CN107578461B - Three-dimensional virtual human body physical motion generation method based on subspace screening - Google Patents
Three-dimensional virtual human body physical motion generation method based on subspace screening Download PDFInfo
- Publication number
- CN107578461B CN107578461B CN201710793725.4A CN201710793725A CN107578461B CN 107578461 B CN107578461 B CN 107578461B CN 201710793725 A CN201710793725 A CN 201710793725A CN 107578461 B CN107578461 B CN 107578461B
- Authority
- CN
- China
- Prior art keywords
- motion
- physical
- joint
- root
- individuals
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000033001 locomotion Effects 0.000 title claims abstract description 127
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000012216 screening Methods 0.000 title claims abstract description 26
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 41
- 238000004088 simulation Methods 0.000 claims abstract description 12
- 238000001914 filtration Methods 0.000 claims abstract description 3
- 238000005070 sampling Methods 0.000 claims abstract description 3
- 230000036544 posture Effects 0.000 claims description 29
- 230000006870 function Effects 0.000 claims description 23
- 239000012636 effector Substances 0.000 claims description 20
- 239000013598 vector Substances 0.000 claims description 12
- 230000005484 gravity Effects 0.000 claims description 9
- 230000009471 action Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000005096 rolling process Methods 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 5
- 230000009191 jumping Effects 0.000 claims description 4
- 238000012952 Resampling Methods 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims 1
- 238000005457 optimization Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000011217 control strategy Methods 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 210000001364 upper extremity Anatomy 0.000 description 2
- 241000251730 Chondrichthyes Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000000988 bone and bone Anatomy 0.000 description 1
- 238000009395 breeding Methods 0.000 description 1
- 230000001488 breeding effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 210000003141 lower extremity Anatomy 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Landscapes
- Manipulator (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a three-dimensional virtual human body physical motion generation method based on subspace screening, belonging to the technical field of computer three-dimensional animation, and comprising the following steps: abstracting a human body into a physical model consisting of multiple rigid bodies and joints with hinge structures, and applying torque to the joints to drive the model to move; carrying out filtering denoising and random sampling variation on given human motion data, and taking the data as an initial population input by an evolutionary algorithm; taking the population individual set as an initial target posture set; solving the evolution algorithm, and performing physical simulation on the physical model by combining the PD controller and the current target posture to obtain a plurality of candidate individuals; screening a plurality of individuals by adopting a subspace-based screening algorithm, and selecting the optimal individual as the initial solution of the next moment; and then, carrying out multiple iterations to obtain a physical control track and generate a physical motion.
Description
Technical Field
The invention relates to the technical field of computer three-dimensional animation, in particular to a three-dimensional virtual human body physical motion generation method based on subspace screening.
Background
The human body movement is widely applied to the fields of animation film and television special effects, interactive games, robot simulation, biomechanics and the like. At present, most of the human motion data come from the dynamic capture equipment. The virtual scene contains different terrains or obstacles, and if the motion capture data is directly applied, abnormal phenomena such as the character wearing the ground, wearing the wall or sliding can be caused, so that manual adjustment and later-stage motion editing are often required by an animator, and the workload involved in the process is heavy. In addition, for some complex movements, such as rolling, jumping, crossing and other movements, a plurality of parts of a character are in contact with the environment, a plurality of collision constraints are involved, although geometric hierarchical constraints can be guaranteed by means of manual adjustment and movement editing, physical constraints of human movement can still be damaged, the problems that the acceleration of joints is too large, or the maximum physiological movement limit is exceeded and the like are caused, and the authenticity of the movement is difficult to guarantee.
In view of the above problems, a physical-based human motion generation method has been proposed and becomes a main solution. The method abstracts a human body into a concrete physical model, and places the human body in a virtual environment for simulation to generate motion meeting physical and geometric constraints. The method models a human body into a hinge structure with a plurality of rigid bodies connected, and solves joint moment and external force required by driving a physical model through a space-time optimization method.
The existing physical motion generation track planning method, the low-dimensional physical model method, the control strategy method, the data driving method and the like.
