frequency of 25 to 60 Hz and an amplitude of 5-9 V was effective in inducing rhythmic, alternating stance and swing phases of the lower limbs. This finding suggests that spinal circuitry in humans has the capability of generating locomotor-like activity even when isolated from brain control, and that externally controlled sustained electrical stimulation of the spinal cord can replace the tonic drive generated by the brain.

In the last years it has become possible to regain some locomotor activity in patients suffering from an incomplete spinal cord injury (SCI) through intense training on a treadmill. The ideas behind this approach owe much to insights derived from animal studies. Many studies showed that cats with complete spinal cord transection can recover locomotor function. These observations were at the basis of the concept of the central pattern generator (CPG) located at spinal level. The evidence for such a spinal CPG in cats and primates (including man) is reviewed in part 1, with special emphasis on some very recent developments which support the view that there is a human spinal CPG for locomotion.

The utilization of passive dynamics to control the swing trajectory is one mechanism which serves to minimize energy costs during locomotion, in addition to reducing the complexity of the neural control. In a reactive situation (e.g. trip or slip during walking), the energy cost may not be a major determinant of the locomotor activity as there is a need for quick corrective action under the threat of a fall. Therefore, we addressed the following question: does the nervous system utilize passive dynamics during the reactive control of locomotion? An unexpected mechanical perturbation was applied to the foot during early and late swing during walking. Video data were input into an inverse dynamics routine to obtain the joint moment and mechanical power profiles and to partition the joint moments into active and passive components. The nervous system still utilized the passive dynamics of the effector system; active control of a single joint, the knee joint, passively facilitated the flexor action at the proximal hip and distal ankle joint following the early swing perturbation. The minimization of the mechanical energy cost was not a major determinant for this task since the total mechanical work during the perturbed steps was greater than during normal steps. A neuromuscular constraint was observed following the late swing perturbation; the active control of the hip and knee joints were increased but the magnitude of the hip extensor/knee flexor moment was invariant and equal to 1.6. The intralimb dynamics identified during these responses may serve to simplify the complexity of the active control of the nervous system.

This research provides converging evidence that infants use exploratory activity to differentiate slant around a horizontal axis before they relate information about slant to consequences for locomotion. In Experiment 1, 14-month-old toddlers walked down safe, shallow 10 degrees hills and slid down or avoided risky, steep 36 degrees hills when height of the hills was held constant. Results indicate that judgments were based an slant. In Experiment 2, 9-month-old crawling infants explored shallow 10 degrees and steep 30 degrees slopes differentially in a nonlocomotor task. Exploration was similar to previous locomotor research with full-size hills, even though crawlers plunged headlong over both shallow and steep hills in the earlier study.

If an oscillating neural circuit is forced by another such circuit via a composite signal, the phase lag induced by the forcing can be changed by changing the relative strengths of components of the coupling. We consider such circuits, with the forced and forcing oscillators receiving signals with some given phase lag. We show how such signals can be transformed into an algorithm that yields connection strengths needed to produce that lag. The algorithm reduces the problem of producing a given phase lag to one of producing a kind of synchrony with a ''teaching'' signal; the algorithm can be interpreted as maximizing the correlation between voltages of a cell and the teaching signal. We apply these ideas to regulation of phase lags in chains of oscillators associated with undulatory locomotion.

Human runners adjust the stiffness of their stance leg to accommodate surface stiffness during steady state running. This adjustment allows runners to maintain similar center of mass movement (e.g., ground contact time and stride frequency) regardless of surface stiffness....

This autonomous biped walking control system is based on reactive force interaction at the foothold. The precise 3D dynamic simulation presented includes: 1) a posture controller which accommodates the physical constraints of the reactive force/torque on the foot with quadratic programming. 2) a real-time COM (center of mass) tracking controller for foot placement, with a discrete inverted pendulum model. 3) a 3D dynamic simulation scheme with precise contact with the environment. The proposed approach realizes robust biped locomotion because environmental interaction is directly controlled. The proposed method is applied to a 20 axes simulation model, and stable biped locomotion with velocity of 0.25 m/sec and a stepping time of 0.5 sec/step is realized.

We demonstrate that an irreducibly simple, uncontrolled two-dimensional two-link model, vaguely resembling human legs, can walk down a shallow slope, powered only by gravity. This model is the simplest special case of the passive-dynamic models pioneered by McGeer (1990a). It has two rigid massless legs hinged at the hip, a point-mass at the hip, and infinitesimal point-masses at the feet. The feet have plastic (no-slip, no-bounce) collisions with the slope surface, except during forward swinging, when geometric interference (foot scuffing) is ignored. After nondimensionalizing the governing equations, the model has only one free parameter the ramp slope gamma. This model shows stable walking modes similar to more elaborate models, bur allows some use of analytic methods to study ifs dynamics. The analytic calculations find initial conditions and stability estimates for period-one gait limit cycles. The model exhibits two period-one gait cycles, one of which is stable when 0 < gamma < 0.015 rad. With increasing gamma, stable cycles of higher periods appear, and the walking-like motions apparently become chaotic through a sequence of period doublings. Scaling laws for the model predict that walking speed is proportional to stance angle, stance angle is proportional to gamma(1/3), and that the gravitational power used is proportional to upsilon(4) where upsilon is the velocity along the slope.

