CN104345735A - Robot walking control method based on foothold compensator - Google Patents
Robot walking control method based on foothold compensator Download PDFInfo
- Publication number
- CN104345735A CN104345735A CN201410521894.9A CN201410521894A CN104345735A CN 104345735 A CN104345735 A CN 104345735A CN 201410521894 A CN201410521894 A CN 201410521894A CN 104345735 A CN104345735 A CN 104345735A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- compensator
- msup
- math
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 93
- 230000008569 process Effects 0.000 claims abstract description 65
- 238000004364 calculation method Methods 0.000 claims abstract description 7
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 206010034701 Peroneal nerve palsy Diseases 0.000 claims description 20
- 230000001133 acceleration Effects 0.000 claims description 16
- 239000008186 active pharmaceutical agent Substances 0.000 claims description 13
- 238000012986 modification Methods 0.000 claims description 9
- 230000004048 modification Effects 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 230000005484 gravity Effects 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 239000003607 modifier Substances 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005021 gait Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Landscapes
- Feedback Control In General (AREA)
- Manipulator (AREA)
Abstract
The invention relates to a robot walking control method based on a foothold compensator. The robot walking control method comprises the following steps of firstly, establishing a constraint dynamic model of a robot; secondly, designing the foothold compensator based on the constraint dynamic model according to the constraint dynamic model; thirdly, establishing a heteroscedastic sparse Gauss process regression model and realizing the mapping calculation from input to output of the foothold compensator; fourthly, locally updating the heteroscedastic sparse Gauss process regression model; fifthly, establishing the foothold compensator based on the heteroscedastic sparse Gauss process regression model; sixthly, performing prediction control over the walking of the robot according to the foothold compensator based on the heteroscedastic sparse Gauss process regression model. Compared with the prior art, the robot walking control method disclosed by the invention has the advantages of accurate prediction, high learning speed and the like.
Description
Technical Field
The invention relates to the field of robot motion control, in particular to a robot walking control method based on a foothold compensator.
Background
The dynamic characteristic and controllability of the humanoid robot can be adjusted by dynamically adjusting the foot-falling point of the robot, so that the suppression capability of the robot on unknown disturbance can be improved. At present, the implementation of the foothold compensator is mostly based on a simple linear controller, and for a humanoid robot, the walking process of the humanoid robot is a complex nonlinear system. Therefore, even if the linear foot-drop point compensator can obtain better gain parameters of the foot-drop point compensator through a learning method, a considerable error still exists between the final foot-drop point compensator strategy and the dynamic characteristics of the robot. In addition, if the learning process of the foothold compensator of the humanoid robot is long, the foothold compensator is not suitable for fast and real-time foothold compensation skill learning of the robot in an unknown environment. Therefore, the invention introduces the nonlinear characteristic and designs the foothold compensator based on the different variance sparse Gaussian process (HspGP) model. In order to realize real-time online adjustment of control of the foothold compensator, a local updating method based on an heteroscedastic Gaussian process model is provided. The local updating method has the advantages of small calculation complexity and high learning rate, and provides a mathematical basis for the online learning of the humanoid robot foothold compensator.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a robot walking control method based on a foothold compensator.
The purpose of the invention can be realized by the following technical scheme:
a robot walking control method based on a foothold compensator comprises the following steps:
1) establishing a constraint dynamic model of the robot;
2) designing a constraint dynamic model foot point compensator according to the constraint dynamic model;
3) establishing a different variance sparse Gaussian process regression model to realize the mapping calculation from input to output of the foothold compensator;
4) locally updating a different variance sparse Gaussian process model;
5) building a foot point compensator based on a variance sparse Gaussian process model;
6) and performing predictive control on the robot walking according to the footfall compensator based on the heteroscedastic sparse Gaussian process model.
