JP3355113B2 - Method for simulating human body movement and method for generating animation using the method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of simulating the movement of a human body and a method of generating an animation of the movement of a human body. In particular, the present invention relates to a simulation method for estimating a motion of a human body based on dynamics, and an animation generation method related to the simulation method.

[0002]

2. Description of the Related Art In recent years, robotics, computer graphics (CG), and virtual reality (V)
Interest in the movement of the human body is increasing among researchers in each field of R). Researchers in the field of robotics are focusing on the movements of human arms and legs that provide smooth and rapid movement. Their stable and efficient human gait has also attracted their attention. To date, great efforts have been made to produce walking devices and the like that move like humans. In fields such as CG and VR, researchers continue to search for methods for creating motion data of the human body. Human body motion data is used to control virtual people in video games, movies, and cyberspace. The demand for human body motion data is increasing.

[0003] Motion capture and physical animation have been used to generate such data. However, these methods cannot automatically generate spontaneous human body movements. The human body is a redundant multi-joined object controlled by many muscles. This requires proper and cooperative movement of these muscles to control the entire body.

The art of creating human animations can be divided into three groups: motion capture, kinematics-based methods, and dynamics-based methods.

[0005] Since motion capture requires special expensive equipment, kinematic trajectories already prepared as a database are usually used. If the desired data is not found in the database, the animators modify the kinematics data to suit their needs. However, in this case you will change the kinematic parameters without taking the dynamics into account, so in most cases
The resulting movement does not conform to physical laws and becomes very unnatural. On the other hand, the motion obtained by the dynamic simulation is more natural.

[0006] Dynamics-based human body animations have been created by many researchers so far. However, when controlling a human body model in a normal environment governed by the laws of physics, animators must specify the changes in torque and force applied to the model instead of kinematic trajectories. The effect of changing each and every kinetic parameter is not clear, and this is a very difficult task.

The difficulty in controlling a human body model results from the redundancy of human body dynamics. Redundancy comes from a large number of joints. Each joint has one, two, or three degrees of freedom (DO
F). Therefore, the total DOF of the human body is close to 50. Although many degrees of freedom allow people to avoid obstacles and move dexterously, controlling the model requires more complex algorithms. For these reasons, there is a strong demand for a controller that calculates the torque at each joint and moves the human body model as desired.

[0008] There are currently two methods for identifying changes in joint torque. The first method uses a control method known in the control theory as proportional derivative (PD) control. However, in this method, a large number of posture key frames and a group of parameters for determining the joint torque are required for each key frame. Since the parameters are completely dependent on the movement, new keyframe poses and parameters are needed for different types of movement. Also, when the size or mass of the human body model changes, the animation creator must carefully set the parameters to obtain the same motion.

The second method uses optimal control theory. This method controls under the assumption that the movement of the human body minimizes the value of some objective function. The calculation of the joint torque at each frame can be performed by the conventional gradient method, dynamic programming, or Rayleigh-Ritz method.
hod, D.E. Schmitt and el "Opt
imal motion programming o
f robot manipulators ", Tra
nsaction on ASME, Journal o
f Mechanisms, Transmissions
and Automation in Desig
n, 107, pp. 239-244 (1985)). Pandy et al. Simulated a maximum height jump using optimal control (MG Pandy and el "An opti).
mal control model for max
imum-height human jumpin
g, Journal of Biomechanics,
23 (12) pp. 1185-1198 (199
0)). The objective function of Pandy et al. Was the height of the jump, but this function cannot be used for general human body movements. Yamaguchi et al. Used a dynamic programming method to simulate a walking motion (GT Yamaguchi and "R").
estoring unassisted natur
al gate to paraplegicsvia
functional neuromuscular
Stimulation: a computer s
"imulation study", IEEE Tra
nsactions on biomedical e
ngineering, 37, pp. 886-902
(1990)). Their objective function is the integral of the square of the difference between the calculated trajectory of the human gait and the actual trajectory captured, and the integral sum of the cube of muscular strength divided by mean muscle cross-sectional area (PCSA). Was the sum. However, their purpose is to calculate strength, and they cannot create animations in their own way.