The aim of our study is to bettern understand the decision mechanisms of human body during the stepping motion over an obstacle.... this approach constitute decision module based on learning process to realize stepping motion. We compare simulation results to experimental data.

Some fundamental properties must be verified when dealing with the dynamic control of legged robots. Actually, to make the robot follow a given gait, the control strategy has to ensure dynamic stability in real time. Moreover, to make the supervisor efficient, the control of each leg has to be highly reliable. Accuracy robustness, and rapidly are then required to follow dynamic motions. To verify these properties, we propose a general control architecture designed for pneumatic legged robots and develop it for a biped. This control architecture is composed of two main levels. The upper one, called the Coordinator level, maintains the robot stability by correcting on line its center of mass acceleration and distributing correctly the forces on each limb. The lower level, called the Limb level, is devoted to the control of each limb according to the desired position and force trajectories given by the Coordinator level. The chosen trajectories are derived from biomechanical results, and a dynamic model that takes into account mechanical and pneumatic effects is presented. We propose to deal with the setting and the raising phases, using a continuous nonlinear joint impedance controller. The asymptotic stability is ensured by using Popov criteria. Simulation results of this new controller are presented, leading to a good behavior of the leg during the two phases at relatively high velocities.

Some basic problems related to the development of goal-directed locomotion in humans are reviewed here. A preliminary study is presented which was aimed at investigating the emergence of anticipatory head orienting strategies during goal-directed locomotion in children. Eight children ranging from 3.5 to 8 years had to walk along a 90 degrees right corner trajectory to reach a goal, both in light and in darkness. The instantaneous orientation in space of the head, trunk, hips and left foot antero/posterior axes was computed by means of an ELITE four-TV camera, 100 Hz system. The results showed that predictive head orienting movements can occur also in the youngest children. The head starts to rotate toward the goal before the corner point of the trajectory is reached. In children, the head peak rotation coincides with the trajectory corner while in adults the peak is attained before. In children, the walking speed is largely decreased in darkness. The results suggest that feedforward control of goal-directed locomotion appears very early in gait development and becomes increasingly important afterwards.

The acquisition process of bipedal walking in humans was simulated using a neuro-musculo-skeletal model and genetic algorithms, based on the assumption that the shape of the body has been adapted for locomotion. The model was constructed as 10 two-dimensional rigid links with 26 muscles and 18 neural oscillators....

Our goal was to document the kinetic strategies for obstacle avoidance in below-knee amputees. Kinematic data were collected as unilateral below-knee traumatic amputees stepped over obstacles of various heights in the walking path. Inverse dynamics were employed to calculate power profiles and work during the limb-elevation and limb-lowering phases. Limb elevation was achieved by employing a different strategy of intra-limb interaction for elevation of the prosthetic limb than for the sound limb, which was similar to that seen in healthy adult non-amputees. As obstacle height increased, prosthetic side knee flexion was increased by modulating the work done at the hip, and not the knee, as seen on the sound side. Although the strength of the muscles about the residual knee was preserved, the range of motion of that knee had previously been found to be somewhat limited. Perhaps more importantly, potential instability of the interface between the stump and the prosthetic socket, and associated discomfort at the stump could explain the altered limb-elevation strategy. Interestingly, the limb-lowering strategy seen in the sound limb and in non-amputees already features modulation of rotational and translational work at the hip: so an alternate strategy was not required. Thus, following a major insult to the sensory and neuromuscular system, the CNS is able to update the internal model of the locomotor apparatus as the individual uses the new limb in a variety of movements, and modify control strategies as appropriate.

The robustness of bipedal walking can be enhanced by the use of adaptive control and learning. This paper describes one such approach, Radial Basis Function (RBF) Neural Network Adaptive Control (NNAC). The adaptive control mechanism is designed in a virtual space utilizing the Virtual Model Control paradigm [3]. The neural network is parameterized and trained in an unsupervised learning mode. There are two advantages to this approach. First, the NNAC can identify the unmodelled dynamics of the robot and ensure asymptotic system stability in a Lyapunov sense. Second, the controller can better accommodate unexpected external disturbances. This system's design is described in this paper and simulation results are presented.