The step 1) comprises the following steps:
11) establishing a three-dimensional linear inverted pendulum mechanical model of the robot, wherein the expression of the mechanical model is as follows:
xz＝Cxc
wherein, <math><mrow>
<msub>
<mi>x</mi>
<mi>c</mi>
</msub>
<mo>=</mo>
<msup>
<mrow>
<mo>[</mo>
<msub>
<mi>x</mi>
<mi>c</mi>
</msub>
<mo>,</mo>
<msub>
<mover>
<mi>x</mi>
<mo>·</mo>
</mover>
<mi>c</mi>
</msub>
<mo>,</mo>
<msub>
<mover>
<mi>x</mi>
<mrow>
<mo>·</mo>
<mo>·</mo>
</mrow>
</mover>
<mi>c</mi>
</msub>
<mo>]</mo>
</mrow>
<mi>T</mi>
</msup>
</mrow></math> position, velocity and acceleration along the x-axis of the robot centroid in the world coordinate system, uxIs the variation of the acceleration of the center of mass and is used to control the acceleration of the center of mass, pc＝[xc，yc，yc]TIs the three-dimensional position of the centroid in the world coordinate system, and pz＝[xz，yz，yz]TThe position of the ZMP in a world coordinate system, and g is the gravity acceleration;
12) according to the balance principle that the virtual ZMP is positioned in the sole support polygon, the controllable acceleration constraint formula of the mass center is obtained as follows:
wherein S iscFor supporting polygonal convex hulls, xi，yiIs ScCoordinates of the inner points.
The step 2) comprises the following steps:
21) establishing a single-foot supporting foot drop point compensator, and defining the output of the compensator as follows:
π＝[△tss，Δxf，Δyf]T
wherein, Δ tssFor the modification of the remaining duration of the monopod support phase, Δ p, from the time of the landing compensation triggerf＝[Δxf，Δyf，0]TThe amount of modification to the landing point for the next step,
from the output of the compensator, the modified trajectory of the reference ZMP is:
wherein γ is (t-t)ss)/tDs，pfTo determine the position of the landing foot point, tssDetermining the time from the moment to the end of the monopod support for the output of the footholdDsFor both feet support time;
22) the inputs defining the one-foot support foothold compensator are:
23) the performance indexes of the single-foot supporting foot-drop point compensator are defined as follows:
wherein,for reference to the ZMP, the method comprises the steps of,for measurement of VZMP, N is the number of steps taken, tNTime taken for N steps, Δ pf(k) And Δ tss(k) For the kth step footprint modification strategy, the corresponding coefficient is set to Qz＝1，Qp100 and Qt＝1。
The step 3) comprises the following steps:
31) establishing an heteroscedastic sparse Gaussian process model according to the constant variance sparse Gaussian process model and the sparse Gaussian process model, wherein an output prediction equation of the heteroscedastic sparse Gaussian process model is as follows:
wherein x istFor the input state, y is the standard Gaussian process model target output, y ═ y1，y2，…，yN]TK is the kernel function, k·＝[κ(xt，x1)，κ(xt，x2)，…，κ(xt，xN)]T，κ·＝κ(x1，xt) K is a square matrix, and each element is K (i, j) ═ K (x)i，xj)，Is the signal variance, W is the weight-diagonal matrix, Λ ═ diag (λ),QM＝KM+KMN(Λ+diag(r))-1KNM，for the uncertainty at each pseudo-input,a pseudo input quantity of a sparse Gaussian process model;
32) and obtaining a final prediction equation according to the unchanged variance at the input state:
wherein, <math><mrow>
<mi>H</mi>
<mo>=</mo>
<msubsup>
<mi>Q</mi>
<mi>M</mi>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<msub>
<mi>K</mi>
<mi>MN</mi>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mi>Λ</mi>
<mo>+</mo>
<mi>diag</mi>
<mrow>
<mo>(</mo>
<mi>r</mi>
<mo>)</mo>
</mrow>
<mo>)</mo>
</mrow>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mi>y</mi>
</mrow></math> and is
The step 4) comprises the following steps:
41) firstly, constructing a single pseudo sample point obtained by training into a KD tree in an off-line manner;
42) given an online instance, finding a nearest pseudo sample point according to a KD tree;
43) updating the prediction vector element corresponding to the nearest pseudo sample point;
44) and updating the heteroscedastic sparse Gaussian process model by adopting a one-time batch updating method.