The Rayleigh-Ritz method is based on B-spline,
Approximate kinematic parameters by Fourier function or the sum of several basis functions such as wavelets,
This is an algorithm for calculating joint torque using inverse dynamics. Rose and colleagues used this method to create human body animations, but did not take into account body muscles (C. Rose and el "Efficie").
nt generation of motion t
ransions using spaceim
e constraints ", Computer G
rapics, 30, pp. 147-152 (199
6)). For this reason, the movement of a person having different muscle strength cannot be processed, and the simulation of the actual movement of the human body cannot be performed.

At present, no algorithm has been elucidated for humans to perform voluntary movements. However, it has been found that two types of controllers, an open loop feedforward control system and a closed loop feedback control system, are involved. Closed loop control systems constantly feed back control variables as inputs to compensate for fluctuations. By analyzing the response of the human body to disturbances, many researchers have investigated the characteristics of human feedback control.

On the other hand, open loop feed forward control plays an important role in voluntary movement. To create fast and flexible hand-like movements, the trajectory of the human body must be pre-programmed with an open-loop control system. Several optimal control models have been proposed for hand trajectory derivation. A famous model is the minimum torque change model proposed by Uno et al. (Y. Uno and
el "Formation and control
of optimal trajectory in
human multi-joint arm mo
element ", Biological Cybern
etics, 61 pp. 89-101 (198
9)). The objective function is represented by the following equation.

[0013]

(Equation 1)

Here, J is the value of the objective function, τ ₀ is the torque produced by the i-th joint, and t ₀ and t _f represent the start and end times of the movement, respectively.

[0015] On the other hand, almost no model of a motion such as walking formed mainly by the entire lower limb has been proposed. Control of the lower limbs is much more complicated because it involves controlling the posture. Chow et al. Created a gait movement of the human body and set the amount of work performed by the muscles as an objective function (CK Chow).
and el "Studies of human
locomotion via optimal p
programming, Mathematical
Biosciences, 10, pp. 239-306
(1971)). Meanwhile, Rose et al. Set an objective function represented by the following equation.

[0016]

(Equation 2)

Here, τ _i represents the torque generated at the i-th joint. However, Chow et al. And Roses did not consider postural stability. On the other hand, Ko et al., When the condition is unstable, translate the pelvis until the posture is stabilized, and make the origin joint (origin).
(H. joint) is rotated to fix the posture of the human body (H.
Ko and el "Animating huma
n locomotion with inverse
dynamics ", IEEE Computer
Graphics and Application
s, March, pp. 50-59 (1996)). Coal's work was based on inverse dynamics, which stabilized each frame while walking, but did not plan the trajectory.

[0018]

The above is the situation relating to the simulation of the motion of the human body or the generation of animation.

The present invention has been made in view of the above various problems, and an object of the present invention is to provide a technique capable of simulating or animating a human body without using an expensive device. Another object of the present invention is to express the voluntary movement of the human body more naturally. Still another object of the present invention is to provide a technique capable of expressing a natural movement of a human body in consideration of muscles with relatively simple and versatile parameter settings. Still another object of the present invention is to provide a technique for expressing movement of a lower limb in which posture control is difficult.

[0020]

The simulation method of the present invention is based on dynamics. The present invention estimates the motion of a human body based on an optimal control theory that employs a minimum muscle motion function as an objective function. Examples of the minimum muscle motion function include those determined in consideration of the action of muscles and tendons, and those determined in consideration of muscle activity levels.

One embodiment of the simulation method according to the present invention includes three steps. First, a step of performing feedforward control based on an optimal control theory using a minimum muscle motion function as an objective function to calculate a motion of a human body. Next, the method includes a step of verifying the posture of the human body in the calculated movement, and a step of correcting the movement of the human body if there is a problem in the posture. Posture may be considered as body stability. That is, if the body is unstable, correct the movement or trajectory.

The animation generation method according to the present invention comprises:
Based on the dynamics, a muscle-based human body model is established, the motion of the human body is estimated based on the optimal control theory, and the motion is generated as an animation. At that time, the minimum muscle motion function may be adopted as the objective function of the optimal control theory.

[0023]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS It is the physical structure of the human body and the neural controller that characterize the movement of the human body. Joint torque is created by the muscles that connect the bones. The configuration space and the maximum output of the human body are largely determined by the attachment position and physical structure of each muscle. Very little research has been done on human body animation using muscle models as actuators for human body models. In the present embodiment, a human body model to which a muscle is added is created.