Locomotion involves repetitive movementsand is often executed unconsciously and automatically.In order to achieve smooth locomotion, the coordina-tion of the rhythms of all physical parts is important.Neurophysiological studies have revealed that basic rhythms are produced in the spinal network called, the central pattern generator (CPG), where some neural oscillators interact to self-organize coordinated rhythms.We present a model of the adaptation of locomotion patterns to a variable environment, and attempt to elucidate how the dynamics of locomotion pattern generation are adjusted by the environmental changes. Recent experimental results indicate that decerebrate cats have the ability to learn new gait patterns in a changed environment. In those experiments, a decere-brate cat was set on a treadmill consisting of three moving belts. This treadmill provides a periodic pertur-bation to each limb through variation of the speed ofeach belt. When the belt for the left forelimb is quickened, the decerebrate cat initially loses interlimb coordination and stability, but gradually recovers themand -nally walks with a new gait. Based on the abovebiological facts, we propose a CPG model whoserhythmic pattern adapts to periodic perturbation fromthe variable environment. First, we design the oscillatorinteractions to generate a desired rhythmic pattern.In our model, oscillator interactions are regardedas the forces that generate the desired motion pattern.If the desired pattern has already been realized, then theinteractions are equal to zero. However, this rhythmicpattern is not reproducible when there is an environ-mental change. Also, if we do not adjust the rhythmicdynamics, the oscillator interactions will not be zero.Therefore, in our adaptation rule, we adjust the mem-orized rhythmic pattern so as to minimize the oscillatorinteractions. This rule can describe the adaptive behav-ior of decerebrate cats well. Finally, we propose amathematical framework of an adaptation in rhythmicmotion. Our framework consists of three types of dynamics: environmental, rhythmic motion, and adap-tation dynamics. We conclude that the time scale ofadaptation dynamics should be much larger than that ofrhythmic motion dynamics, and the repetition of rhyth-mic motions in a stable environment is important for theconvergence of adaptation.

Children voluntarily adopt a frequency and movement pattern for walking. The force-driven harmonic oscillator (FDHO) model was used in this study for accurate prediction of the preferred walking frequency of nondisabled children and children with spastic hemiplegic cerebral palsy. Four potential optimality criteria with which the preferred walking pattern was forced to comply were examined: minimization of physiological costs, maximization of mechanical energy conservation, minimization of asymmetry in lower limb movements and minimization of variability of interlimb and intralimb coordination. Age and gender-matched nondisabled children (n = 6) and children with spastic hemiplegic cerebral palsy (n = 6) were tested under six frequency conditions of walking at a constant speed on a treadmill. For the nondisabled children, the results indicated that their preferred walking frequency could be accurately predicted by the FDHO model. They freely adopted a walking pattern that minimized physiological costs, asymmetry, and variability of inter- and intralimb coordination. For the children with spastic hemiplegic cerebral palsy, the prediction of preferred overground walking frequency required that the FDHO model be modified to account for muscle mass and leg length discrepancies between limbs and increased stiffness. Most of the children achieved the same optimality goals as the nondisabled when walking at the preferred frequency. However, the children were found to use different mechanisms to attain these goals: for example, a steeper increase observed in physiological cost at higher frequencies; a lowered center of gravity Of the body, which allowed for angular symmetry; and greater variability of between-joint coordination in the nonaffected limb and less variability in the affected limb.

This is the first of two articles in which we describe how infants adapt their spontanteous leg movements to changes in posture or to elicitation of behaviors by a mechanical treadmill.... This increased correlation between muscle torques at the hip and knee implicates anatomical and energetic constraints - the intrinsic limb dynamics - in creating coordinated limb behavior out of nonspecific muscle activations.

In this article, the development of the increasingly differentiated control of the joints necessary to transform the spontaneous leg movements of early infancy into adaptive and functional actions is described. The hypothesis-that increasing joint control requires the capability for disassociation of joint action, the active modulation of joint stiffness, and a transition from proximal to distal control of the joints-is proposed. Kinematic and kinetic analyses of the vertical kicks of infants 2 weeks, 3 months, and 7 months of age (as well as a comparative group of adults) indicated increasing joint independence as well as phase-dependent dent and joint-dependent control modifications. The kicks of the younger infants were dominated by a proximal control strategy and minimal adjustments of the limb energetics during the flexion and extension phases of the kick. By 7 months of age, much larger modulations of the kick phases were observed as well as increasing evidence of distal control. These results revealed kinematic and kinetic patterns of emerging limb control between 2 weeks and 7 months of age.

Automatic prediction of gait events (e.g., heel contact, flat foot, initiation of the swing, etc.) and corresponding profiles of the activations of muscles is important for realtime control of locomotion. This paper presents three supervised machine learning (ML) techniques for prediction of the activation patterns of muscles and sensory data, based on the history of sensory data, for walking assisted by a functional electrical stimulation (FES), Those ML's are: 1) a multilayer perceptron with Levenberg-Marquardt modification of backpropagation learning algorithm; 2) an adaptive-network-based fuzzy inference system (ANFIS); and 3) a combination of an entropy minimization type of inductive learning (IL) technique and a radial basis function (RBF) type of artificial neural network with orthogonal least squares learning algorithm. Here we show the prediction of the activation of the knee flexor muscles and the knee joint angle for seven consecutive strides based on the history of the knee joint angle and the ground reaction forces, The data used for training and testing of ML's was obtained from a simulation of walking assisted with an FES system [39], The ability of generating rules for an FES controller was selected as the most important criterion when comparing the ML's. Other criteria such as generalization of results, computational complexity, and learning rate were also considered, The minimal number of rules and the most explicit and comprehensible rules were obtained by ANFIS. The best generalization was obtained by the IL and RBF network.