The step 5) comprises the following steps:
51) the output of the different variance sparse Gaussian process model foot point compensator obtained according to the nonparametric mapping capability of the different variance sparse Gaussian process model is as follows:
π＝[min(Δtssx，△tssy)，Δxf，Δyf]T
wherein, Δ tssx＝Gα(s1) And Δ tssx＝Gα(s1) Is an independent HSpGP model of the heteroscedastic sparse Gaussian process model footpoint compensator along the x-axis, Δ tssy＝Gty(s2) And Δ yf＝Gy(s2) Is an independent HSpGP model of the heteroscedastic sparse Gaussian process model foot point compensator along the x axis,
52) training a different variance sparse Gaussian process model foot point compensator in an off-line mode according to the kinetic model foot point compensator and updating the foot point compensator of the different variance sparse Gaussian process model on line;
53) using the vector from the reference ZMP position to the measured ZMP position as an update output by the foot-drop compensator, calculating the desired foot-drop position as:
yt=Δyf+βΔyz
xt=Δxf+βΔxz
wherein x istΔ p for a given predicted input statef＝[Δxf，Δyf]TThe current output foot-drop point position modifier of the foot-drop point compensator of the heteroscedastic sparse Gaussian process model is obtained.
Compared with the prior art, the invention has the following advantages:
firstly, prediction is accurate, a nonlinear heteroscedastic sparse Gaussian process model foot point compensator is adopted, real-time online adjustment is achieved, and the method is more accurate.
Secondly, the learning speed is high, a local updating method based on the heteroscedastic Gaussian process model is adopted, the calculation complexity is reduced, and the learning speed is improved.
And thirdly, the model is convenient to train, and the overfitting problem of the parameter regression model is avoided.
And fourthly, the model stability is high, and the prediction accuracy of the whole model cannot be damaged due to local updating.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a tracking trajectory diagram of a constrained dynamics model.
FIG. 3 is a tracking trajectory diagram of an ideal linear inverted pendulum model.
FIG. 4 is a diagram of the effect of a landing point compensator based on a constrained dynamical model.
FIG. 5 is a block diagram of a humanoid robot walking control employing an HSpGP foothold compensator.
FIG. 6 is a Y-axis trajectory tracking diagram using the HSpGP foothold compensator.
FIG. 7 is a plan trace using HSpGP landing point compensators.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments.
Example (b):
as shown in fig. 1, a robot walking control method based on a foothold compensator includes the following steps:
1) establishing a constraint dynamic model of the robot;
2) designing a constraint dynamic model foot point compensator according to the constraint dynamic model;
3) establishing a different variance sparse Gaussian process regression model to realize the mapping calculation from input to output of the foothold compensator;
4) locally updating a different variance sparse Gaussian process model;
5) building a foot point compensator based on a variance sparse Gaussian process model;
6) and performing predictive control on the robot walking according to the footfall compensator based on the heteroscedastic sparse Gaussian process model.
The step 1) comprises the following steps:
11) establishing a three-dimensional linear inverted pendulum mechanical model of the robot, wherein the expression of the mechanical model is as follows:
xz＝Cxc
wherein, <math><mrow>
<msub>
<mi>x</mi>
<mi>c</mi>
</msub>
<mo>=</mo>
<msup>
<mrow>
<mo>[</mo>
<msub>
<mi>x</mi>
<mi>c</mi>
</msub>
<mo>,</mo>
<msub>
<mover>
<mi>x</mi>
<mo>·</mo>
</mover>
<mi>c</mi>
</msub>
<mo>,</mo>
<msub>
<mover>
<mi>x</mi>
<mrow>
<mo>·</mo>
<mo>·</mo>
</mrow>
</mover>
<mi>c</mi>
</msub>
<mo>]</mo>
</mrow>
<mi>T</mi>
</msup>
</mrow></math> position, velocity and acceleration along the x-axis of the robot centroid in the world coordinate system, uxIs the variation of the acceleration of the center of mass and is used to control the acceleration of the center of mass, pc＝[xc，yc，yc]TIs the three-dimensional position of the centroid in the world coordinate system, and pz＝[xz，yz，yz]TThe position of the ZMP in a world coordinate system, and g is the gravity acceleration;
12) according to the balance principle that the virtual ZMP is positioned in the sole support polygon, the controllable acceleration constraint formula of the mass center is obtained as follows:
wherein S iscFor supporting polygonal convex hulls, xi，yiIs ScCoordinates of the inner points.
As shown in figures 2 and 3, the role of the constrained dynamic model in the walking control of the humanoid robot is represented vividly, the humanoid robot is subjected to simulation comparison based on the track tracking conditions of the constrained linear inverted pendulum model and the unconstrained linear inverted pendulum model when being impacted, the impact is about 2.5s, and the micro impact is resisted successfully through pre-observation control when the constraint condition of a supporting polygon is not considered. However, in the practical case of considering support for polygon constraints, the robot will not be able to obtain sufficient acceleration of the centroid to control the centroid deviating from the desired trajectory, and therefore the robot inevitably loses balance.
