WO2019218805A1

WO2019218805A1 - Motion closed-loop control method for quadruped robot

Info

Publication number: WO2019218805A1
Application number: PCT/CN2019/081969
Authority: WO
Inventors: 刘厚德; 王翔; 王学谦; 梁斌; 高学海; 朱晓俊
Original assignee: 清华大学深圳研究生院
Priority date: 2018-05-16
Filing date: 2019-04-09
Publication date: 2019-11-21
Also published as: CN108572553A; CN108572553B

Abstract

A motion closed-loop control method for a quadruped robot, comprising the following steps: using a CPG network to generate a periodic control signal for each leg portion of a quadruped robot respectively so as to generate an advancing signal for the quadruped robot; and simultaneously using proportional differential control to adjust the pose of a machine body of the quadruped robot. The described motion closed-loop control method for a quadruped robot is concise, highly efficient, and easy to implement, and overcomes the defect in existing algorithms wherein control difficulty is high.

Description

Motion closed-loop control method for quadruped robot

Technical field

The invention relates to the field of robot control, and in particular to a motion closed loop control method for a quadruped robot.

Background technique

Robotics is an interdisciplinary subject involving knowledge in the fields of mechanics, electronics, bionics, computer science, and artificial intelligence. Its flexible mobility can replace human beings under harsh working conditions (high temperature, toxic, underwater, etc.). Complex tasks, and thus have important social value for the study of robotics.

Currently the most popular mobile robots are divided into wheeled, crawler and foot robots. Wheeled and crawler robots are able to walk stably on flat roads, and their off-road capabilities have also made significant breakthroughs. However, their walking on rough terrain still has great limitations. In contrast, the foot robot can still achieve fast and stable travel under certain load conditions. These features are increasingly favored by researchers.

According to the number of legs, the foot robot can be divided into a single-legged jumping robot, a biped robot, a four-legged robot, and six enough robots. Compared with single-legged and bipedal robots, quadruped robots are more stable and have a higher load capacity. At the same time, compared with multi-foot (six-foot and above) robots, the mechanical structure is simpler and the control difficulty is lower. In addition, from the perspective of bionics, mammals, as the most advanced vertebrate animals, use the four-legged walking method. Therefore, constructing a four-legged bionic robot is more conducive to learning the gait pattern when the quadruped is walking.

At present, most of the quadruped robots still stay in the indoor test or even the simulation stage, and can only walk on a flat ground, the terrain adaptability is poor, and the anti-interference ability is weak. Although the quadruped robot has good motion capability and wide application prospects, its structural design and control algorithm are also relatively complex, especially the coordinated control of four-foot gait timing and attitude control during travel, plus high precision and high real-time. Sexual sensing feedback makes the control system more complex. In general, the related technology of the quadruped robot is still in the development stage, and the corresponding theoretical research still needs further improvement and development.

Among them, the concept of CPG (Central Pattern Generator) comes from the field of biology and bionics, which refers to the nerve reflex current that produces rhythmic motion in animals. It is a distributed neural network, which is essentially the occurrence of periodic signals. Device. Since the joint angle between the four legs of the quadruped robot during normal walking is essentially only a phase difference, many researchers in the prior art use it to control the walking of the quadruped robot. The use of CPG open-loop control can not effectively control the attitude of the robot, so the anti-disturbance ability is poor; the closed-loop CPG network with feedback signal is too complicated, the parameter setting is very cumbersome, can only rely on engineering experience, and the upper control command needs to go through more The link can be converted into the joint control signal, so the execution efficiency is low; when the characteristic parameters of the robot itself and the external environment change, the parameter setting needs to be re-established; these shortcomings greatly limit the further expansion of the CPG application.