The mechanism by which the human body is controlled by the central nervous system (CNS) has not yet been elucidated. The process of trajectory generation is more complex than for other parts of the body, especially since the legs need to be balanced. Most of the conventional leg controllers can be applied only to a specific operation. In the present embodiment, a synthesis controller for a muscle-added human body model that can rationally interpolate between keys only by specifying key postures and maintain a balance is realized.

The present embodiment accurately simulates human dynamics and a feedforward control system.
The power system of the entire human body including the muscle groups of the entire lower limb is targeted. This method can easily absorb physical individual differences such as muscle strength and body size. This model can be used for analysis and simulation of actual human body movement.

This embodiment is based on the optimal control theory. That is, it is assumed that the feedforward controller can minimize the value of a certain objective function while maintaining a stable body balance. An arbitrary posture is interpolated using this objective function. In the present embodiment, the minimum muscle motion function taking into account the muscle-tendon dynamics and the muscle activity level is used as the objective function. Once the trajectory is calculated in this way, the dynamic stability of the movement of the human body is then calculated. Stabilize if unstable. Muscles produce reverse kinetics and Crowninshield.
d) Calculation using the prediction algorithm proposed by R. D. (R.C.
"A phisiologically based
criterion of muscle forc
e prediction inlocomatio
n ", Journal of Biomechanic
s, 14, pp. 793-800 (1981)).

Finally, the effect of the present embodiment will be confirmed by experiments. As a result of comparing the motion data of the human body obtained by the motion capture with the trajectory obtained in the present embodiment, the present embodiment shows that the dynamic animation of the human body can be obtained by specifying only a few key postures. Turned out to be possible. This embodiment can also be applied to analysis of human body motion data obtained by a motion capture system. Also, since the history of muscle movement can be obtained at the same time, these can be used directly as input to the dynamics system. Further, the system according to the present embodiment can consider the muscle strength of each muscle. Hereinafter, the present embodiment will be described with reference to the drawings as appropriate.

FIG. 1 is a flowchart showing an outline of the processing procedure of the present embodiment. As shown in the figure, first, feedforward control is performed based on the optimal control theory that employs the minimum muscle motion function as the objective function, and the motion of the human body is calculated (S1). Next, the posture of the human body on the calculated trajectory is verified (S2). If there is a problem with the posture, the movement of the human body is corrected (S3). In the embodiment, verification and correction of the posture are performed at two levels. That is, verification and correction are performed first on the boundary posture (the first and last postures of the movement) and then on the middle posture. When desired motion data is completed in this way, it is animated, displayed on a display, and stored in a storage device (S4). Animation is performed by interpolating between key postures specified by the user.

[0029] 1. A human body model necessary for processing the muscle-based human body model S1 is constructed based on the following considerations.

1.1. Muscle parameters in biomechanics In the field of biomechanics, many researchers have used muscle-added models to calculate body movements.
In contrast, few computer graphics researchers have realized a human body system controlled by muscles. In the present embodiment, a research result of biomechanics is applied.

In order to create a muscle model, data of a parameter group for specifying a muscle attachment position and a muscle is essential. The former includes an origin and a muscle insertion position in each body coordinate system. For muscle origin and insertion position data, most of the models for research in the field of biomechanics use data provided by Brand et al. (RA Brand).
and el "The sensitivity o
f muscle force prediction
s to changes in physiolog
ic cross-sectional area ",
Journal of Biomechanics, 1
9, pp. 589-596 (1986)). Also, the data of Hoy et al. Is frequently used (MG Ho
y and el "A musculoskelet
al model of the human low
erextremity: The effect of
muscle, tendon and moment
arm on the moment-angle
relationship of musculote
ndon actors at the hi
p, knee and ankle ”, Journal
of Biomechanics, 23 (2), p
p. 157-169 (1992)). Hui et al.'S data include other muscle-specific parameters, such as muscle fiber length, relaxed tendon length, tendon elasticity, maximum muscular strength, and penation angle (muscle fiber direction and force). The angle formed by the direction).

As shown in FIG. 2, the human body model used in this embodiment has 15 parts, 30 DOF, and 47 ×
Consists of two muscles. Each part is considered as one rigid body. The size, weight, and moment of inertia of each part used the data of Yamaguchi et al. The foot consists of two rigid bodies, the heel and the toe. When the angle between the foot and the floor is positive, it is considered that only the toe of the foot is in contact with the floor.