The step 2) comprises the following steps:
21) establishing a single-foot supporting foot drop point compensator, and defining the output of the compensator as follows:
π＝[△tss，Δxf，△yf]T
wherein, Δ tssFor the modification of the remaining duration of the monopod support phase, Δ p, from the time of the landing compensation triggerf＝[Δxf，Δyf，0]TThe amount of modification to the landing point for the next step,
from the output of the compensator, the modified trajectory of the reference ZMP is:
wherein γ is (t-t)ss)/tDS，pfTo determine the position of the landing foot point, tssDetermining the time from the moment to the end of the monopod support for the output of the footholdDSFor both feet support time;
22) the inputs defining the one-foot support foothold compensator are:
23) the performance indexes of the single-foot supporting foot-drop point compensator are defined as follows:
wherein,for reference to the ZMP, the method comprises the steps of,for measurement of VZMP, N is the number of steps taken, tNTime taken for N steps, Δ pf(k) And Δ tss(k) For the kth step footprint modification strategy, the corresponding coefficient is set to Qz＝1，Qp100 and Qt＝1。
Such asDrawing (A)4, given the input, output definition, and performance metrics of the footprint compensator, the optimal footprint compensator output can be calculated from each footprint compensator input using a constrained nonlinear optimization Calculation (CNOP) method, since the constrained dynamics model is directly calculable. It can be seen that after the foothold compensator is used, the robot can successfully resist the disturbance and continue to stably advance.
The step 3) comprises the following steps:
31) establishing an heteroscedastic sparse Gaussian process model according to the constant variance sparse Gaussian process model and the sparse Gaussian process model, wherein an output prediction equation of the heteroscedastic sparse Gaussian process model is as follows:
wherein x istFor the input state, y is the standard Gaussian process model target output, y ═ y1，y2，…，yN]TK is the kernel function, k·＝[κ(xt，x1)，κ(xt，x2)，…，κ(xt，xN)]T，κ·＝κ(xt，xt) K is a square matrix, and each element is K (i, j) ═ K (x)i，xj)，Is the signal variance, W is the weight-diagonal matrix, Λ ═ diag (λ),QM＝KM+KMN(Λ+diag(r))-1KNM，for the uncertainty at each pseudo-input,a pseudo input quantity of a sparse Gaussian process model;
32) and obtaining a final prediction equation according to the unchanged variance at the input state:
wherein, <math><mrow>
<mi>H</mi>
<mo>=</mo>
<msubsup>
<mi>Q</mi>
<mi>M</mi>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<msub>
<mi>K</mi>
<mi>MN</mi>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mi>Λ</mi>
<mo>+</mo>
<mi>diag</mi>
<mrow>
<mo>(</mo>
<mi>r</mi>
<mo>)</mo>
</mrow>
<mo>)</mo>
</mrow>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mi>y</mi>
</mrow></math> and is
The step 4) comprises the following steps:
41) firstly, constructing a single pseudo sample point obtained by training into a KD tree in an off-line manner;
42) given an online instance, finding a nearest pseudo sample point according to a KD tree;
43) updating the prediction vector element corresponding to the nearest pseudo sample point;
44) and updating the heteroscedastic sparse Gaussian process model by adopting a one-time batch updating method.
Step 5) comprises the following steps:
51) the output of the different variance sparse Gaussian process model foot point compensator obtained according to the nonparametric mapping capability of the different variance sparse Gaussian process model is as follows:
π＝[min(Δtssx，Δtssy)，Δxf，Δyf]T
wherein, Δ tssx＝Gα(s1) And Δ tssx＝Gα(s1) Is an independent HSpGP model of the heteroscedastic sparse Gaussian process model footpoint compensator along the x-axis, Δ tssy＝Gty(s2) And Δ yf＝Gy(s2) Is an independent HSpGP model of the heteroscedastic sparse Gaussian process model foot point compensator along the x axis,
52) training a different variance sparse Gaussian process model foot point compensator in an off-line mode according to the kinetic model foot point compensator and updating the foot point compensator of the different variance sparse Gaussian process model on line;
53) using the vector from the reference ZMP position to the measured ZMP position as an update output by the foot-drop compensator, calculating the desired foot-drop position as:
yt＝Δyf+βΔyz
wherein x isfΔ p for a given predicted input statef＝[Δxf，Δyf]TThe current output foot-drop point position modifier of the foot-drop point compensator of the heteroscedastic sparse Gaussian process model is obtained.
