JP7150306B2 - Motion generation method, device and program - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Act Proceedings of the International Conference "IEEE International Conference on Robotics and Automation (ICRA) 2018" (at Brisbane Convention and Exhibition Center, Brisbane, Queensland, Australia), May 21, 2018 ( Publication date) [Publications, etc.] Proceedings of the 23rd Robotics Symposia (Venue: Kuroshio Onsen Shofukaku, Yaizu City, Shizuoka Prefecture), March 14, 2018 (Publications, etc.) [Publications, etc.] ROBOMECH Journal, 2018, vol. 5, no. 10, May 18, 2018 (date of issue)

The present invention relates to a method, apparatus and program for generating motion.

A “SLIP (Spring Loaded Inverted Pendulum) model” is available as a method of simulating the complex motion of the human body using a simple model consisting of mass points, springs, and dampers (see, for example, Non-Patent Document 1).

"The spring-mass model for running and hopping" Blickhan, R. Journal of Biomechanics Vol. 22, No. 11/12 pp. 1217-1227 (1989)

Kinematics analysis is a technique for measuring movements of various parts of the human body, applying a kinematics model to the measured data, and executing analysis. Such kinetic analysis has demonstrated its power in the fields of sports science and biomechanics, and has given rise to various applications such as the analysis of sports performance and clarification of the occurrence of injuries.

Conventional methods in kinematics analysis make it possible to analyze and verify actual performance and the movement of the human body before injury occurs. For example, when applied to sports science, it is possible to ask, ``What posture did the athlete take and what movements did each part of the body perform when a good performance was achieved, or what is the difference from a bad performance? ” can be verified. Also, when applied to biomechanics, it is possible to elucidate to some extent how humans normally behave in a safe environment such as a laboratory.

On the other hand, there is a very strong demand for direct knowledge on how to achieve performances that have not yet been realized, such as performances that exceed world records or personal records of certain athletes. In addition, it is useful to elucidate the mechanism of injury occurrence based on the movement of the body and the way the body collapses when an injury actually occurs, in order to take preventive measures against injuries.

However, motions that have not yet been realized cannot be analyzed using conventional methods because there is, of course, no measurement data for the motions. Similarly, regarding the occurrence of injury, it is practically impossible to carry out an experiment in which the subject is in danger, so it is impossible to perform a motion analysis when an injury actually occurs.

Therefore, when attempting to analyze such unrealized motion, it is necessary to generate the motion computationally.
For example, in the case of robots,
1) Joint angle and task-based motion generation,
2) forward dynamics simulation;
can generate a new motion.
On the other hand, in the case of human whole-body exercise,
1) Human body parts such as joints have extremely complicated structures, and the number of degrees of freedom of each part is much higher than that of robots.
2) Since the human body is not fixed to the environment such as the floor, it is necessary to estimate the contact force with the environment, but the calculation is generally difficult and the result is unstable.
Because of these factors, it has been difficult to generate new movements. Based on these findings, the inventors have come to recognize that generating a mechanically consistent human whole-body motion is an important issue in this field.

The present invention has been made in view of such circumstances, and its object is to computationally generate movements that have not yet been realized.

In order to solve the above problems, a method according to one aspect of the present invention includes the steps of acquiring motion data of motion in a plurality of parts of the body, reproducing whole-body motion from the motion data, reproducing the whole-body motion a step of extracting the feature quantity of the first group from, applying the feature quantity of the first group to the motion expression model to calculate the model parameters of the first group, and the model of the first group generating a second group of model parameters by changing parameter values; applying the second group of model parameters to the motion expression model to calculate the second group of feature quantities; Prepare.

Another aspect of the invention is an apparatus. This apparatus includes a data acquisition unit that acquires motion data of motion in a plurality of parts of the body, a reproduction unit that reproduces the whole-body motion from the motion data, and a feature amount of the first group from the reproduced whole-body motion. a feature amount extraction unit for extracting; a model parameter calculation unit for applying the feature amount of the first group to the motion expression model to calculate the model parameters of the first group; A model parameter changing unit that changes and generates model parameters of the second group, and a feature amount calculation unit that applies the model parameters of the second group to the motion expression model and calculates the feature amount of the second group. And prepare.

Yet another aspect of the present invention is a program. This program includes the steps of acquiring motion data of motion in multiple parts of the body, reproducing whole-body motion from the motion data, and extracting a first group of features from the reproduced whole-body motion. and applying the feature amount of the first group to the motion expression model to calculate the model parameters of the first group; The step of generating model parameters and the step of applying the second group of model parameters to the motion expression model to calculate the second group of feature quantities are executed by a computer.