A discussion is going on in cognitive science about the use of representations to explain how intelligent behavior is generated. In the traditional view, an organism is thought to incorporate representations. These provide an internal model that is used by the organism to instruct the motor apparatus so that the adaptive and anticipatory characteristics of behavior come about. So-called interactionists claim that this representational specification of behavior raises more problems than it solves. In their view, the notion of internal representational models is to be dispensed with. Instead, behavior is to be explained as the intricate interaction between an embodied organism and the specific make up of an environment. The problem with a non-representational interactive account is that it has severe difficulties with anticipatory, future oriented behavior. The present paper extends the interactionist conceptual framework by drawing on ideas derived from the study of morphogenesis. This extended interactionist framework is based on an analysis of anticipatory behavior as a process which involves multiple spatio-temporal scales of neural, bodily and environmental dynamics. This extended conceptual framework provides the outlines for an explanation of anticipatory behavior without involving a representational specification of future goal states.

A three-dimensional musculoskeletal database of the lower extremities has been developed for use in human musculoskeletal models. The locations of idealized muscle attachments on the pelvis, both femurs, both tibias and fibulas, and both feet were accurately digitized for 52 dried skeletal specimens. The mean specimen heights were 177.5 cm (male) and 166.2 cm (female) and the mean specimen age at the time of death was 48.8 yr. Statistical accumulation and scaling techniques were used to generate highly representative normative models, which were divided into groups and tested for differences based on gender and race. From the test results, the pelvis was divided into a male model(RMS = 8.6 mm), a black female model(RMS = 7.0 mm) and a white female model (RMS = 7.3 mm). The foot was separated into black (RMS = 3.7 mm) and white models (RMS = 3.6 mm). Single models were used for the femur (RMS = 6.5 mm) and the tibia/fibula (RMS = 3.7). Containing over 12000 anatomical landmarks digitized from 52 dried skeletons, this study represents an improvement over previous databases by an order of magnitude.

Collective dynamics of three globally coupled Hodgkin-Huxley neurons with a symmetric synap-tic coupling has been studied as a paradigm of one of the simplest but nontrivial example ofphysiology-based coupled oscillator networks capable of displaying synchrony and clustering.Rich phase dynamics have been observed and the phase diagram for various phase states, inparticular the synchronized state and clustered states, have been constructed by direct numer-ical simulations of the full system. We find that the computed phase diagram in the synapticparameter space of the reversal potential and the signal propagation time delay reveals a richbifurcation structure and agrees with one from bifurcation analysis of the reduced phase model.Implication of our findings in connection with two-coupled neurons and larger neural networksis discussed.

We examined the lower-limb electromyographic (EMG) activity from a patient with clinically complete spinal cord injury during orthotic gait. A newly developed gait orthosis was used to obtain bipedal locomotion. The surface EMG data during the gait together with the biomechanical variables were collected by way of a radio EMG system. A cyclic EMG activation pattern corresponding to the gait cycles were observed in each of the paralyzed lower-limb muscles during the orthotic gait. Although the EMG activation did not seem to contribute toward generating the gait, it showed some similarities to that of the infant stepping or immature gait. These results might be regarded as one of the indirect pieces of evidence that suggest the existence of a spinally originating motor mechanism underlying human locomotion.

This study analyses the incidence of anatomical (mass, height, inertia) and mechanical (gravity) parameters on the duration of gait initiation, from a standing posture, in children. Twenty-one children, aged 4, 6 and 8 years, participated in the study. Experimental and theoretical values of the duration of gait initiation are compared. The experimental data are computed from children's gait executed on a force plate. The theoretical data are computed by using an inverted-pendulum model. The results show that (1) duration of gait initiation is independent of gait velocity, as it is in adults; (2) the experimental values are very close to the theoretical values. These findings suggest that children's biomechanical constants are determining factors for initiating movement. It is hypothesized that the capacity to combine and adapt properties of the body with dynamics of the context is acquired through practice of independent walking.

This study analyses the anticipatory postural adjustments during the gait initiation process in children aged 2.5, 4, 6 and 8 years. In adults, anticipation during gait initiation includes a shift in the centre of foot pressure (CP) both backwards and towards the stepping foot. Backward displacement and the duration of the anticipation phase covary with the gait progression velocity reached by the subject at the end of the first step. In the present study, the children walked on a force plate that allowed us to calculate the acceleration of the centre of mass and the displacements of the CP. The results showed three main characteristics of the development of anticipatory behaviour: (1) The occurrence of anticipatory displacements of the CP increased progressively with age. Systematic backward anticipation was found for all children except one of the youngest, whereas the lateral displacement was systematically observed later, in the 6-year group; (2) the amplitude of the spatial parameters showed a significant increase with age; (3) contrary to the adult, the amplitude of the backward shift did not covary with the forthcoming velocity in the youngest groups. This covariation became significant at 6 years and remained significant at 8 years. The results showed that even if anticipatory behaviour was present in 2.5-year-old children it is only later that the child is able of more accurate tuning of feedforward control, probably due to better control of the overall postural adjustments.