The human-simulated robot walking control framework adopting the HSpGP model as the foot-landing point compensator model is shown in fig. 5, the subsequent foot-landing position is planned according to the walking instruction sent by the robot decision layer by the gait planning, the ZMP track is planned according to the foot-landing position, and the expected centroid track is calculated by using the pre-observation controller. And the walking of the robot is realized through a reverse motion model and a PD joint controller. After the robot finishes the current motion instruction, the current state of the mass center is estimated through airborne inertial unit (IMU) measurement information and joint feedback values. The drop point compensator re-modifies subsequent drop points based on the centroid state and the drop point modifier predicted by HspGP and the single-foot support phase time.
To illustrate the role of the updated landing point compensation controller in disturbance rejection, the robot is here made to use the final updated HSpGP model in the simulation software to control the landing point compensation, as shown in FIGS. 6 and 7. In the process of robot walking, a small ball with the mass of 0.4kg is adopted to impact the shoulder of the robot at the speed of 4 m/s. Because gravity does work on the small balls in the flying process of the small balls, the actual impulse generated by the small balls acting on the robot is more than 2 Ns. The impact occurs at around 2.5s and the robot experiences a large displacement of the ZMP after impact. The robot takes a large step to the left under the action of the foothold compensator and most impact disturbances are suppressed. The robot then performs a second step of right foot adjustment to accomplish the suppression of the amount of residual disturbance. From the trajectory diagram in the x-y plane, it can be seen that after the robot has suffered an impact, its ZMP has moved far out of the robot's sole support polygon, if nothing is done the robot will lose balance and fall.
Claims (6)
1. A robot walking control method based on a foothold compensator is characterized by comprising the following steps:
1) establishing a constraint dynamic model of the robot;
2) designing a constraint dynamic model foot point compensator according to the constraint dynamic model;
3) establishing a different variance sparse Gaussian process regression model, and realizing the mapping calculation from input to output of the constraint dynamic model foot point compensator;
4) locally updating a variance sparse Gaussian process regression model;
5) establishing a different variance sparse Gaussian process regression model-based footpoint compensator;
6) and performing predictive control on the robot walking according to the different-variance-based sparse Gaussian process regression model foot-drop point compensator.
2. The robot walking control method based on the foothold compensator according to claim 1, wherein the step 1) specifically comprises the following steps:
11) establishing a three-dimensional linear inverted pendulum mechanical model of the robot, wherein the expression of the mechanical model is as follows:
xz＝Cxc
wherein, position, velocity and acceleration along the x-axis of the robot centroid in the world coordinate system, uxIs the variation of the acceleration of the center of mass and is used to control the acceleration of the center of mass, pc＝[xc，yc，yc]TIs the three-dimensional position of the centroid in the world coordinate system, and pz＝[xz，yz，yz]TThe position of the ZMP in a world coordinate system, and g is the gravity acceleration;
12) according to the balance principle that the virtual ZMP is positioned in the sole support polygon, the controllable acceleration constraint formula of the mass center is obtained as follows:
wherein S iscFor supporting polygonal convex hulls, xi，yiIs ScCoordinates of the inner points.