Any combination of the above constituent elements, and any conversion of the expressions of the present invention into devices, methods, systems, recording media, computer programs, etc. are also effective as embodiments of the present invention.

According to the present invention, it is possible to generate new motions, including motions that have not yet been realized.

1 is a flow diagram of a motion generation method according to one embodiment; FIG. 1 is a schematic diagram showing the concept of a SLIP model; FIG. FIG. 5 is a flow diagram of a motion generation method according to a further embodiment; FIG. 5 is a flow diagram of a motion generation method according to a further embodiment;

(Embodiment 1)
FIG. 1 is a flow diagram of a motion generation method according to one embodiment.
At step S10, the method acquires motion data of movements at multiple parts of the body. One example of a means of acquiring motion data is optical motion capture. In optical motion capture, a camera captures the movement of a subject with markers attached to multiple parts of the body. Motion data, such as position, velocity, acceleration, etc., of the site to which the marker is attached can be obtained from the captured image. Note that the motion capture is not limited to the optical type, and may be a mechanical type, a magnetic type, a video type, or the like. Another example of a means of acquiring motion data is a force plate. The force plate measures the force and moment applied to the upper surface of the force plate according to the motion state of the subject, such as standing, stepping, and jumping. In addition, motion data may be acquired using existing means such as wearable IMU sensors and pressure sheets (installed type, wearable type) (detailed description is omitted).

In step S20, the method reproduces whole-body motion using a digital human model or the like from the motion data of each marker site acquired in step S10. For example, body posture, movement speed, stride length, knee angle, fingertip movement, joint angle and joint torque of each joint, and the like are reproduced. These reproductions are performed by "inverse kinematics calculation" based on inverse kinematics and "inverse dynamics calculation" based on inverse dynamics.

At step S30, the method extracts features for the motion to be generated from the whole-body motion reproduced at step S20. For example, when it is desired to generate an unknown walking motion or running motion, the feature quantity is the locus of the center of mass, the locus of the sole pressure center, the leg contact force, and the like.
Hereinafter, the feature amount (singular or plural) extracted in step S30 may be collectively referred to as "first group feature amount".

In step S40, the method applies the first group of features extracted in step S30 to the motion expression model to calculate model parameters. This makes it possible to express complex whole-body movements in the form of a simple model. An example of a motion representation model is the SLIP model. As described above, the SLIP model models the human body as consisting of mass points, springs, and dampers.

The SLIP model will be described in detail below.
FIG. 2 is a schematic diagram showing the concept of the SLIP model. A body model 100 representing a human body comprises mass points 110 , springs 120 , dampers 130 , contact points 140 , first shaft 150 and second shaft 160 . Mass point 110 is the center of mass of the body. Spring 120 and damper 130 are arranged in parallel. The contact point 140 is located on the opposite side of the mass point 110 across the spring 120 and the damper 130, and can come into contact with the environment such as the floor surface. A first shaft 150 connects mass point 110 , spring 120 and damper 130 . A second shaft 160 connects spring 120 and damper 130 with contact point 140 .

Hereinafter, the terms "right" and "left" refer to orientations when the page is viewed from the front. At time T1, the body model 100 is jumping with its posture tilted slightly to the left. At this time, the contact point 140 is away from the environment. At time T2, the physical model 100 lands with its posture tilted slightly to the left. That is, at time T2, contact point 140 makes contact with the environment. From time T2 to T3, body model 100 tilts its posture to the right while contact point 140 remains in contact with the environment. At time T3, the body model 100 jumps while tilting its posture slightly to the right. That is, at time T3, contact point 140 leaves the environment. From time T3 to time T4, body model 100 tilts its posture to the left while keeping contact point 140 away from the environment. At time T4, the body model 100 is jumping with its posture tilted slightly to the left.

With respect to such motion, by experimentally measuring the leg contact force F _leg (t) and the leg length L _leg (t) at times T1, T2, T3 and T4, the spring coefficient K _leg and the damper coefficient D _leg , leg natural length L _{leg, 0,} etc. can be calculated. Specifically, K _leg , D _leg , L _{leg, 0} are calculated by optimizing to minimize E _f represented by the following equation (1).