Lower-limb movements and muscle-activity patterns were assessed from seven normal and seven ambulatory subjects with incomplete spinal-cord injury (SCI) during level and uphill treadmill walking (5, 10 and 15 degrees). Increasing the treadmill grade from 0 degrees to 15 degrees induced an increasingly flexed posture of the hip, knee and ankle during initial contact in all normal subjects, resulting in a larger excursion throughout stance. This adaptation process actually began in mid-swing with a graded increase in hip flexion and ankle dorsiflexion as well as a gradual decrease in knee extension. In SCI subjects, a similar trend was found at the hip joint for both swing and stance phases, whereas the knee angle showed very limited changes and the ankle angle showed large variations with grade throughout the walking cycle. A distinct coordination pattern between the hip and knee was observed in normal subjects, but not in SCI subjects during level walking. The same coordination pattern was preserved in all normal subjects and in five of seven SCI subjects during uphill walking. The duration of electromyographic (EMG) activity of thigh muscles was progressively increased during uphill walking, whereas no significant changes occurred in leg muscles. In SCI subjects, EMG durations of both thigh and leg muscles, which were already active throughout stance during level walking, were not significantly affected by uphill walking. The peak amplitude of EMG activity of the vastus lateralis, medial hamstrings, soleus, medial gastrocnemius and tibialis anterior was progressively increased during uphill walking in normal subjects. In SCI subjects, the peak amplitude of EMG activity of the medial hamstrings was adapted in a similar fashion, whereas the vastus lateralis, soleus and medial gastrocnemius showed very limited adaptation during uphill walking. We conclude that SCI subjects can adapt to uphill treadmill walking within certain limits, but they use different strategies to adapt to the changing locomotor demands.

This paper examined the feasibility of using different optimization criteria in inverse dynamic optimization to predict antagonistic muscle forces and joint reaction forces during isokinetic flexion/extension and isometric extension exercises of the knee. Both quadriceps and hamstrings muscle groups were included in this study. The knee joint motion included flexion/extension, varus/valgus, and internal/external rotations. Four linear, nonlinear, and physiological optimization criteria were utilized in the optimization procedure. All optimization criteria adopted in this paper were shown to be able to predict antagonistic muscle contraction during flexion and extension of the knee. The predicted muscle forces were compared in temporal patterns with EMG activities (averaged data measured from five subjects). Joint reaction forces were predicted to be similar using all optimization criteria. In comparison with previous studies, these results suggested that the kinematic information involved in the inverse dynamic optimization plays an important role in prediction of the recruitment of antagonistic muscles rather than the selection of a particular optimization criterion. Therefore, it might be concluded that a properly formulated inverse dynamic optimization procedure should describe the knee joint rotation in three orthogonal planes.

Successful performance of a sensorimotor task arises from the interaction of descending commands from the brain with the intrinsic properties of the lower levels of the sensorimotor system, including the dynamic mechanical properties of muscle, the natural coordinates of somatosensory receptors, the interneuronal circuitry of the spinal cord, and computational noise in these elements. Engineering models of biological motor control often oversimplify or even ignore these lower levels because they appear to complicate an already difficult problem. We modeled three highly simplified control systems that reflect the essential attributes of the lower levels in three tasks: acquiring a target in the face of random torque-pulse perturbations, optimizing fusimotor gain for the same perturbations, and minimizing postural error versus energy consumption during low- versus high-frequency perturbations. The emergent properties of the lower levels maintained stability in the face of feedback delays, resolved redundancy in over-complete systems, and helped to estimate loads and respond to perturbations. We suggest a general hierarchical approach to modeling sensorimotor systems, which better reflects the real control problem faced by the brain, as a first step toward identifying the actual neurocomputational steps and their anatomical partitioning in the brain.

The control of locomotion has been studied from various perspectives related to the tasks of pattern generation, equilibrium control or adaptation to the environment. The last of these locomotor components has received comparably less attention, specifically pertaining to anticipatory adjustments. Continuing the work which has been conducted on both humans and cats, the present paper explores the nature of the differences in anticipatory locomotor adjustments for obstacle avoidance versus the accommodation to level changes. Six subjects walked in six different environments including no obstructions, a simple obstacle, two different level changes (a platform and stairs), and a combination of an obstacle with each respective level change. Full dynamic analyses allowed comparison of muscle torques as well as muscle power generated and absorbed at the lower limb joints across conditions. It was found that the previously shown robust lower limb reorganization characterized by a knee flexor generation strategy was upheld in all conditions when the obstacle was present. Pure level changes involved an augmentation of the ongoing hip strategy inherent In normal level walking. In the compound environment of obstructed level changes, subjects chose to combine an augmentation of hip flexor power with a reorganization to active knee flexion. The results are discussed from the point of view of general principles of mechanical coordination and the exploitation of intersegmental dynamics for foot transport.

CPG (central pattern generator) and entrainment dynamics together form a promising framework for robust and adaptive behavior generation for a high degree of freedom system in unstructured environment. This paper investigates its possibility in the domain of biped robot locomotion....

In this paper, we propose a learning method for implementing human-like sequential movements in robots. As an example of dynamic sequential movement, we consider the stand-up task for a two-joint, three-link robot. In contrast to the case of steady walking or standing, the desired trajecory for such a transient behavior is very difficult to derive....