The trajectory planning method sets some target constraint functions by combining with kinematics rules, adopts algorithms such as manual adjustment or offline space-time optimization and the like, adjusts physical parameters or trajectories of the motions to obtain target motions, and can generate motions such as walking, running and the like; the trajectory planning method needs to manually adjust key frames, has large workload, cannot generate complex motion and cannot meet the requirement of motion diversity.
The low-dimensional physical model method adopts a first-order inverted pendulum or spring inverted pendulum model to simulate the gravity center of a human body, combines reverse kinematics and dynamic information to generate motion, and can generate different styles of motion which are suitable for different terrains and have a steering function. Because the low-dimensional physical model method only calculates important joint information such as gravity center, feet and the like for lower limb physical modeling, and other joints reversely calculate by utilizing bone constraint and kinematics rules, only walking motion related to footprints can be generated, and complex actions such as overturning, jumping and the like cannot be generated; in addition, the motion of the upper limbs cannot be truly restored without the help of external input.
The control strategy method is also used to generate physical movements of the human body, and may generate physical movements of the upper limbs of the human body. The control strategy method generates motion by setting a set of motion rules and employing multiple sets of optimal controls. Different walking motions of the role, including motions of different styles and speeds, are generated by setting a group of objective functions such as momentum, end effector, floor point opportunity evaluation and the like and by a priority-based optimization algorithm. The motion method generated by the control strategy requires a researcher to be familiar with key points of a motion sequence and state transition conditions; in addition, the types of generated motions have certain limitations, and the creation of a new motion requires redesigning a set of complex motion rules.
The above method adopts an optimization algorithm or generates motion based on rules, which inevitably results in that the obtained motion is relatively stiff and resembles the motion of a robot. Along with the popularization of motion capture equipment and the establishment of a motion database, motion data is relatively easy to acquire, researchers guide the generation of physical motion by utilizing the motion data, the problem that the physical motion is not natural enough is solved, and vacated complex motion similar to rolling and the like can be generated. Due to the multi-degree of freedom and the strong coupling characteristic of the physical model of the human body, the solving is difficult, so that the solving time of the existing method is long; for the generation of complex physical motion (such as rolling jump) of human body, local minimum value is easy to enter, resulting in failure of final solution; second, motion judder is generated.
Disclosure of Invention
The invention aims to provide a three-dimensional virtual human body physical motion generation method based on subspace screening, and aims to solve the problems that a track planning method and a low-dimensional physical model method in the prior art are difficult to generate complex motion and the generated motion is not natural enough, and the problems that the motion generated by the traditional dynamic capture data driving method is large in jitter, long in optimization time and has certain requirements on the quality of the dynamic capture data.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a three-dimensional virtual human body physical motion generation method based on subspace screening is characterized by comprising the following steps: the method sequentially comprises the following steps:
10) abstracting a human body into a physical model of a skeleton formed by a plurality of rigid bodies and a human body joint formed by a hinge joint, wherein the mass of each rigid body is defaulted to be equal density, and the geometric shape of each rigid body is a cylinder; calculating the difference value between the whole current posture and the target posture of the human body to generate moment by adopting a proportional differential controller (PD controller), and driving the physical model to move by the hinge joint under the action of the moment;
20) acquiring a motion segment from motion data collected from motion capture equipment or key frame data generated by editing of an animator, filtering and denoising the motion segment, removing an error frame in the motion segment, and resampling to obtain motion data; 30) acquiring an initial solution of the attitude structure evolutionary algorithm by adopting equal time intervals according to the motion data acquired in the step 20), and increasing random disturbance on each channel of the acquired attitude to acquire an initial population of the evolutionary algorithm in order to prevent the close-up propagation of population individuals and increase diversity;
40) setting a target function according to the initial population obtained in the step 30) as an initial solution of the evolutionary algorithm, inputting the initial solution into the evolutionary algorithm, generating a moment by the PD controller to carry out evolutionary physical simulation on the human physical model, and calculating to obtain a target function value, namely a fitness value, corresponding to each individual;
50) screening the individuals by adopting a screening algorithm based on space segmentation according to each individual and the corresponding fitness value obtained in the step 40), selecting a plurality of better individuals, recording the target postures and the control tracks of the individuals to form a new population, increasing the time iteration by 0.1 second, turning to the step 60 if the time reaches the motion data end time, and turning to the step 30) if the time does not reach the motion data end time, and performing the next iteration;
60) and driving the human physical model by adopting a PD controller according to the target posture and the control track recorded in each iteration step 50) to generate the human physical motion.