Virtual Model Control (VMC) is actually an intuitive control method. The so-called intuitive control method is a kind of control method that people establish based on their own feelings and the existing knowledge of the system. This type of method does not have a fixed model formula and derivation process, but simply based on the difference between the expected state of the robot and the real-time state, the controller gives something like "fast (slow) point, "high (low) point", and Instructions such as "large (small) points" guide the robot to make corresponding behavior adjustments. The basic idea is to compare the desired motion state of the upper layer of the robot and compare it with the real-time state measured by the sensor. According to the difference between the two, the required force or moment is calculated by a specific intuition law, and then applied to the joint. Thereby the robot is moved towards the desired state. The core idea of VMC is to connect different points of action of the robot with imaginary virtual components (spring, damping, etc.) or to connect the point of action with the external environment to generate corresponding virtual forces to drive the robot to perform the desired motion. The virtual force is mapped to the joint torque by the robot Jacobian matrix, which drives the robot motion to produce the same effect as the virtual component. A schematic diagram of the 3D model of the virtual model control of the quadruped robot is shown in Fig. 1. As can be seen from the figure, the virtual spring damper supports the robot to maintain a certain posture when supporting the phase; when the phase is oscillated, the virtual spring damper pulls the leg along the planned trajectory. The control method requires a large number of sensing feedback signals, including the displacement, velocity and attitude angle of the robot body and the leg, and the angular velocity of the attitude. In practice, the application is very cumbersome and costly, and the GPS and inertial measurement unit (IMU) need to be configured. , gyro and force sensors and other components.

The above disclosure of the background art is only for assisting in understanding the concepts and technical solutions of the present invention, and it does not necessarily belong to the prior art of the present patent application, and there is no clear evidence that the above content has been disclosed on the filing date of the present patent application. In this case, the above background art should not be used to evaluate the novelty and inventiveness of the present application.

Summary of the invention

In order to solve the above technical problem, the present invention proposes a motion closed-loop control method for a quadruped robot, which is simple, efficient and easy to implement, and overcomes the defects of high control complexity of the existing algorithm.

In order to achieve the above object, the present invention adopts the following technical solutions:

The invention discloses a motion closed-loop control method for a quadruped robot, comprising the steps of: generating a periodic control signal for each leg of the quadruped robot by using a CPG network to generate a forward of the quadruped robot The signal is simultaneously adjusted by the proportional differential control to the posture of the body of the quadruped robot.

Preferably, each of the legs of the quadruped robot is provided with three joints: a hip yaw joint, a hip forward joint and a knee forward joint, wherein the CPG network is used for the four Controlling the hip forward joint and the knee forward joint of each of the legs of the foot robot; the proportional differential control is for controlling the joints of the legs of the quadruped robot in the support phase .

Preferably, each of the legs of the quadruped robot is controlled by the following formula:

q=q _a +q _c

Wherein q represents a total signal applied to the quadruped robot, q _a represents an attitude compensation angle for controlling each joint of the quadruped robot in the leg of the supporting phase, q _c is a control of the quadruped robot Control signals for the hip forward joint and the knee forward joint of each of the legs.

Preferably, the CPG network therein comprises CPG units respectively correspondingly controlling the respective legs of the quadruped robot, and the expression of the model adopted by the CPG unit is:

Δ _ji =y _j cosθ _ji -x _j sinθ _ji

Where i and j represent the leg numbers of the quadruped robot,

ω _i represents the motion frequency of the i-th leg, α and β are convergence factors, μ is a parameter, and x _i is the control signal of the i-th CPG unit to the hip forward joint of the leg, y _i is a control signal of the i-th CPG unit to the knee forward joint of the leg,

with

Representing the differentiation of the signals corresponding to x _i and y _i respectively, θ _ji represents the phase difference between different CPG units, k Σ _j Δ _ji is a coupling polynomial, and ω _stance and ω _swing represent the frequency of the supporting phase and the frequency of the oscillating phase, respectively, b Represents a constant factor; where q _c = (x ₁ , y ₁ , x ₂ , y ₂ , x ₃ , y ₃ , x ₄ , y ₄ ) ^T .