1.2. Hill Muscle Model As a model of muscle and tendon, the famous Hill (Hi
11) three-element model was used. This model has three components: a contractile component (CE, muscle fiber), a series elastic component (SEE, muscle tendon), and a parallel elastic component (PEE,
Fibers and connecting structures around the fiber bundles). The force (f ^ce ) exerted by CE is a function of its element length l ^ce , contraction rate v ^ce , and the level of muscle activation (a) controlled by the central nervous system (CNS). C
The force produced by E is proportional to the muscle activity level, and can be written as ^fce = f ( ^lce , ^vce , a). SEE and PEE output forces (f ^T ,
f ^pe ) relates only to the length (l ^T , l ^pe ) of each element, and can be written as f ^T , = f (l ^T ) and f ^pe = f (l ^pe ), respectively. FIG. 4 shows f ^ce and its length l ^ce , and l ^ce
It shows the relationship between ^pe and its length ^lpe . The relationship between the forces generated by each element is f ^T , = f ^ce + f ^pe . The muscle - the length of the tendon ^{(l MT)} is the sum of the length of the tendon of the muscle ^{fibers, l MT = l ce + l} T, it is l ce ^{= l pe} established. The relationship between tendon force and length can be approximated by the following primary elasticity equation:

[0034] ^{f t = k T (l T} -l T o) k T In this equation the elasticity of the ^tendon, the ^l _{T o} indicating the length of time of relaxation. the set value of k ^T here is 37.5 ^{/ f} _{M o.}
On the other hand, f ^M _o denotes the peak isometric force muscles, average cross-sectional area of the (PCSA) can be calculated by the following equation using.

[0035] In ^f _M o = kA _i this equation, _{A i} is PCSA, k represents a constant, here is set to 30 Nm ^-2. The data of Brand et al. Was used for the shape of the muscle (RA Bran
d and el "A model of low
r extreme muscle anat
omy ", J. Biomech, 104, pp. 304.
-310 (1982)). For a group of parameters for specifying muscles such as muscle contraction element length, pennation angle, and average cross-sectional area, see Friedrich
h) using the data of (JA Friedric)
hand el "Muscle fiber arc"
hitcture in the human lo
wer limb ”, The Journalna B
iomechanics, 23, pp. 91-95 (1
989)). Length at the tendon relaxed (l T _^o) was calculated using the method proposed in Hoira. This method is based on a moment (hereinafter referred to as a passive moment, hereinafter referred to as a "passive moment") in which the length ^lce of the actuator is longer than the natural length and is generated in a direction opposite to the bending. in the joint angle corresponding to) that point to things selects its original length l ^M _o l ^T _o so close.
The passive elastic moment of the joint was calculated as shown in FIGS.

1.3. Independent parameters of the generalized coordinate system When the movement of the body is captured as a still image at a certain moment, the entire body can be expressed by specifying the joint angle and the position of the body. These are called generalized coordinates, where q = ｑq ₁ ,
q ₂ ,. . . , Q _n }. Although there are dependent components in the generalized coordinate system, only independent components are needed to determine the state of the body. Let the independent components be q ₁ = ｛q ₍₁₎ ,
q ₍₂₎ ,. . . , Q _(nk) }. Here, k is the number of dependent coordinates. As shown in FIGS. 6A and 6B, different sets of independent parameters are selected for the one-legged state and the two-legged state.

1.4. Inverse Dynamics The forces and torques generated at each joint can be calculated using a generalized set of coordinate parameters and their derivatives. This operation is a known inverse kinetics. Here, conventional Newton-
The Euler method was used. The Newton-Euler dynamics algorithm is O (n) complex. Here, n indicates the number of DOFs in the system. Since the amount of calculation is linear, it is possible to calculate a complex multi-part system such as a human body.

1.5. Closed loop problem In the state of standing on both feet, the torque generated by both feet is redundant, so that calculation using inverse dynamics is impossible. For this reason, here the forces and torques are distributed to the legs in proportion to the reciprocal of the distance between the legs and the point where the center of gravity of the body is projected on the ground.