A model calculation is presented simulating the coordinated interaction between the walking legs of a multi-legged animal. The neural network consists of separate modules with oscillatory capabilities. It has the ability to adjust the necessary parameters for producing a coordinated interaction between the modules in a self-organizing fashion. Some sort of reinforcement comparison learning is used to train the network. It starts oscillations in a completely uncoupled state. After about 100 learning steps, the generation of a stable alternating pattern is usually terminated. Then, the network is able to maintain synchronization, even when disturbances are applied to single agents or to the network as a whole.

The purpose of this study was to investigate the influence of midsole hardness and running velocity on external impact forces in heel-toe running.... The result showed that running velocity does influence external impact force peaks (linear connection) and that midsole hardness does not influence magnitude and loading rate of the external vertical impact forces. Changes in kinematic and kinetic data can be used to explain this result mechanically. However, the neuromuscular control mechanism to keep external impact forces constant are not known.

It has been suggested that control using a skill-based expert system can be applicable to gait restoration. Rule-based systems have several advantages for this application: they generate a fast response (they are not computationally intensive) and they are easy to comprehend and implement. A major problem with using such systems is the inability of users to determine its rules. In this study, an automatic method for obtaining the production rules from a set of examples is described, The rule base was automatically induced from a model which used external sensor signals as inputs and electromyogram (EMG) patterns as outputs. The method is based on the minimization of entropy. A production rule estimated the muscle activity pattern using the sensor information. The algorithm was tested using data recorded from six able-bodied individuals during ground level walking, with and without ankle-foot orthoses. The data show ed that gait variability will increase in able-bodied subjects when the motion of ankle joints is restricted, thus, providing a good test for generalization. The experimental results illustrate performance of the production rule that estimates quadriceps muscle group activity pattern for ground level walking in able-bodied subjects.

A learning model for coupled oscillators is proposed. The proposed learning rule takes a simple form by which the intrinsic frequencies ofthe component oscillators and the coupling strength between them are changed according to the effects of the input signals on the dynamicsof the oscillator. In the learning mode, each component oscillator receives a teacher signal of desired phase and frequency, and a desiredparameter set for generating the desired pattern is acquired by the proposed learning rule. It is known that the basic locomotor patterns ofmany living bodies are generated by coupled neural oscillators. The proposed learning rule could be a learning model used by such neuralsystems to acquire an adequate parameter set for generating a desired locomotor pattern.

Slipping during various kinds of movement often leads to potentially dangerous incidents of falling. The purpose of this study was to determine whether there was evidence to support the theory that movement strategies could be used by individuals to regain stability during an episode of slipping and whether forced sliding from a moving platform accurately simulated the effect of slipping on stability and balance. A single-link-plus-foot biomechanical model was used to mathematically simulate base of support (BOS) translation and body segment rotation during movement termination in sagittal plane. An optimization routine was used to determine region of stability [defined at given COM locations as the feasible range of horizontal velocities of the center of mass (COM) of human subject that can be reduced to zero with respect to the BOS while still allowing the COM to traverse within the BOS limits]. We found some 30 of stability for slipping and non-slipping conditions. This finding supports the theory that movement strategies can be sought for restoring stability and balance even if slipping unexpectedly occurs. We also found that forced sliding produces effects on stability that are similar to those of slipping, indicated by over 50 two conditions. In addition, forced sliding has distinctive effects on stability, including a shift of the region of stability extended beyond the BOS in the direction of sliding. These findings may provide quantifiable guidance for balance training aimed at reducing fall incidents under uncertain floor surface conditions.

This paper presents a general method for simulating the movement of the lower extremity during human walking. It is based upon two separate algorithms: one for single support (am open kinematic chain), and the other for the double support phase (a closed-loop linkage).... The attractiveness of the method is that it offers a compact alternative to manually deriving the equations defining a mathematical model for human gait.

Trunk motions are typically used to stabilize the motions of biped robots, which can be very large in some leg trajectories. This paper proposes a method to reduce the motion range of the trunk by generating a desired trajectory of the ZMP [zero-moment point]. The trajectory is determined by a fuzzy logic based on the leg trajectories that are arbitrarily selected. The resulting ZMP trajectory is similar to human's one and the ZMP continuously moves forward. The proposed scheme is simulated on a 7-degree-of-freedom biped robot. Its results indicate that the proposed ZMP trajectory increases the stability of the locomotion and thus resulting in reduction of motion range of the trunk.

The MIT Leg Lab is best known for the seminal work of Marc Raibert, who showed in the 1980s that robotic running could be accomplished using a few simple, decoupled control laws.... The question was: what to do next?

A neural network model has been developed to represent the shaping function of a central pattern generator (CPG) for human locomotion. The model was based on cadence and electromyographic data obtained from a single human subject who walked on a treadmill. The only input to the model was the fundamental timing of the gait cycle (stride rate) in the form of sine and cosine waveforms whose period was equal to the stride duration. These simple signals were then shaped into the respective muscle activation patterns of eight muscles of the lower limb and trunk. A network with a relatively small number of hidden units trained with back-propagation was able to produce an excellent representation of both the amplitude and timing characteristics of the EMGs over a range of walking speeds. The results are further discussed with respect to the dependence of some muscles upon sensory feedback anti other inputs not explicitly presented to the model.