Preferably, wherein the y _i signal is filtered by:

Where c is a constant factor and θ _ki is the control signal of the filtered i-th CPG unit to the knee forward joint of the leg; where q _c = (x ₁ , θ _k1 , x ₂ , θ _K2 , x ₃ , θ _k3 , x ₄ , θ _k4 ) ^T .

Preferably, the control law for adjusting the posture of the body of the quadruped robot by using proportional differential control is:

Where K=diag(k ₁ , k ₂ , k ₃ ) and B=diag(b ₁ , b ₂ , b ₃ ), where k _l and b _l are the first of the body of the quadruped robot, respectively The proportion of the pose and the differential constant, where l = 1, 2, 3;

with

An angular vector and an angular rate vector respectively representing the actual posture of the body of the quadruped robot,

with

An angular vector and an angular rate vector representing a desired posture of the body of the quadruped robot, respectively.

Preferably, the motion closed-loop control method of the quadruped robot further includes: when the posture of the body of the quadruped robot is consistent with a desired posture, the PD controller does not function; when the quadruped robot When the posture of the body is deviated, the PD controller generates an error compensation signal according to the control law of the proportional differential control, so that the body of the quadruped robot stably travels in a desired posture.

Preferably, wherein the PD controller receives the foot touch signal from the force sensor to determine the ground contact state of the bottom end of each of the legs.

Compared with the prior art, the present invention has the beneficial effects that the motion closed-loop control method of the quadruped robot proposed by the present invention generates a periodic signal according to the bionics principle to control the periodic motion of the quadruped robot compared to other dynamics. The method is more efficient and easy to understand; and the proportional differential control law is used to adjust the posture of the quadruped robot body so that it can stably travel in a suitable attitude. Compared with other methods such as virtual model control, it is no longer necessary to calculate one leg in real time. Jacques matrix and matrix inversion, eliminating the complicated and cumbersome derivation and solution, simple and efficient.

DRAWINGS

1 is a schematic diagram of a 3D model of a virtual model control of a quadruped robot in the prior art;

2 is a schematic diagram of a model of a quadruped robot in accordance with a preferred embodiment of the present invention;

3 is a block diagram of a control strategy of a quadruped robot in accordance with a preferred embodiment of the present invention.

Detailed ways

The invention will now be further described with reference to the drawings in conjunction with the preferred embodiments.

As shown in FIG. 2, the quadruped robot of the preferred embodiment of the present invention includes a body 10 and four legs 20, each leg 20 having three joints including two joints of the hip 21 (one hip side) The pendulum joint and the 1 hip front joint) and the 1 joint of the knee 22 (the knee forward joint), that is, the quadruped robot, include a total of 12 joints.

The four-legged robot travels in two stages: the supporting phase and the oscillating phase, wherein the supporting phase means that the legs of the legs are in contact with the ground, providing support for the remaining legs to move forward, and the legs that move forward are called In the swing phase; the free switching between the two states is achieved by determining whether the leg touches the ground during the movement of the foot robot.

Referring to FIG. 3, a cyclic control signal is generated for each leg portion 20 of the quadruped robot by using a CPG network to generate a forward signal of the quadruped robot for controlling the body 10 to travel forward at a certain speed while using proportional differentiation. Control effectively corrects the posture of the body 10 using the legs 20 in the support phase. When the posture of the quadruped robot body 10 is consistent with the desired posture, the PD controller does not function, and this is a simple open loop control; when the posture of the quadruped robot body 10 is deviated, the PD controller will according to the ratio. The control law of the differential control generates an error compensation signal so that the body 10 of the quadruped robot can stably travel in a desired posture. Wherein the PD controller receives the foot touch signal from the force sensor to determine the ground contact state of the bottom end 23 of each leg portion 20, so as to switch between the support phase and the swing phase, that is, the PD controller is only The posture of the leg portion 20 in the support phase is adjusted.