1.6. Prediction of muscle strength Many muscles are used to control one joint. Therefore, even if the torque at the joint is calculated, the force generated by each muscle is still not determined. One way to solve this is to use a numerical optimization procedure to predict muscle movements. The dynamics of the joints and muscles of the legs can be expressed by the following equations.

[0040]

(Equation 3)

Here, τ ^hip , τ ^knee , τ
^Angle represents the component of the moment obtained between the rigid bodies in the hips, knees, and ^ankles , respectively.
r ^hip , r ^knee , and ^rank indicate muscle moment arms for hips, knees, and ^ankles , respectively. _{i h,} i k, i _a show hip, knee, and the reference count value for the ankle. f _ih ,
f _ik and f _ia indicate the force generated by each muscle. x
Represents an outer product. Have proposed the following objective function:

[0042]

(Equation 4)

Here, A _i represents the average sectional area of muscle i, and n was 1, 2, 3, or 100. According to their report, a good prediction was obtained when n = 3 when the obtained results were compared with electromyographic data. Also, n = 2
But I could make enough predictions.

The problem with this method is that muscle-tendon dynamics are not taken into account in the muscle strength prediction phase. Even if the muscle activity level is 0, if the distance between the origin of the muscle and the insertion point is sufficiently long, the muscle-tendon exerts elasticity and pulls these two rigid bodies together. Joint angle muscles begin to exert an elastic force is determined by the optimum muscle fiber length (l M _^o) and the length of time of tendon relaxed (l T _^o). Muscle contraction element, peak isometric force at the time of the optimal muscle fiber length (f M _^o) can be exhibited. The strength of the passive force is determined by the elasticity of the muscle passive element (k ^pe ) and the stiffness of the tendon (k ^T ). The passive force of each element can be estimated by the following equation.

[0045]

(Equation 5)

Here, the force exerted by each muscle when the muscle activity level is 0 is defined as f ^pass . f ^T and f
^pass is the same when the muscle activity level is zero. Muscle - length ^{l MT} tendon elastic force is formed becomes longer than its original length ^l _{M o} ^{+ l} _T o.

The force generated by each muscle is at least f
must be greater than ^pass . Therefore, in the embodiment, f ^pass of the muscle group is first calculated, and the total moment τ ^pass around the joint generated by the muscle group is subtracted from the joint torque. This is represented by the following equation.

[0048]

(Equation 6)

As a result, the problem of muscle strength prediction results in prediction of the force (f ^con ) generated by the muscle contraction element. As shown in the following equation, ^{f T} is the sum of the ^{f con} and ^{f pass.}

F ^T = f ^con + f ^pass Here, f ^con is predicted by minimizing an objective function u represented by the following equation. In the embodiment, n =
Set to 2.

[0051]

(Equation 7)

Once f ^T is calculated, l ^MT and v ^MT
Is used to obtain the muscle activity level a as shown by the following equation.

A = f (f ^T , 1 ^MT , v ^MT ) 1.7. Optimal control of human body model The optimal trajectory of the system is calculated using the known Riley-Litz method. In the embodiment, an algorithm of Nagurka et al. Is used (ML Nagurka a).
nd el "Fourier-based opti
mal control of nonlinear
dynamic system ", Transact
ions of the ASME, Journal
of Dynamic Systems, Measur
element and Control, 112, pp.
17-26 (1990)). A Fourier series and a polynomial function are used as the basis functions. Thereby, the independent parameter group of the human body is represented by the following equation.

[0054]

(Equation 8)

The trajectory of q _i (t) is represented by the coefficient c =
(A _im ... A _nm , b _im ... B _nm ). q (t) is determined so as to minimize the following objective function J.

[0056]

(Equation 9) The coefficients of the polynomial are boundary conditions, q _i (t _o ) = q _o q _i (t _f ) = q _f ∂qi (t _o ) / ∂t = ∂q _o / ∂t∂q _i (t _f ) / ∂ It is adjusted so as to satisfy t = ∂q _f / ∂t.

Using inverse dynamics, the torque τ generated at each joint during motion is given by τ = F (q, ∂q / ∂t, ∂ ² q
/ ∂t ² , t). Torque τ is decomposed into muscle force f _i as described above. Finally, the muscle activity level is calculated.