3. The robot walking control method based on the foothold compensator according to claim 1, wherein the step 2) specifically comprises the following steps:
21) establishing a constraint dynamic model foot point compensator, and defining the output of the compensator as follows:
π＝[Δtss，Δxf，Δyf]T
wherein, Δ tssFor the modification of the remaining duration of the monopod support phase, Δ p, from the time of the landing compensation triggerf＝[Δxf，Δyf，0]TThe amount of modification to the landing point for the next step,
from the output of the compensator, the modified trajectory of the reference ZMP is:
wherein γ is (t-t)SS)/tDS，pfTo determine the position of the landing foot point, tSSDetermining the time from the moment to the end of the monopod support for the output of the footholdDSFor both feet support time;
22) the inputs defining the constraint dynamical model footpoint compensator are:
23) defining the performance indexes of the constraint dynamic model foot point compensator as follows:
wherein,for reference to the ZMP, the method comprises the steps of,for measurement of VZMP, N is the number of steps taken, tNTime taken for N steps, Δ pf(k) And Δ tss(k) For the kth step footprint modification strategy, the corresponding coefficient is set to Qz＝1，Qp100 and Qt＝1。
4. The robot walking control method based on the foothold compensator of claim 1, wherein the step 3) specifically comprises the following steps:
31) establishing an heteroscedastic sparse Gaussian process model according to the constant variance sparse Gaussian process model and the sparse Gaussian process model, wherein an output prediction equation of the heteroscedastic sparse Gaussian process model is as follows:
wherein x istFor the input state, y is the standard Gaussian process model target output, y ═ y1，y2，…，yN]TK is the kernel function, k*＝[κ(xt，x1)，κ(xt，x2)，…，κ(xt，xN)]T，κ*＝κ(xt，xt) K is a square matrix, and each element is K (i, j) ═ K (x)i，xj)，Is the signal variance, W is the weight-diagonal matrix, Λ ═ diag (λ),QM＝KM+KMN(Λ+diag(r))-1KNM，for the uncertainty at each pseudo-input,a pseudo input quantity of a sparse Gaussian process model;
32) and obtaining a final prediction equation according to the unchanged variance at the input state:
wherein, <math>
<mrow>
<mi>H</mi>
<mo>=</mo>
<msubsup>
<mi>Q</mi>
<mi>M</mi>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msubsup>
<msub>
<mi>K</mi>
<mi>MN</mi>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mi>Λ</mi>
<mo>+</mo>
<mi>diag</mi>
<mrow>
<mo>(</mo>
<mi>r</mi>
<mo>)</mo>
</mrow>
<mo>)</mo>
</mrow>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mi>y</mi>
</mrow>
</math> and is
5. The robot walking control method based on the foothold compensator according to claim 1, wherein the step 4) specifically comprises the following steps:
41) firstly, constructing a single pseudo sample point obtained by training into a KD tree in an off-line manner;
42) given an online instance, finding a nearest pseudo sample point according to a KD tree;
43) updating the prediction vector element corresponding to the nearest pseudo sample point;
44) and updating the heteroscedastic sparse Gaussian process model by adopting a one-time batch updating method.
6. The robot walking control method based on the foothold compensator according to claim 1, wherein the step 5) specifically comprises the following steps:
51) the output of the different variance sparse Gaussian process model foot point compensator obtained according to the nonparametric mapping capability of the different variance sparse Gaussian process model is as follows:
π＝[min(Δtssx，Δtssy)，Δxf，Δyf]T
wherein, Δ tssx＝Gts(st) And Δ tssx＝Gtx(s1) Is a heteroscedastic sparse Gaussian processIndependent HSpGP model, Δ t, of model footpoint compensator along the x-axisssy＝Gty(s2) And Δ yf＝Gy(s2) Is an independent HSpGP model of the heteroscedastic sparse Gaussian process model foot point compensator along the x axis,
52) training a different variance sparse Gaussian process model foot point compensator in an off-line mode according to the kinetic model foot point compensator and updating the foot point compensator of the different variance sparse Gaussian process model on line;
53) using the vector from the reference ZMP position to the measured ZMP position as an update output by the foot-drop compensator, calculating the desired foot-drop position as:
yt＝Δyf+βΔyz
xt＝Δxf+βΔxz