Split-belt locomotion (i.e., walking with unequal leg speeds) requires a rapid adaptation of biomechanical parameters and therefore of leg muscle electromyographic (EMG) activity. This adaptational process during the first strides of asymmetric gait as well as learning effects induced by repetition were studied in 11 healthy volunteers. Subjects were switched from slow (0.5 m/s) symmetric gait to split-belt locomotion with speeds of 0.5 m/s and 1.5 m/s, respectively. All subjects were observed to adapt in a similar way: (1) during the first trial, adaptation required about 12-15 strides. This was achieved by an increase in stride cycle duration, i.e., an increase in swing duration on the fast side and an increase in support duration on the slow side. (2) Adaptation of leg extensor and flexor EMG activity paralleled the changes of biomechanical parameters. During the first strides, muscle activity was enhanced with no increase in coactivity of antagonistic leg muscles. (3) A motor learning effect was seen when the same paradigm was repeated a few minutes later - interrupted by symmetric locomotion - as adaptation to the split-belt speeds was achieved within 1-3 strides. (4) This short-time learning effect did not occur in the ''mirror'' condition when the slow and fast sides were inverted. In this case adaptation again required 12-15 strides. A close link between central and proprioceptive mechanisms of interlimb coordination is suggested to underlie the adaptational processes during split-belt conditions. It can be assumed that, as in quadrupedal locomotion of the cat, human bipedal locomotion involves separate locomotor generators to provide the flexibility demanded. The present results suggest that side-specific proprioceptive information regarding the dynamics of the movement is necessary to adjust the centrally generated locomotor activity for both legs to the actual needs for controlled locomotion. Although the required pattern is quickly learned, this learning effect cannot be transferred to the contralateral side.

A neural network mechanism is proposed to modify the gait of a biped robot that walks on sloping surfaces using sensory inputs. The robot climbs a sloping surface from a level surface with no priori knowledge of the inclination of the surface. By training the neural network while the robot is walking, the robot adjusts its gait and finally forms a gait that is as stable as when it walks on the level surface. The neural network is trained by a reinforcement learning mechanism while proportional and integral (PI) control is used for position control of the robot joints. Experiments of static and pseudo dynamic learning are performed to show the validity of the proposed reinforcement learning mechanism.

The authors of this study are a part of a joint project, involving four French laboratories, whose goal is the design and construction of a mechanical biped robot with anthropomorphic characteristics. Motivations for this project, named BIP, are addressed in [7], In the first section of this paper, we will examine mechanical architectures of some representatives of state-of-the art biped robots by focusing on their kinematic arrangement. It is widely known that the existence of natural gaits is closely linked to the intrinsic dynamic characteristics of the mechanical structure of the biped robot. In order to further develop this idea, two studies will be presented in the second section: the first is relative to the lateral instability of the system while the second deals with the existence of passive pendular gaits during the swing phase of walking in the sagittal plane. In the last section, in correlation with the observations made in the Sections I and II, we will gain insight into main characteristics of the mechanical architecture that we have designed for the BIP project: 15 active degrees of freedom (DOF), joints actuated by special transmission system, anthropometric mass distribution.

The skill of rhythmic juggling a ball on a racket is investigated from the viewpointof nonlinear dynamics. The difference equations that model the dynamical system areanalyzed by means of local and non-local stability analyses. These analyses yield that thetask dynamics offer an economical juggling pattern which is stable even for open-loopactuator motion. For this pattern, two types of predictions are extracted: (i) Stable peri-odic bouncing is sufficiently characterized by a negative acceleration of the racket at themoment of impact with the ball; (ii) A nonlinear scaling relation maps different jugglingtrajectories onto one topologically equivalent dynamical system. The relevance of theseresults for the human control of action was evaluated in an experiment where subjectsperformed a comparable task of juggling a ball on a paddle. Task manipulations involveddifferent juggling heights and gravity conditions of the ball. The predictions were con-firmed: (i) For stable rhythmic performance the paddle+s acceleration at impact is nega-tive and fluctuations of the impact acceleration follow predictions from global stabilityanalysis; (ii) For each subject, the realizations of juggling for the different experimentalconditions are related by the scaling relation. These results allow the conclusion that forthe given task, humans reliably exploit the stable solutions inherent to the dynamics ofthe task and do not overrule these dynamics by other control mechanisms. The dynamicalscaling serves as an efficient principle to generate different movement realizations fromonly a few parameter changes and is discussed as a dynamical formalization of the prin-ciple of motor equivalence.