In the motion control of the quadruped robot, the CPG model is mainly used to generate a stable periodic oscillation signal. In this embodiment, the CPG network is used for the hip forward joint and the knee front of each leg of the quadruped robot. Control the joints. The CPG network includes CPG units respectively correspondingly controlling the legs of the quadruped robot. In this embodiment, the CPG unit uses a Hopf oscillator as a signal generator, and the computational complexity is low and the number of parameters is small. The model is as follows:

Δ _ji =y _j cosθ _ji -x _j sinθ _ji

Where i and j represent the leg numbers of the quadruped robot, i=1, 2, 3, 4, j=1, 2, 3, 4,

ω _i represents the motion frequency of the i-th leg, α and β are convergence factors, μ is a parameter used to adjust the amplitude of the oscillator output signal, indicating the joint angle of the leg of the quadruped robot, x _i is the i The control signal of the CPG unit to the hip forward joint of the leg, y _i is the control signal of the i-th CPG unit to the knee forward joint of the leg,

with

Representing the differentiation of the signals corresponding to x _i and y _i respectively, θ _ji represents the phase difference between different CPG units, k Σ _j Δ _ji is a coupling polynomial for adjusting the smoothness of the output signal curve, ω _stance and ω _swing respectively Represents the supporting phase frequency and the wobble phase frequency, and b represents a constant factor.

Where x _i and y _i are control signals for the two forward joints of the legs of the quadruped robot. In a further embodiment, the y _i signal is subjected to a certain filtering process as a control signal for the knee joint, as follows:

Where c is a constant factor and θ _ki is the control signal of the filtered i-th CPG unit to the knee forward joint of the leg.

The proportional differential control is used to adjust the attitude of the body of the quadruped robot (that is, the proportional differential control is used to control the joints of the legs of the four-legged robot in the support phase):

In the above formula, q _l represents the attitude joint compensation angle of the l-th posture of the body of the quadruped robot generated by the controller, and a _l represents the actual attitude angle of the l-th posture of the body of the quadruped robot,

The actual attitude angular velocity of the l-th posture of the body of the quadruped robot, and a _dl represents the desired attitude angle of the l-th posture of the body of the quadruped robot,

The expected attitude angular velocity of the l-th pose of the body of the quadruped robot, k _l and b _l are the ratio and the differential constant of the l-th pose of the body of the quadruped robot, respectively, where l=1, 2, 3.

The above formula is expressed as a matrix:

Where K=diag(k ₁ , k ₂ , k ₃ ) and B=diag(b ₁ , b ₂ , b ₃ ) are third-order square matrix, respectively controlling the three attitude angles and angular rates of the body of the quadruped robot ;

with

An angular vector and an angular rate vector representing the actual posture of the body of the quadruped robot, respectively.

with

The angular vector and the angular rate vector of the desired posture of the body of the quadruped robot are respectively indicated, and q _a represents the attitude compensation angle of each joint of the leg of the four-legged robot generated by the PD controller.

In summary, in the present embodiment, the legs of the quadruped robot are controlled by the following formula:

q=q _a +q _c

Where q represents the total signal applied to the quadruped robot, q _a represents the attitude compensation angle of each joint controlling the quadruped robot in the leg of the supporting phase, and q _c is the front of the hip that controls each leg of the quadruped robot Control signals to the joint and knee forward joints, wherein, in a preferred embodiment, q _c = (x ₁ , θ _k1 , x ₂ , θ _k2 , x ₃ , θ _k3 , x ₄ , θ _k4 ) ^T .

The existing CPG control method is not easy to combine with the external feedback signal, and the VMC method is too cumbersome and complicated. The preferred embodiment of the present invention discloses a simple and efficient control method, which can overcome the defects of the prior art and realize the expected control. aims.

The motion closed-loop control method of the quadruped robot proposed by the preferred embodiment of the present invention generates periodic signals according to the bionics principle to control the periodic motion of the quadruped robot, which is more efficient and easier to understand than other dynamic methods; The differential control law adjusts the attitude of the quadruped robot body so that it can travel stably in a suitable attitude. Compared with other methods such as virtual model control, it is no longer necessary to calculate the single-leg Jacobian matrix and matrix inversion in real time, and eliminate the complexity. The tedious derivation and solution is simple and efficient.