Consider the following objective functions: integration of the square of the torque, integration of the work done by the muscle, integration of the square of the torque change, and finally the integration of the square of the muscle movement (action). The first function is used by Rose et al., Supra. The second function was proposed by Chow et al. For creating a gait. This can be expressed as the following equation.

[0059]

(Equation 10) Here v _i represents the contraction rate of the muscle i. The third function is represented by Equation 1. The last function is expressed by the following equation.

[0060]

[Equation 11] Here, a _i represents the activity level of muscle i. w (a _i )
Is a penalty added when a _i is out of the original range (0 ≦ a _i ≦ 1). The objective function determined in this way is called a minimum muscle motion function.

Since the trajectory of the independent parameter group is a function of the coefficient c of the Fourier series, this objective function is also a function of c, that is, J = f (c). The minimum is calculated using a known BFGS optimization algorithm. The BFGS algorithm is one of the quasi-Newton algorithms.

[0062] 2. A human body model is completed by consideration and preparation beyond posture stability .
Next, consider the stability of the posture that should be considered when the trajectory of the human body is determined. Figure 1 shows the stability considerations.
Provides the basis for S2 and S3 of

When performing a movement such as standing up, walking, and running, a human unconsciously maintains stability. Frank (F
In the posture and movement adjustment method proposed by Rank et al., the central nervous system (CNS) simultaneously plans the main operation and the accompanying posture adjustment operation, and generates each movement at an appropriate time (JS Frank). and el "C
ordination of posture an
d movement ", Physical Ther
apy, 70, pp. 855-863 (1990)).
In order to create an accurate feedforward control model of the whole body, it is necessary to consider this type of auxiliary posture adjustment operation.

2.1. Stability function First, a function to evaluate the stability of the posture is defined. Here, the definition of the stability function is "0 moment point" (ZMP)
Use When a person stands on one or both feet, there is a point where the torque applied to the body from the ground becomes zero. When standing on one foot, this point is on the back of the support foot. In the state of standing on both feet, this point exists in a region surrounded by two feet. Since there is no joint between each foot and the ground, the torque between the sole and the ground is limited. If the torque exceeds that limit, the body falls to the ground.

To check for this constraint, the ZMP
And check if it is included in the support area. This is also done with inverse kinetics. If the ZMP is in the support area, the posture is stable. Outside the area, the supporting feet must be torqued to keep the feet from leaving the ground (this torque is referred to as additional torque). Here, the additional torque is used in the following stability function s.

[0066]

(Equation 12)

Here, τ + represents the minimum additional torque externally applied to the support foot to keep the support foot on the ground.

2.2. Stabilization of Boundary Posture If the key posture, that is, the first and last postures of the motion, is given by the user, there is no problem as long as the ZMP is within the support area. However, if neither posture satisfies this constraint, the values of the independent parameters must be adjusted to stabilize the body. The value s relating to stability described above is used as the objective function. s is minimized until the ZMP is within the support area. FIGS. 7A and 7B show an example in which the boundary posture is stabilized in this way.

2.3. Stabilization of body movement After calculating the trajectory of the body by optimal control, the next step is to stabilize the movement. Even if the posture of the human body model is stable at the beginning and end, it cannot be said that it is stable during operation. In Frank's attitude control method, an additional operation is added to a focus operation for adjusting the attitude. In the embodiment, again, the stability function s is used for adjusting the movement of the whole body. As described above, the trajectory q _m (t) of each generalized parameter is a function of the coefficient c of the Fourier function. The stability of all orbits is defined by the following equation.

[0070]

(Equation 13)

Here, if S is positive, there is some unstable period during operation. s is q, ∂q / ∂t, and ∂ ² q /
Since it is a function of Δt ² , S is also a function of the coefficient c of the Fourier function. Here, c is adjusted to stabilize the operation. This can be achieved by minimizing S. Again, the BFGS algorithm is used for minimization. FIG. 8 shows the trajectory obtained as a result of stabilizing the motion when the person curves backward.

[0072] 3. A desired motion can be obtained by the processing described above. Next, the obtained movement is animated and displayed (S4).
Here, the results of the experiment are shown. In the experiment, three body movements,
That is, squats, kick-ups, and walking were simulated and animated.