The impact of manipulating the physical layout and the perceptual appearance of the environment on motor behavior and perceptual judgments for action was investigated by presenting 12- to 30 month old toddlers with a visually guided locomotion task involving stepping over a barrier varying in its height. Experiment 1 observed changes in infants' crossing behavior as a function of barrier height, with successful crossing at low heights, failures in crossing as the height increased, and refusals to attempt crossing at the highest barrier heights for younger infants. Experiment 2 manipulated the barrier's perceived transparency and spatial extent, with a finding of increased crossing thresholds at all ages for barriers of greater spatial extent relative to lesser spatial extent, and increased thresholds for opaque relative to transparent barriers at younger ages. Both studies found that crossing thresholds were strongly related to walking experience, suggesting that for infants and toddlers action in the world is experientially scaled.

Significant advances have occurred over the past two decades in issues related to the mechanical design of legged robots and the coordination and control of legs during locomotion. The performance of current legged robots, however, remains far below even simplest counterparts in the biological world. Naturally, this has led to a search by researchers for biologically motivated approaches to the design and control of legged robots in order to improve their performance and robustness, In this paper the use of central pattern generators (CPGs) will be examined for the control of locomotion. In doing so, a biologically consistent mathematical model that has the ability to emulate arbitrary gaits is developed. A simple algorithm qfor encoding the characteristic gaits of bipeds, quadrupeds and hexapeds is presented. The paper concludes with a brief description of a locomotion control architecture for actually realizing leg movements that correspond to various gaits.

Analytical techniques are presented for the motion planning and control of a 12 degree-of-freedom biped walking machine. From the Newton-Euler equations, joint torques are obtained in terms of joint trajectories, and the inverse dynamics are developed for both the single-support and double-support cases. Physical admissibility of the biped trajectory is characterized in terms of the equivalent force-moment and zero-moment point. This methodology has been used to obtain reference inputs and implement the feedforward control of walking robots. A simulation example illustrates the application of the techniques to plan the forward-walking trajectory of the biped robot. The implementation of a prototype mechanism and controller is also described.

This paper presents the energy analysis of a bipedal walking system. The main goal is to understand the movement strategies in walking and to search for the optimal locomotion variables that minimise a cost function related to energy. Three diffferent indices are proposed: mean absolute power, mean power dispersion and mean power lost. In order to accomplish this goal, in the description and analysis of the motion it is used a set of locomotion variables, namely: step length, hip height, hip ripple, hip offfset, foot clearance and link lengths. Based on these variables and their influence on the energy flow, the performance measures are discussed and the results compared with those observed in nature.

Although there is increasing agreement that the cerebellum plays an important role in motor learning, the basic substance of what constitutes motor learning has been difficult to define. Unless motor learning is somehow radically different from other forms of learning, it must involve relatively simple stimulus-stimulus and stimulus-response associations. All forms of learning, including purely sensory associations and cognitive learning as well as motor learning, effect changes in behavior. However, a singular characteristic of motor learning is that it adjusts joint and limb mechanics by altering the neural input to muscles through practice and mental rehearsal. The hypothesis proposed here is that the cerebellum plays an important role in motor learning by forming and storing associated muscle activation patterns for the time-varying control of limb mechanics. By modulating the cocontraction of agonist-antagonist muscles through adjustments in the timing and amplitude of muscle activity, the viscoelastic properties of joints can be appropriately regulated throughout movement and adapted for transitions between postures and movements. Optimal control of joint viscoelastic properties cannot be achieved by online corrections initiated by reflex feedback because of the delays and consequent instabilities incurred. Instead, strategies for optimizing muscle activation patterns or synergies must be learned from the temporal association of proprioceptive stimuli signaling muscle lengths and forces and the rates of changes in these parameters, with reinforcement occurring when the movement achieves its objective. Such strategies would involve varying degrees of cocontraction or reciprocal inhibition of agonist-antagonist muscles that ultimately contribute to joint and limb stiffness. Evidence from neural recordings and clinical and experimental lesion studies are presented, suggesting that the cerebellum uses teleceptive and proprioceptive feedback as feedforward conditioned stimuli for specific muscle activation patterns contributing to joint stiffness (i.e., agonist-antagonist muscle synergies) for particular tasks and postures. A wide variety of observations are thought to be consistent with such a role for the cerebellum, but ultimately additional experiments could confirm or disconfirm this hypothesis.

The mobility of above-knee amputees (A/K) is limited, in part, due to the performance of A/K prostheses during the stance phase. Currently stance phase control of most conventional A/K prostheses can only be achieved through leg alignment and choise of the SACH (Solid Ankle Cushioned Heel) foot. This paper examines the role of the knee controller in relation to a SACH foot during the stance phase of level walking.... Analysis and interpretation of the data indicate the following: (1) SACH foot design can strongly influence the walking mechanics independent of the knee controller; (2) knee controller design and SACH foot design are mutually interdependent; and (3) normal kinematics imposed on the prosthetic knee does not necessarily produce normal hip kinematics (e.g., reduce the abnormal rise in the prosthetic side hip trajectory)....

A principle of locomotor control in an unpredictably changing environment is presented on the basis of neurophysiology and biomechanics from the perspective of nonlinear dynamics theory. Locomotor movements emerge as a limit cycle generated through global entrainment among the neuro-musculo-skeletal system and the environment. A computer simulation of a specific model of bipedal locomo