wherein x istΔ p for a given predicted input statef＝[Δxf，Δyf]TThe current output foot-drop point position modifier of the foot-drop point compensator of the heteroscedastic sparse Gaussian process model is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410521894.9A CN104345735A (en) | 2014-09-30 | 2014-09-30 | Robot walking control method based on foothold compensator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410521894.9A CN104345735A (en) | 2014-09-30 | 2014-09-30 | Robot walking control method based on foothold compensator |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104345735A true CN104345735A (en) | 2015-02-11 |
Family
ID=52501581
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410521894.9A Pending CN104345735A (en) | 2014-09-30 | 2014-09-30 | Robot walking control method based on foothold compensator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104345735A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106564055A (en) * | 2016-10-31 | 2017-04-19 | 金阳娃 | Stable motion planning method of simulation humanoid robot and control device thereof |
| CN106597843A (en) * | 2015-10-20 | 2017-04-26 | 沈阳新松机器人自动化股份有限公司 | Front-wheel driving robot safety control method and front-wheel driving robot safety control system |
| CN109352655A (en) * | 2018-11-28 | 2019-02-19 | 清华大学 | A kind of deformation-compensated method of robot returned based on multi output Gaussian process |
| CN109445444A (en) * | 2018-12-25 | 2019-03-08 | 同济大学 | A kind of barrier concentrates the robot path generation method under environment |
| CN111144644A (en) * | 2019-12-24 | 2020-05-12 | 淮阴工学院 | Short-term wind speed prediction method based on variation variance Gaussian process regression |
| CN112318509A (en) * | 2020-10-30 | 2021-02-05 | 东南大学 | Trajectory tracking control method for Gaussian process of space robot |
| CN112975941A (en) * | 2019-12-13 | 2021-06-18 | 深圳市优必选科技股份有限公司 | Robot control method, device, computer readable storage medium and robot |
| CN114253260A (en) * | 2021-12-08 | 2022-03-29 | 深圳市优必选科技股份有限公司 | Robot gait planning method and device, motion planning equipment and storage medium |
| CN114723807A (en) * | 2021-01-04 | 2022-07-08 | 腾讯科技（深圳）有限公司 | Centroid trajectory determination method and device, foot type robot, equipment and medium |
| CN116819973A (en) * | 2023-08-29 | 2023-09-29 | 北京成功领行汽车技术有限责任公司 | Track tracking control method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030229419A1 (en) * | 1999-11-25 | 2003-12-11 | Sony Corporation | Legged mobile robot and method and apparatus for controlling the operation thereof |
| JP2004314250A (en) * | 2003-04-17 | 2004-11-11 | Honda Motor Co Ltd | Floor reaction working point estimating method of bipedal walk moving body and joint moment estimating method of bipedal walk moving body |
| US20070220637A1 (en) * | 2006-02-09 | 2007-09-20 | Gen Endo | Robot apparatus and method of controlling the same |
| CN101408435A (en) * | 2008-10-31 | 2009-04-15 | 北京理工大学 | Method and apparatus for movement planning of apery robot ankle |
| CN103116354A (en) * | 2013-01-30 | 2013-05-22 | 同济大学 | Method for generating real-time gait path of biped robot |
-
2014
- 2014-09-30 CN CN201410521894.9A patent/CN104345735A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030229419A1 (en) * | 1999-11-25 | 2003-12-11 | Sony Corporation | Legged mobile robot and method and apparatus for controlling the operation thereof |
| JP2004314250A (en) * | 2003-04-17 | 2004-11-11 | Honda Motor Co Ltd | Floor reaction working point estimating method of bipedal walk moving body and joint moment estimating method of bipedal walk moving body |
| US20070220637A1 (en) * | 2006-02-09 | 2007-09-20 | Gen Endo | Robot apparatus and method of controlling the same |
| CN101408435A (en) * | 2008-10-31 | 2009-04-15 | 北京理工大学 | Method and apparatus for movement planning of apery robot ankle |
| CN103116354A (en) * | 2013-01-30 | 2013-05-22 | 同济大学 | Method for generating real-time gait path of biped robot |
Non-Patent Citations (2)
| Title |
|---|
| ANDREW RUSHTON: "《用于逻辑综合的VHDL 原书第3版》", 31 January 2014, 北京航空航天大学出版社 * |
| XU TAO 等: "Rebalance Strategies for Humanoids Walking by Foot Positioning Compensator Based on Adaptive Heteroscedastic SpGPs", 《2011 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106597843A (en) * | 2015-10-20 | 2017-04-26 | 沈阳新松机器人自动化股份有限公司 | Front-wheel driving robot safety control method and front-wheel driving robot safety control system |
| CN106597843B (en) * | 2015-10-20 | 2019-08-09 | 沈阳新松机器人自动化股份有限公司 | A kind of front driving wheel formula robot security control method and system |
