US 12,741,710 B2 · Whole-body compliance control method applied to fast walking biped robot

Abstract

The present invention provides a whole-body compliance control method applied to a fast and stable walking biped robot, and the control method includes: obtaining a plantar stress and a plantar moment based on information of a foot force sensor to estimate an actual ZMP position of the robot; designing a centroid compliance controller based on the actual ZMP position of the robot to correct a centroid acceleration; designing a plantar position compliance controller based on a double spring-damper model according to the plantar stress to correct a foot height; designing a plantar posture compliance controller based on the plantar moment to correct a plantar posture; adding corrected centroid and foot values into an originally desired trajectory; and obtaining motor angles of various joints. In the method, the plantar position compliance controller is used to reduce the landing impact; the plantar posture compliance controller is used to ensure that the robot lands flat to prevent the robot from rollover; the centroid compliance controller is used to adjust a centroid position to prevent a robot body from bending forward and backward when the robot walks fast, thereby increasing the stability of the biped robot during fast walking.

A whole-body compliance control method applied to a fast and stable walking biped robot, comprising the following steps:

Step 1:
acquiring plantar stress data according to information of a foot force sensor, wherein the plantar stress data includes a plantar stress and a plantar moment;
Step 2:
calculating an actual zero moment point (ZMP) position of the robot according to the plantar stress data;
Step 3:
using a centroid compliance controller to calculate a centroid acceleration adjustment variable according to a difference between the actual ZMP position and a desired ZMP position;
and integrating the centroid acceleration adjustment variable to obtain a centroid position adjustment variable;
Step 4:
using a double spring-damper model to establish a state equation of a plantar force and a plantar position according to the plantar stress data;
and adjusting spring and damper coefficients to calculate a plantar position adjustment variable of the robot in different landing periods in a vertical direction;
wherein the double spring-damper model comprises two spring-damper units, the rigidity coefficients and damping coefficients of the two spring-damper units are selected according to different characteristics of the foot sole of the robot in different periods;
when the foot sole just lands, relationships between the rigidity coefficients and the damping coefficients of the two spring-damper units need to be opposite, that is, one of the spring-damper units has a large rigidity coefficient and a small damping coefficient, and the other thereof has a small rigidity coefficient and a large damping coefficient;
during foot support, the two spring-damper units have same effects and same relationships between the rigidity coefficients and the damping coefficients;
Step 5:
using a plantar posture compliance controller to calculate a plantar posture adjustment variable according to a difference between the plantar moment and a desired plantar moment, wherein the plantar posture includes a rolling direction posture and a pitching direction posture;
and the plantar posture adjustment variable is calculated by the plantar posture compliance controller according to the following formula:
k
d
⁢
δ
˙
+
k
p
⁢
δ
=
τ
d
-
τ
,
wherein τ and τd respectively denote the plantar moment and the desired plantar moment;
δ denotes the plantar posture to be adjusted, wherein the plantar posture comprises the rolling direction posture and the pitching direction posture;
and kP and kd denote gain parameters to be adjusted;
Step 6:
respectively superimposing the adjustment variables of the centroid position, plantar position and plantar posture to the originally desired centroid position, plantar position and plantar posture;
and obtaining angle information of various joints adjusted through an inverse kinematics solution to drive the robot to walk forward;
wherein sensor information is collected by a moment sensor array or a force sensor array that is uniformly distributed on a foot sole;
wherein in the Step 2, according to the plantar stress data, the actual ZMP positions of the robot in x and y directions are calculated through the following formula:
p
x
=
∑
j
=
1
N
{
-
τ
jx
-
(
p
jz
-
p
z
)
⁢
f
jy
+
p
jy
⁢
f
jz
}
∑
j
=
1
N
f
jz
p
y
=
∑
j
=
1
N
{
τ
jy
-
(
p
jz
-
p
z
)
⁢
f
jx
+
p
jx
⁢
f
jz
}
∑
j
=
1
N
f
jz
,
wherein fji and τji respectively denote a measured force and moment of a j (th) sensor in the i direction, pji denotes a position coordinate of the sensor relative to an origin of a coordinate system, j=1, . . . , N,i=x,y, and the origin of a coordinate system is set as a projective spot of an ankle located on the foot sole;
wherein in the Step 3, the biped robot is simplified as a table-cart model to design the centroid compliance controller, wherein a table height is corresponding to a centroid height relative to the ground;
a cart acceleration is a robot centroid acceleration, and the centroid compliance controller is used to calculate the centroid position adjustment variable;
and such calculation comprises:
firstly, calculating the centroid acceleration adjustment variable:
Δ
⁢
x
¨
=
-
k
1
⁢
Δ
⁢
p
-
k
2
⁢
Δ
⁢
x
-
k
3
⁢
Δ
⁢
x
.
,
wherein Δp denotes the difference between the desired ZMP position of the robot and the actual ZMP position thereof;
Δx, Δ{dot over (x)} and Δ{umlaut over (x)} respectively denote cart position, velocity and acceleration adjustment variables;
k1, k2 and k3 are determined by coefficient matrices
[
-
1
/
T
1
/
T
0
0
0
1
0
0
0
]
and
[
-
z
c
g
⁢
T
0
1
]
and weight matrices Q and R of a state variable and a control variable, T denotes a sampling period of a sensor, zc denotes a table height to ground, and g denotes a gravity acceleration;
secondly, performing twice numerical integration on Δ{umlaut over (x)} to obtain the centroid position adjustment variable;
wherein a linear quadratic regulator is used to calculate coefficients k1, k2 and k3;
wherein in the Step 4, the double spring-damper model is expressed as:
(
k
s
⁢
1
+
k
s
⁢
2
)
⁢
f
𝓏
+
(
d
1
+
d
2
)
⁢
f
˙
𝓏
=
k
s
⁢
1
⁢
k
s
⁢
2
⁢
𝓏
a
+
(
k
s
⁢
1
⁢
d
2
+
k
s
⁢
2
⁢
d
1
)
⁢
𝓏
.
a
wherein fz denotes an ankle stress in the z axis, za denotes an ankle position in the z direction, and ks1, ks2 and d1, d2 respectively denote rigidity coefficients and damping coefficients of two spring-damper units;
the double spring-damper model is transformed into the form of the state equation, and the linear quadratic regulator is used to calculate the acceleration adjustment variable of the foot sole in a vertical direction:
Δ
⁢
𝓏
¨
a
=
-
k
𝓏
⁢
1
⁢
Δ
⁢
f
𝓏
-
k
𝓏
⁢
2
⁢
Δ
⁢
𝓏
a
-
k
𝓏3
⁢
Δ
⁢
𝓏
˙
a
,
wherein Δfz denotes a difference between a desired stress value of the ankle and an actual stress value thereof along the z axis, Δza, Δża and Δ{umlaut over (z)}a respectively denote the position, velocity and acceleration adjustment variables of the ankle in the z direction, kz1, kz2 and kz3 are determined by coefficient matrices
[
-
k
s
⁢
1
+
k
s
⁢
2
d
1
+
d
2
k
s
⁢
1
⁢
k
s
⁢
2
d
1
+
d
2
k
s
⁢
1
⁢
d
2
+
k
s
⁢
2
⁢
d
1
d
1
+
d
2
0
0
1
0
0
0
]
and
[
d
1
⁢
d
2
d
1
+
d
2
0
1
]
and weight matrices Q and R of a state variable and a control variable, and ks1, ks2 and d1, d2 respectively denote rigidity coefficients and damping coefficients of the two spring-damper units;
and
wherein a linear quadratic regulator is used to calculate coefficients kz1, kz2 and kz3′.

353 more elements

Whole-body compliance control method applied to fast walking biped robot

Life of this patent

Whole-body compliance control method applied to fast walking biped robot

Life of this patent

Abstract

Claims2