The model parameters calculated in this step S40 reflect physical features such as body shape and physique of the subject, and exercise features such as physical ability, posture, habits, and the like. These parameters therefore contain just enough information about the subject's whole-body motion, which is the basis for generating new motion.

The calculation executed in step S40 is a "mathematical optimization calculation" for calculating model parameters based on motion-related feature amounts such as the locus of the center of mass, the locus of the center of sole pressure, and the leg contact force.
Hereinafter, the model parameters (which may be singular or plural) calculated in step S40 may be collectively referred to as "first group model parameters".

In step S50, the method varies the values of the first group of model parameters K _leg , D _leg , L _{leg, 0} calculated in step S40 to generate new values of the model parameters. At this time, in addition to the values of the model parameters, the values of the environment variables (for example, the initial position of the center of mass H _COM (0) and the initial velocity of the center of mass V _COM (0)) are changed, and the new values of the environment variables are changed. may be generated. The process of step S50 means that the motion of the subject is changed from the motion when the data was acquired in step S10 to a new motion. Specifically, by changing the model parameters K _leg , D _leg , L _{leg, and 0} , it is possible to change the joint hardness and the like. Also, by changing the environmental variables H _COM (0) and V _COM (0), the height of the floor surface, etc. can be changed.

In this way, by changing various model parameters relating to the subject, it is possible to variously transform the exercise to be achieved by the subject. These altered movements include any new movements, such as movements the subject has not yet achieved, or movements when an injury occurs.
Hereinafter, the new model parameters (which may be singular or plural) generated in step S50 may be collectively referred to as "second group of model parameters".

At step S60, the method applies the second group of model parameters and environmental variables to the model of step S40 to calculate features related to the motion transformed at step S50. This feature amount is a feature amount related to new motion based on the second group of model parameters. As described in step S30, for example, when the exercise is walking or running, the feature amounts to be calculated are the center-of-mass trajectory, the center-of-sole-pressure trajectory, and the leg contact force.

In this way, by changing the model parameters related to a certain subject, a new motion of the subject is generated. Is it okay?" becomes clear. In other words, by finding appropriate model parameters, it is possible, for example, to generate exercises that achieve high performance that the subject has not yet achieved.

Note that while the calculation executed in step S20 is an inverse dynamics calculation, the calculation executed in step S60 is based on the model parameters, and is related to motion such as the center-of-mass trajectory, the center-of-sole-pressure trajectory, and the leg contact force. It is a "forward dynamics calculation" that calculates feature values.
Hereinafter, the motion feature amount (singular or plural) calculated in step S60 may be collectively referred to as "second group motion feature amount".

This completes the description of the present embodiment. According to this embodiment, by acquiring the actual exercise data of the subject, it is possible to generate new exercises including exercises that the subject has not yet achieved.

(Embodiment 2)
FIG. 3 is a flow diagram of a motion generation method according to a further embodiment. This embodiment is the method shown in FIG. 1, and further comprises a step S70 of reconstructing the whole-body motion based on the motion-related feature amount calculated in step S60. Since the processing from step S10 to step S60 is common to the method of FIG. 1, description thereof is omitted.

ここで、Ｐ_ｍａｒ：マーカ位置、Ｐ_ＣＯＭ：質量中心位置、Ｐ_ＣＯＰ：足底圧力中心位置、である。 In step S70, the method reconstructs the whole-body motion based on the motion features of the second group. Specifically, the parameters α and β are set so as to minimize _Em represented by the following equation (2) in the motion data acquired in step S10 and the first group feature amount extracted in step S30. Optimize. Then, by substituting the motion-related feature amount calculated in step S60 into Equation (2), the motion trajectories of all the markers can be reconstructed.

Here, P _mar : Marker position, P _COM : Mass center position, P _COP : Plantar pressure center position.

According to this embodiment, the whole-body motion of the subject can be reconstructed and expressed with respect to the new motion generated in step S60.

(Embodiment 3)
FIG. 4 is a flow diagram of a motion generation method according to a further embodiment. This embodiment is the method shown in FIG. 3, further comprising a step S80 of analyzing the whole body motion reconstructed in step S70. Since the processing from step S10 to step S70 is common to the method of FIG. 3, description thereof is omitted.

At step S80, the method analyzes the reconstructed whole-body motion. Analysis of whole-body motion may include comparing and evaluating the newly reconstructed whole-body motion with the original measured data. Analysis of whole-body motion may also include comparing, evaluating, etc., the newly reconstructed whole-body motion with an ideal motion that achieves the target performance.