The three-dimensional virtual human body physical motion generation method based on subspace screening is characterized by comprising the following steps: in the step 40), the target function setting is composed of a posture penalty term, a root joint penalty term, an end-effector penalty term, a motion stability penalty term and a motion balance penalty term in a weighting mode.
The three-dimensional virtual human body physical motion generation method based on subspace screening is characterized by comprising the following steps: in the step 40), the evolutionary algorithm optimizes the target postures at multiple moments, slides forward for one moment to continue solving after solving is finished, and solves by adopting a variable-scale sliding window, wherein the size of the window can be set by a user.
The three-dimensional virtual human body physical motion generation method based on subspace screening is characterized by comprising the following steps: in the step 40), the evolutionary algorithm is realized by adopting a covariance matrix adaptive evolutionary strategy algorithm.
The three-dimensional virtual human body physical motion generation method based on subspace screening is characterized by comprising the following steps: in the step 50), based on a screening algorithm of space segmentation, the individuals in the population are segmented in a parameter space by adopting K-Means, and then the individuals in each subspace select the individuals with the optimal fitness value.
According to the invention, a user can generate physical-based real human body motion similar to input data without knowing knowledge of motion rules, human body skeleton physiology and the like and only adjusting parameters such as optimized subspace scale, optimized step length and the like, so that the motion effect and motion generation range generated by the controller are improved.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides an intelligent optimization algorithm framework based on subspace screening, and the method can be used for physically generating a plurality of different physical frameworks and a plurality of different motions.
Aiming at the problem of low convergence speed of the existing physical generation method, the invention ensures that the individuals are optimal in subgroups on the basis of ensuring the individual difference, and performs subspace screening. The method effectively reduces the number of candidate individuals, and compared with the traditional method, the number of candidate solutions is reduced by one order of magnitude, so that the convergence speed of the optimization solution is improved.
Aiming at the problems of stiff physical generated motion, large jitter and entering of a local minimum value, the invention provides a set of objective function punishment system, adopts a sliding window to carry out unified optimization to avoid discontinuity during posture switching, generates more stable motion, has small jitter and improves the motion generation effect.
Drawings
Fig. 1 is a schematic diagram of a virtual human body motion generation method of the present invention.
FIG. 2 is a schematic diagram of an evolutionary algorithm for solving a sliding window of an individual target pose.
FIG. 3 is a schematic diagram of subspace-based screening for the 2D parameter.
Fig. 4 is a diagram of the result of generating the physical motion of the three-dimensional virtual human body.
Detailed Description
The invention integrates the evolution strategy and the space-time optimization algorithm, realizes the physical simulation of the human body movement, is suitable for various different types of human body movements, and improves the movement effect compared with the original method. The invention is further described with reference to the following figures and detailed description.
In one embodiment, the three-dimensional virtual human body physical motion generation method based on subspace screening is realized on a computer with an Intel Xeon CPU E5-2609v3 as a processor, 8GB as a memory and a GeForce Titan X as a display card, wherein the computer adopts a Windows10 operating system, an Open Dynamic Engine (ver0.14) is adopted as a physical bottom layer in software, an Open source project Shark development library is adopted as a CMAES solving algorithm, and the physical simulation time step length is 0.002 seconds. On the basis of the platform, the algorithm schematic diagram shown in fig. 1 is referred to, and the invention is specifically described as follows:
10) the physical model is formed by abstracting a human body into a skeleton formed by multiple rigid bodies and hinge joints, wherein the mass of the rigid bodies is defaulted to be equal in density, and the geometric shape of the rigid bodies is a cylinder. And a proportional differential controller (PD controller) calculates the difference value between the current posture and the target posture to generate a moment, and then the hinge joint drives the physical model to move under the action of the moment.
20) The motion segment is from dynamic capture data or key frame data, the motion segment may have 'jitter' or have an error frame and cannot be directly optimized and solved, and the motion segment needs to be filtered and denoised, the error frame is removed, and resampling is performed.