| CN106564055B (en) * | 2016-10-31 | 2019-08-27 | 金阳娃 | Human simulation robot stabilization motion planning method and control device |
| CN106564055A (en) * | 2016-10-31 | 2017-04-19 | 金阳娃 | Stable motion planning method of simulation humanoid robot and control device thereof |
| CN109352655A (en) * | 2018-11-28 | 2019-02-19 | 清华大学 | A kind of deformation-compensated method of robot returned based on multi output Gaussian process |
| CN109445444B (en) * | 2018-12-25 | 2021-05-11 | 同济大学 | Robot path generation method under barrier concentration environment |
| CN109445444A (en) * | 2018-12-25 | 2019-03-08 | 同济大学 | A kind of barrier concentrates the robot path generation method under environment |
| CN112975941A (en) * | 2019-12-13 | 2021-06-18 | 深圳市优必选科技股份有限公司 | Robot control method, device, computer readable storage medium and robot |
| CN111144644A (en) * | 2019-12-24 | 2020-05-12 | 淮阴工学院 | Short-term wind speed prediction method based on variation variance Gaussian process regression |
| CN112318509A (en) * | 2020-10-30 | 2021-02-05 | 东南大学 | Trajectory tracking control method for Gaussian process of space robot |
| CN112318509B (en) * | 2020-10-30 | 2022-04-29 | 东南大学 | Trajectory tracking control method for Gaussian process of space robot |
| CN114723807A (en) * | 2021-01-04 | 2022-07-08 | 腾讯科技（深圳）有限公司 | Centroid trajectory determination method and device, foot type robot, equipment and medium |
| CN114253260A (en) * | 2021-12-08 | 2022-03-29 | 深圳市优必选科技股份有限公司 | Robot gait planning method and device, motion planning equipment and storage medium |
| CN114253260B (en) * | 2021-12-08 | 2023-08-18 | 深圳市优必选科技股份有限公司 | Robot gait planning method and device, motion planning equipment and storage medium |
| CN116819973A (en) * | 2023-08-29 | 2023-09-29 | 北京成功领行汽车技术有限责任公司 | Track tracking control method |
| CN116819973B (en) * | 2023-08-29 | 2023-12-12 | 北京成功领行汽车技术有限责任公司 | Track tracking control method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104345735A (en) | Robot walking control method based on foothold compensator | |
| CN106586033B (en) | The multistage linear puppet of adaptive segmentation composes broad sense mark control miss distance reentry guidance method | |
| Lu et al. | Predictive approach for sensorless bimanual teleoperation under random time delays with adaptive fuzzy control | |
| CN103116275B (en) | Based on the gyroscope Robust Neural Network Control system and method that sliding formwork compensates | |
| Kajita et al. | Biped walking | |
| CN103303495B (en) | Method for estimating disturbance moment in power decreasing process | |
| Yang et al. | Impact invariant control with applications to bipedal locomotion | |
| Morlando et al. | Whole-body control with disturbance rejection through a momentum-based observer for quadruped robots☆ | |
| Ramos et al. | Generalizations of the capture point to nonlinear center of mass paths and uneven terrain | |
| JP6781101B2 (en) | Non-linear system control method, biped robot control device, biped robot control method and its program | |
| Jeong et al. | Biped walking stabilization based on foot placement control using capture point feedback | |
| Wittmann et al. | State estimation for biped robots using multibody dynamics | |
| CN116449715A (en) | Tracking control method for surface bouncing movement track of complex small celestial body | |
| Wittmann et al. | Real-time nonlinear model predictive footstep optimization for biped robots | |
| Wang et al. | A new spacecraft attitude stabilization mechanism using deep reinforcement learning method | |
| CN108415443B (en) | Control method for forced flight-around of non-cooperative target | |
| CN104318071A (en) | Robot walking control method based on linear foothold compensator | |
| Wang et al. | Fast Terrain-Adaptive Motion of Humanoid Robots Based on Model Reference One-Step-Ahead Predictive Control | |
| Roscia et al. | Reactive landing controller for quadruped robots | |
| Cisneros et al. | Partial yaw moment compensation using an optimization-based multi-objective motion solver | |
| Orozco-Soto et al. | Motion control of humanoid robots using sliding mode observer-based active disturbance rejection control | |
| CN103197558A (en) | Microgyroscope fuzzy self-adaptation control method based on T-S model | |
| Lowrey et al. | Real-time state estimation with whole-body multi-contact dynamics: A modified UKF approach | |
| Li et al. | Neural adaptive tracking control for wheeled mobile robots | |
| JP7221833B2 (en) | Nonlinear model predictive controller |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20150211 |