The above is a further detailed description of the present invention in connection with the specific preferred embodiments, and the specific embodiments of the present invention are not limited to the description. It will be apparent to those skilled in the art that <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt;

Claims

A motion closed-loop control method for a quadruped robot, comprising the steps of: generating a periodic control signal for each leg of the quadruped robot by using a CPG network to generate a forward of the quadruped robot The signal is simultaneously adjusted by the proportional differential control to the posture of the body of the quadruped robot.
The motion closed-loop control method for a quadruped robot according to claim 1, wherein each of said legs of said quadruped robot is provided with three joints: a hip sway joint, a hip forward joint, and a knee forward joint, wherein the CPG network is for controlling a hip forward joint and a knee forward joint of each of the legs of the quadruped robot; the proportional differential control is used for The quadruped robot is in control of each joint of the leg of the support phase.
The motion closed-loop control method for a quadruped robot according to claim 2, wherein each of said legs of said quadruped robot is controlled by the following formula:

q=q a +q c

Wherein q represents a total signal applied to the quadruped robot, q a represents an attitude compensation angle for controlling each joint of the quadruped robot in the leg of the supporting phase, q c is a control of the quadruped robot Control signals for the hip forward joint and the knee forward joint of each of the legs.
The motion closed-loop control method for a quadruped robot according to claim 3, wherein the CPG network includes a CPG unit that respectively controls each of the legs of the quadruped robot, and a model adopted by the CPG unit The expression is:

Δ ji =y j cosθ ji -x j sinθ ji

Where i and j represent the leg numbers of the quadruped robot,
ω i represents the motion frequency of the i-th leg, α and β are convergence factors, μ is a parameter, and x i is the control signal of the i-th CPG unit to the hip forward joint of the leg, y i is a control signal of the i-th CPG unit to the knee forward joint of the leg,
with
Representing the differentiation of the signals corresponding to x i and y i respectively, θ ji represents the phase difference between different CPG units, k ∑ j Δ ji is a coupling polynomial, and ω stance and ω swing represent the frequency of the supporting phase and the frequency of the oscillating phase, respectively. b represents a constant factor; wherein q c = (x 1 , y 1 , x 2 , y 2 , x 3 , y 3 , x 4 , y 4 ) T .
The motion closed-loop control method for a quadruped robot according to claim 4, wherein the y i signal is filtered by the following formula:

Where c is a constant factor and θ ki is the control signal of the filtered i- th CPG unit to the knee forward joint of the leg; where q c = (x 1 , θ k1 , x 2 , θ K2 , x 3 , θ k3 , x 4 , θ k4 ) T .
The motion closed-loop control method for a quadruped robot according to claim 3, wherein the control law for adjusting the posture of the body of the quadruped robot by using proportional differential control is:

Where K=diag(k 1 , k 2 , k 3 ) and B=diag(b 1 , b 2 , b 3 ), where k l and b l are the first of the body of the quadruped robot, respectively The proportion of the pose and the differential constant, where l = 1, 2, 3;
with
An angular vector and an angular rate vector respectively representing the actual posture of the body of the quadruped robot,
with
An angular vector and an angular rate vector representing a desired posture of the body of the quadruped robot, respectively.
The motion closed-loop control method for a quadruped robot according to any one of claims 1 to 6, further comprising: when the posture of the body of the quadruped robot is consistent with a desired posture, PD control The device does not function; when the posture of the body of the quadruped robot is deviated, the PD controller generates an error compensation signal according to the control law of the proportional differential control, so that the body of the quadruped robot follows a desired posture Stable travel.
The motion closed-loop control method for a quadruped robot according to claim 7, wherein the PD controller receives the foot touch signal from the force sensor to determine the ground contact state of the bottom end of each of the legs.