30) And acquiring an initial solution of the attitude structure evolutionary algorithm by adopting equal time intervals according to the motion data acquired in the step 20), and increasing random disturbance on each channel of the acquired attitude to acquire an initial population of the evolutionary algorithm in order to prevent the population individuals from close-up breeding and increase diversity, wherein the formula is as follows:
θi(j,ti) Indicates the j-th joint time tiThe angle value of (c). m (j, t)i) Representing input motion data at time tiJ th jointThe value of the angle is such that,
representing the time of the interval, and taking the experience value of 0.1 s; and the unit of the random term added behind the xi-N (0,4) motion data is an angle value in order to ensure the difference of population individuals.
40) Setting an objective function as an initial solution of the evolutionary algorithm according to the initial population obtained in the step 30), inputting the initial solution into the evolutionary algorithm, generating a moment by the PD controller to drive the physical model of the human body to perform physical simulation, and calculating to obtain an objective function value (fitness value) corresponding to each individual, wherein the method is implemented as follows:
401) and because many variables are involved in the optimization solving process and the coupling is strong, the single target posture is solved, and the local minimum value is easy to enter, so that the solving fails. In order to avoid trapping in a local minimum value, simultaneously, under the premise of not reducing the sampling frequency of the target postures, the target postures are optimized in a unified mode, and then the sliding window is used for continuously optimizing. As shown in fig. 2, in the present invention, a single window solves 4 target poses at a time, and after the solution is finished, the solution is continued by sliding 2 distances.
402) The method adopts a covariance matrix adaptive evolution strategy (CMA-ES) to solve, and the evolutionary algorithm can well adapt to the high-dimensional constraint problem. Because the acting force between the physical model and the external environment is unpredictable, the controller mainly provides torque for each joint by depending on the target posture to ensure the normal motion of the physical model. The constraint variables are therefore:
wherein
The orientation of each joint representing the ith target attitude, ds is the number of solved target attitudes, and the default is 4. The function Γ (q) is used to transform a unit quaternion point q in a 4-dimensional space to the corresponding 4-dimensional hypersphere coordinates. The invention adopts the offspring number of 20 deltads as the initial stepThe length was 0.2.
403) Setting constraint conditions of a constraint solving algorithm, wherein the physical human model is a plurality of rigid body connecting rod link models, and the motion and mechanics relation between rigid bodies needs to satisfy a robot dynamic equation:
wherein M is,V,G and tau respectively represent an inertia matrix, a non-inertia term, a gravity term and a space generalized force in a dynamic system, F is an external acting force (friction force, supporting force and the like), and theta represents a generalized coordinate vector. J is a Jacobian matrix from a generalized coordinate system to a world coordinate system.
The friction force generated by the physical model and the ground adopts a coulomb friction cone model, wherein V is a linear basis of four groups of friction cones, the weighting coefficient of delta friction cone is, and F is V delta.
404) And then setting an objective function, and weighting and measuring the quality degree of the sample by adopting various objective functions. Wherein the objective function (fitness) is defined as follows:
E＝wpEp+wrEr+weEe+wsEs+wbEb，
attitude penalty term Ep: and (3) examining the difference between the physical motion and the posture in the reference sequence, measuring by using the quaternion difference value of each joint, and weighting and averaging a plurality of joints. Wherein wjFor each joint weight in the overall pose penalty, DqA j-th joint quaternion q for the current posturejJ-th joint quaternion q with reference attitudej_refDistance between jnumAll joints are considered equivalent in the present invention for the number of all joints, and the joint weight coefficient is 1, which has the following formula:
root joint penalty term Er: the difference of the physical motion from the root joint of the reference sequence is examined. The purpose of the control of the joint is to ensure imitationThe overall orientation of the role of the true sequence and the reference sequence is as close as possible. Wherein q isrootQuaternion of the root joint of the current pose, qroot_refFor the quaternion of the root joint of the reference pose, there is the following equation:
Er＝Dq(qroot,qroot_ref)，
end-effector penalty term Ee: examine the closeness of end effectors such as feet. Ensuring that the end part reaches the correct position at the correct moment. For walking movement, the end effectors are both feet; for jumping and rolling, the head and hands are considered. Wherein enumNumber of end effectors, DvFor the e-th end effector vector p of the current poseeAnd reference attitude e-th end effector vector pe_refThe distance between, having the formula:
penalty term for smooth movement Ep: the stability of the movement is considered, and the effect of suppressing the excessive angular velocity is achieved to a certain extent. In the walking control, the motion amplitude is not easy to be too large, although a certain deviation between the current sequence and the reference sequence is allowed, the corresponding angular speed needs to be controlled within a certain amplitude, and a higher weight needs to be set for the item. However, for actions with large angular velocities, such as skip, the weight needs to be reduced. Wherein ω isroot、ωroot_refAngular velocity, omega, of the root joint for the current attitude and the reference attitudej、ωj_refAngular velocity, v, of j-th joint for current attitude and reference attitudee、ve_refThe velocity of the e-th end effector for the current pose and the reference pose is given by:
motion balance penalty term Ee: looking at the degree of offset of the end effector with respect to the center of gravity,the center of gravity is ensured to be in a stable area of the respective step support, and the term is used for controlling walking movement. Wherein com, comrefThe vector for the center of gravity of the current attitude and the reference attitude has the following formula:
50) screening the individuals by adopting a screening algorithm based on space segmentation according to the individuals obtained in the step 40) and the corresponding fitness values, selecting a plurality of better individuals, recording the target postures and the control tracks of the individuals to form a new population, increasing the time iteration by 0.1 second, turning to the step 60 if the time reaches the motion data end time, and turning to the step 30) if the time does not reach the motion data end time, and specifically implementing the following steps:
and respectively optimizing different initial values by a plurality of groups of CMA-ES evolutionary algorithms to obtain a group of physical motion postures. The physical state of each sample is
Jointly determined by the first several target poses:
where Dynamic is the physical simulation process function, pr,vr,arRespectively, the displacement, linear velocity, linear acceleration, Q of the root jointT,
Then the angular displacement vector, angular velocity vector, angular acceleration vector, E corresponding to all joints (including the root joint) of the character is the objective function value of the current simulation time period.
And k is the number of target postures required from the starting time to the simulation time t.
In order to ensure the diversity of subsequent optimization solution, all filial generations generated by the CMA-ES need to be screened, and two preconditions, namely the difference between individuals and the individual adaptability are ensured to be optimal as much as possible.
Firstly, primarily screening all offspring individuals, discarding 30% of individuals with backward fitness ranking, and rejecting individuals with particularly poor fitness; then, the remaining individuals are selected to be the 10 closest to the current target pose (less than 10 total selections), and the parameter subspace is constructed by using the group of target poses as follows:
the function Γ transforms a set of object poses into a set of independent parameters pariAnd the vectors are used as characteristic parameters of the ith individual in the subspace, all the individual parameter vectors are clustered by using a K-Means algorithm to obtain n subclasses, and then the individual with the optimal fitness is searched in each subclass and is used as an initial solution of the next iteration. FIG. 3 shows the parameter pariAnd the dimension is 2, the number n of the subclasses is 5, and the thickened vertex is the initial solution of the next iteration.
60) And driving the physical human body model by adopting a PD controller according to the target postures and the control tracks recorded in each iteration step 50) to generate physical human body motions, as shown in figure 4.