According to this embodiment, kinematics analysis can be performed based on the new whole-body motion reconstructed in step S70.

(Embodiment 4)
A further embodiment is the method shown in FIGS. 1, 3 and 4, wherein in step S10 motion data for motion at multiple sites of the artificial limb are combined in addition to motion data for motion at multiple sites of the body. may be obtained. At this time, the model parameters to be changed in step S50 may include model parameters related to the artificial limb (for example, the viscoelastic coefficient of the artificial limb).

People who have had their upper or lower limbs amputated (hereinafter referred to as "limb amputees") generally wear artificial limbs such as artificial arms and legs. When an amputee wearing an artificial limb runs, the movement involves not only the movement of the body, but also the complex interaction between the body and the artificial limb. This makes the kinematics analysis more difficult. On the other hand, there is a high demand for comfortable artificial limbs that are free from dangers such as falls. Furthermore, in recent years, the number of amputees who wear artificial limbs has increased and their performance has improved remarkably. In order to develop such a safe and high-performance artificial limb, it is necessary to computationally generate motion while wearing an artificial limb, especially one that is not actually worn by the subject or has not yet been realized. It is desirable to be able to analyze this. The present embodiment meets such needs.

According to this embodiment, regarding the motion of a human wearing an artificial limb, it is possible to generate a motion that the human has never realized, or a motion in a state of wearing an artificial limb different from the artificial limb.

The present invention has been described above based on the examples. Those skilled in the art will understand that these examples are illustrative only, and that various modifications can be made to combinations of each component and each treatment process, and such modifications are within the scope of the present invention. be.

For example, in addition to the above applications, the present invention can be applied to various applications such as sports training, development of sports equipment, design of buildings and facilities with low risk of injury, medical care for the elderly, rehabilitation, and control of walking assist suits. . The method according to the present invention can be applied to such various fields and has high industrial applicability.

S10 Step of acquiring motion data S20 Step of reproducing whole body motion S30 Step of extracting feature amount S40 Step of calculating model parameter S50 Step of changing model parameter S60 Step of calculating feature amount S70 Whole body Reconstructing motion S80 Analyzing whole body motion

Claims (5)

obtaining motion data for motion in multiple parts of a subject's body;
recreating whole-body motion from the motion data;
extracting a first group of features from the reproduced whole-body motion;
a step of applying the first group of feature quantities to a motion expression model to calculate a first group of model parameters;
varying the values of the first group of model parameters to generate a second group of model parameters for a new movement of the subject ;
a step of applying the model parameters of the second group to the motion expression model to calculate a feature amount of the second group;
a step of reconstructing whole-body motion based on the second group of features;
analyzing the reconstructed whole body motion;
with
The method of claim 1, wherein analyzing the reconstructed whole-body motion is performing a kinematics analysis . wherein the motion data includes motion data of motion at multiple parts of the artificial limb;
The method of claim 1 . wherein the first group of model parameters includes model parameters relating to an artificial limb;
3. The method of claim 2 . a data acquisition unit that acquires motion data of motion in multiple parts of the subject's body;
a reproduction unit that reproduces a whole-body movement from the movement data;
a feature extraction unit that extracts a feature of the first group from the reproduced whole-body motion;
a model parameter calculation unit that applies the feature amount of the first group to the motion expression model to calculate model parameters of the first group;
a model parameter changing unit that changes the values of the first group of model parameters to generate a second group of model parameters for a new movement of the subject ;
a feature quantity calculation unit that applies the model parameters of the second group to the motion expression model to calculate the feature quantity of the second group;
a whole-body motion reconstruction unit that reconstructs a whole-body motion based on the feature amount of the second group;
an analysis unit that analyzes the whole-body motion reconstructed in the whole-body motion reconstruction unit;
prepared,
The analysis unit is a device that performs kinematics analysis . obtaining motion data for motion in multiple parts of a subject's body;
recreating whole-body motion from the motion data;
extracting a first group of features from the reproduced whole-body motion;
a step of applying the first group of feature quantities to a motion expression model to calculate a first group of model parameters;
varying the values of the first group of model parameters to generate a second group of model parameters for a new movement of the subject ;
a step of applying the model parameters of the second group to the motion expression model to calculate a feature amount of the second group;
a step of reconstructing whole-body motion based on the second group of features;
analyzing the reconstructed whole body motion;
on the computer, and
The program , wherein the step of analyzing the reconstructed whole-body motion performs a kinematics analysis .

Images