As an experimental procedure, first, for each movement, the values of the generalized independent parameter group were specified for the first and last postures. The time for these two positions was also specified. If the movement involved either a foot landing on the ground or a foot lifting off the ground, the time of that moment was also identified. When any of the boundary postures was dynamically unstable, the adjustment was performed by the method described above.

3.1. Squat behavior The squat boundary attitude was dynamically stable and did not require modification. 9 (a) to 9 (d) and FIG. 10 (a)
(D) respectively show a trajectory based on a conventional objective function considering the minimum torque and a trajectory automatically generated using the minimum muscle motion function according to the present invention as an objective function.

FIGS. 9 (a) to 9 (d) and FIGS.
The biggest difference in (d) is the movement of both feet. FIG. 10 (a)
In cases (d), the human body model first stands on the toes, stretches the knees, and then puts the soles on the ground again. On the other hand, in FIGS. 9A to 9D, the soles of the feet are always in contact with the ground. As can be seen, the present embodiment using the minimum muscle motion function generates a more natural motion. The reason for this is that the objective function with the minimum torque does not take into account the physical structure of the human body. When only the minimum torque is considered, the human body is operated like a robot, and the dependence of the joint torque on the joint angle, speed, and acceleration is ignored. As a result, the effectiveness of the human body model and the objective function of the present embodiment was shown.

3.2. Kick motion Figure 11 shows the first and last postures of the kick motion.
(A) and (b) show. The time required for the operation was 0.5 seconds. It is assumed that the right foot leaves the ground 0.1 seconds after the movement starts. Since the final posture was not stable, the stabilization algorithm was applied. FIG. 12A shows a dynamically stable posture, and FIG. 12B shows a motion trajectory.

[0077] Motion optimization includes adjustments to reduce the addition of support legs as they move off the ground. Before leaving the ground, the trunk bends toward the support legs and shifts the center of gravity. This adjustment is also found in actual human movements. Looking at the measured data on the floor load surface of a real human kick motion, before raising the leg, slightly press the ground with the raising leg, move the center of gravity of the body to the support leg, and then the leg starts to separate from the ground I understand. Also in this example, the effectiveness of the present embodiment was found.

3.3. Walking motion A half cycle of walking motion was created. First, the first and last positions were identified. FIG. 13A shows the first posture of walking at t = 0.0, and FIG. 13B shows the last posture of walking at t = 0.5. In addition, the time for the swing leg to leave the ground and reach the ground was also entered. The swing leg is
It is assumed that the player leaves the ground when t = 0.1 and lands when t = 0.4. No stabilization was needed for the first and last positions. The optimal operation was stabilized by minimizing the integration of the stabilization function as described above.

The motion obtained in this way has the following features that are in good agreement with real human motion. That is, FIG.
(A) Sine trajectory of pelvis, trunk swing (swing)
And roll (roll). The pelvic sinusoidal trajectory was also observed in actual motion obtained using a motion capture system, as shown in FIG. Trunk swings seem to work to reduce the load on the support legs as they swing off the ground. Trunk rolls help the swing leg to leave the ground and counteract the torque created by the swing leg.

In the experiment, a walking motion of the infant model was also created. The size and ratio of each rigid body were changed. According to the size of each rigid body, a parameter group for specifying a muscle was estimated. The PCSA of each muscle was set to 1/10 the original value.
The pelvis trajectory is shown in FIG. The infant's pelvic trajectory was arched compared to the adult sinusoidal trajectory. Also, when walking with infants, the swing of the center of gravity was larger than that of adults. This phenomenon is derived from the size ratio of each rigid body and the weakness of muscle strength. This example shows that the present embodiment can be easily and accurately applied to different persons.

[0081] 4. Summary Experience suggests that humans stand and walk to minimize energy. However, this idea has no scientific basis and is currently based on intuition. The motion created from the minimum muscle motion function in the present invention was very close to real human motion. Among the motions using this objective function, a motion known as a preceding posture adjustment also appeared.

As a result of the experiment, the objective function based on the minimum torque change and the like succeeded in the trajectory of the hand, but the natural movement of the leg could not be created. The objective function for minimum muscle work proposed by Chau et al. Also failed. On the other hand, the minimum muscular motion function of the present invention was able to make a good approximation in the lower limb feedforward control.

The system of the present embodiment can be used for education and simulation for improving the performance of athletic sports. In that case, the competitors are more effective because their muscles can be used more efficiently and their movements can be learned. Competitors can also simulate how the results change as they build muscle.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a processing procedure according to an embodiment.

FIG. 2 is a diagram showing a human body model used in the embodiment.

FIG. 3 is a diagram showing a hill-shaped muscle composed of three elements.

FIG. 4: Active muscle element force f ^ce and its length l
^ce and the force l ^pc of the passive muscular element and its length l
It is a figure showing the relationship of ^pc .

5 (a), 5 (b) and 5 (c) show waist,
FIG. 5 is a diagram showing passive elastic moments of the knee and ankle, and FIG.
(A) Passive moment of the knee at 50 degrees of flexion, FIG. 5 (b)
FIG. 5C is a diagram showing the passive moment of the foot when the sole is bent 10 degrees and the waist angle is 12 degrees, and FIG. 5C is a diagram showing the passive moment of the ankle without bending the knee.

FIG. 6A is a diagram showing a state of standing on one leg, and FIG. 6B is a diagram showing a state of standing on both legs.

FIG. 7A is a diagram illustrating an unstable posture when warping behind, and FIG. 7B is a diagram illustrating a stable posture by minimizing a stabilizing function s.

FIG. 8 is a diagram illustrating a trajectory of an operation that is backward of FIG. 7;

FIGS. 9A to 9D are diagrams showing squat operations obtained by a conventional minimum torque method.

FIGS. 10A to 10D are diagrams showing a squat operation obtained in the embodiment.

FIG. 11A shows an initial posture of a kick operation;
FIG. 11B is a diagram illustrating the final posture of the kick operation before stabilization.

12A is a diagram illustrating a final posture of a kick operation after stabilization, and FIG. 12B is a diagram illustrating a trajectory of a lower limb during the kick operation.

13A is a diagram illustrating a first posture of walking at t = 0.0, and FIG. 13B is a diagram illustrating a last posture of walking at t = 0.5.

FIG. 14A is a diagram showing a walking trajectory based on a minimum muscle movement, and FIG. 14B is a diagram showing a trajectory based on an infant model.

FIG. 15 is a diagram illustrating a trajectory of a pelvis during walking.

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-8-329272 (JP, A) JP-A-4-270470 (JP, A) JP-A 8-221599 (JP, A) JP-A-6-221599 342459 (JP, A) JP-A-8-55233 (JP, A) JP-A-4-195476 (JP, A) JP 11-508491 (JP, A) JP 2000-506637 (JP, A) ITO Masami, Introduction to Automatic Control [below], Japan, Shokodo Co., Ltd., October 10, 1996, First Edition, 7th edition, P.E. 254-255 (58) Fields surveyed (Int. Cl. ⁷ , DB name) G05B 13/02 G05B 13/04

Claims (11)

(57) [Claims] 1. A method for simulating the motion of a human body based on dynamics, wherein the motion of the human body is estimated based on an optimal control theory using a minimum muscle motion function as an objective function. 2. The method according to claim 1, wherein the minimum muscular motion function is determined in consideration of muscle and tendon effects. 3. The method according to claim 1, wherein the minimum muscle movement function is determined in consideration of a muscle activity level. 4. The method according to claim 1, further comprising taking into account the balance of the human body and estimating the motion of the human body including the lower limbs. 5. The method according to claim 1, wherein an animation of the human body is generated based on the estimated motion of the human body. 6. A method for simulating the motion of a human body based on dynamics, comprising the steps of: performing feedforward control based on an optimal control theory having a minimum muscle motion function as an objective function; and calculating the motion of the human body. Verifying the posture of the human body in the motion, and correcting the motion of the human body if there is a problem with the posture. 7. The method according to claim 6, wherein an animation of the human body is generated based on the finally obtained motion of the human body. 8. A method of animating the motion of a human body, comprising constructing a muscle model of the human body based on dynamics, estimating the motion of the human body based on optimal control theory, and generating the motion as an animation. Characteristic animation generation method. 9. The method according to claim 8, wherein a minimum muscle motion function is adopted as an objective function of the optimal control theory. 10. The method according to claim 9, wherein the minimum muscular motion function is determined taking into account the action of muscles and tendons. 11. The method according to claim 9, wherein the minimum muscle movement function is determined by considering a muscle activity level.