Claims (1)
1. A three-dimensional virtual human body physical motion generation method based on subspace screening is characterized by comprising the following steps: the method sequentially comprises the following steps:
10) abstracting a human body into a physical model of a skeleton formed by a plurality of rigid bodies and a human body joint formed by a hinge joint, wherein the mass of each rigid body is defaulted to be equal density, and the geometric shape of each rigid body is a cylinder; calculating the difference value between the whole current posture and the target posture of the human body to generate moment by adopting a proportional differential controller (PD controller), and driving the physical model to move by the hinge joint under the action of the moment;
20) acquiring a motion segment from motion data collected from motion capture equipment or key frame data generated by editing of an animator, filtering and denoising the motion segment, removing an error frame in the motion segment, and resampling to obtain motion data;
30) acquiring an initial solution of the attitude structure evolutionary algorithm by adopting equal time intervals according to the motion data acquired in the step 20), and increasing random disturbance on each channel of the acquired attitude to acquire an initial population of the evolutionary algorithm in order to prevent the close-up propagation of population individuals and increase diversity;
40) setting a target function according to the initial population obtained in the step 30) as an initial solution of the evolutionary algorithm, inputting the initial solution into the evolutionary algorithm, generating a moment by the PD controller to carry out evolutionary physical simulation on the human physical model, and calculating to obtain a target function value, namely a fitness value, corresponding to each individual;
50) screening the individuals by adopting a screening algorithm based on space segmentation according to each individual and the corresponding fitness value obtained in the step 40), selecting a plurality of better individuals, recording the target postures and the control tracks of the individuals to form a new population, increasing the time iteration by 0.1 second, turning to the step 60 if the time reaches the motion data end time, and turning to the step 30) if the time does not reach the motion data end time, and performing the next iteration;
60) driving a human physical model by adopting a PD controller according to the target posture and the control track recorded in each iteration step 50) to generate human physical motion; in the step 40), the target function setting is composed of a posture penalty term, a root joint penalty term, an end-effector penalty term, a motion stability penalty term and a motion balance penalty term in a weighting mode; in the step 40), the evolutionary algorithm optimizes the target postures at a plurality of moments, slides forward for a moment to continue solving after solving is finished, and adopts a variable-scale sliding window to solve, wherein the size of the window can be set by a user;
in the step 40), the evolutionary algorithm is realized by adopting a covariance matrix self-adaptive evolutionary strategy algorithm;
in the step 50), based on a screening algorithm of space segmentation, individuals in the population are segmented in a parameter space by adopting K-Means, and then the individuals in each subspace select the individuals with the optimal fitness value;
the obtained initial population is used as an initial solution of an evolutionary algorithm, a target function is set, the initial solution is input into the evolutionary algorithm, a PD controller generates a moment to carry out evolutionary physical simulation on a human physical model, a target function value, namely a fitness value, corresponding to each individual is obtained through calculation, the PD controller is adopted to drive the human physical model according to a target posture and a control track recorded in each iteration, and human physical motion is generated;
the evolutionary algorithm comprises:
on the premise of not reducing the sampling frequency of the target postures, uniformly optimizing a plurality of target postures, then continuously optimizing by sliding a window, solving 4 target postures by a single window once, and continuously solving by sliding 2 distances after solving is finished;
weighting and measuring the quality degree of the sample by adopting a plurality of objective functions, wherein the objective functions are defined as follows:
E＝wpEp+wrEr+weEe+wsEs+wbEb，
wherein,
attitude penalty term Ep: examining the difference between the physical motion and the posture in the reference sequence, measuring by using the quaternion difference value of each joint, and weighting and averaging a plurality of joints; w is ajFor each joint weight in the overall pose penalty, DqA j-th joint quaternion q for the current posturejJ-th joint quaternion q with reference attitudej_refDistance between jnumAll joints are equivalent for the number of all joints, the joint weight coefficient is 1, and there is the following formula:
root joint penalty term Er: examining the difference of the physical motion and the root joint of the reference sequence; wherein q isrootQuaternion of the root joint of the current pose, qroot_refFour of root joints as reference posturesThe number of elements, having the formula:
Er＝Dq(qroot，qroot_ref)，
end-effector penalty term Ee: inspecting the similarity degree of foot end effectors; ensuring that the tail end part reaches the correct position at the correct time; for walking movement, the end effectors are both feet; corresponding to jumping and rolling actions, head and hand effectors need to be considered; wherein emumNumber of end effectors, DvFor the alpha end effector vector p of the current poseeAnd reference attitude e-th end effector vector pe_refThe distance between, having the formula:
motion stationary penalty term Es: investigating the stability of the movement; wherein ω isroot、ωroot_refAngular velocity, omega, of the root joint for the current attitude and the reference attitudej、ωj_refAngular velocity, v, of j-th joint for current attitude and reference attitudee、ve_refThe velocity of the e-th end effector for the current pose and the reference pose is given by:
motion balance penalty term Eb: the deviation degree of the end effector relative to the gravity center is inspected, the gravity center is ensured to be in the stable area of the respective footstep support, and the item is used for controlling the walking movement; wherein com, comrefThe vector for the center of gravity of the current attitude and the reference attitude has the following formula:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710793725.4A CN107578461B (en) | 2017-09-06 | 2017-09-06 | Three-dimensional virtual human body physical motion generation method based on subspace screening |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710793725.4A CN107578461B (en) | 2017-09-06 | 2017-09-06 | Three-dimensional virtual human body physical motion generation method based on subspace screening |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107578461A CN107578461A (en) | 2018-01-12 |
| CN107578461B true CN107578461B (en) | 2021-05-04 |
Family
ID=61031259
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710793725.4A Active CN107578461B (en) | 2017-09-06 | 2017-09-06 | Three-dimensional virtual human body physical motion generation method based on subspace screening |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107578461B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805965B (en) * | 2018-05-18 | 2022-03-15 | 合肥工业大学 | Human physical motion generation method based on multi-objective evolution |
| CN109318230A (en) * | 2018-09-29 | 2019-02-12 | 鲁东大学 | Robot motion optimization method, device, computer equipment and storage medium |
| CN110782513A (en) * | 2019-10-30 | 2020-02-11 | 北京中科深智科技有限公司 | Method for real-time motion capture data debouncing composite algorithm |
| CN111445475B (en) * | 2020-02-26 | 2022-06-24 | 清华大学 | Human model attitude prediction method based on reverse dynamics |
| CN112001989B (en) * | 2020-07-28 | 2022-08-05 | 完美世界(北京)软件科技发展有限公司 | Virtual object control method and device, storage medium and electronic device |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106773739A (en) * | 2017-02-28 | 2017-05-31 | 山东大学 | Method for planning track of robot based on hereditary chaotic optimization algorithm |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2852932A1 (en) * | 2012-05-22 | 2015-04-01 | Telefónica, S.A. | A method and a system for generating a realistic 3d reconstruction model for an object or being |
| CN104331911A (en) * | 2014-11-21 | 2015-02-04 | 大连大学 | Improved second-order oscillating particle swarm optimization based key frame extraction method |
-
2017
- 2017-09-06 CN CN201710793725.4A patent/CN107578461B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106773739A (en) * | 2017-02-28 | 2017-05-31 | 山东大学 | Method for planning track of robot based on hereditary chaotic optimization algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107578461A (en) | 2018-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107578461B (en) | Three-dimensional virtual human body physical motion generation method based on subspace screening | |
| US11928765B2 (en) | Animation implementation method and apparatus, electronic device, and storage medium | |
| JP7061238B2 (en) | Reinforcement learning to train characters using heterogeneous target animation data | |
| CN111292401B (en) | Animation processing method and device, computer storage medium and electronic equipment | |
| CN108115681B (en) | Simulation learning method and device for robot, robot and storage medium | |
| JP5836565B2 (en) | Robot tracking and balancing system and method for mimicking motion capture data | |
| Hämäläinen et al. | Online motion synthesis using sequential monte carlo | |
| CN110764416A (en) | Humanoid robot gait optimization control method based on deep Q network | |
| Kenwright | Everything must change with character-based animation systems to meet tomorrows needs | |
| CN115464659A (en) | Mechanical arm grabbing control method based on deep reinforcement learning DDPG algorithm of visual information | |
| Ruchanurucks et al. | Humanoid robot motion generation with sequential physical constraints | |
| Ha et al. | Evolutionary optimization for parameterized whole-body dynamic motor skills | |
| Brogan et al. | Simulation level of detail for multiagent control | |
| Dong et al. | Passive bimanual skills learning from demonstration with motion graph attention networks | |
| Naderi et al. | A reinforcement learning approach to synthesizing climbing movements | |
| Wei | A comprehensive approach to the generation of human-like arm movements on robot NAO | |
| Van de Panne | Control techniques for physically-based animation. | |
| CN108805965B (en) | Human physical motion generation method based on multi-objective evolution | |
| Kouchaki et al. | Balance control of a humanoid robot using deepreinforcement learning | |
| CN118012077B (en) | Four-foot robot motion control method and system based on reinforcement learning motion simulation | |
| WO2024060833A1 (en) | Image processing method | |
| Greer | Physics-based character locomotion control with large simulation time steps. | |
| CN116062059A (en) | Single-leg robot continuous jump control method based on deep reinforcement learning | |
| Mellema | Exoskeleton controller design for cybathlon by parallel reinforcement learning | |
| Ling | Techniques in learning-based approaches for character animation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |