Anti-sliding control method for quadruped robot

Abstract

The invention relates to the technical field of robots, and discloses a quadruped robot anti-sliding control method, which comprises a sliding detector for estimating the foot end contact condition, and a sliding compensation controller activated after the occurrence of sliding is recognized. The sliding detector judges whether sliding occurs or not by estimating the position and the speed of the contact feet of the quadruped robot in a world coordinate system; realizing a sliding compensation controller of the quadruped robot based on model predictive control, and constructing a model predictive control problem according to a kinetic model of the quadruped robot; and solving the model prediction control problem to obtain the optimal ground control force. According to the method, the speed and displacement errors of the end effector are introduced into the detection model, so that accurate detection of foot end sliding in different gait modes is realized. Meanwhile, the hybrid controller provided by the invention decomposes a control task of stabilizing the sliding leg into horizontal plane position control and vertical force control.

Description

Technical Field

[0001] The invention relates to the technical field of robots, and in particular to an anti-slip control method for a quadruped robot. Background Art

[0002] Due to the flexibility and adaptability of legged locomotion, quadruped robots have been widely used in challenging tasks in various terrains and conditions, such as mine exploration, industrial inspection and rescue. Unlike wheeled robots, quadruped robots, as typical floating base underactuated systems, need to obtain the external input force required for control through the contact between the foot end and the ground. Therefore, reliable estimation and execution of contact force are crucial for motion strategies. However, when moving on slippery surfaces such as sand, ice and snow, the lack of friction can cause the foot end of the legged robot to slip between the ground, which can cause the robot to become unstable and fall. Therefore, developing an anti-slip control algorithm for quadruped robots on sliding surfaces is of positive significance to improving their stability and safety.

[0003] Conventional quadruped robot control algorithms are mostly based on the static contact assumption to derive and control the foot contact force, that is, adding a kinematic constraint that the contact point velocity and acceleration are zero. By substituting this constraint into the dynamics of the floating base, the contact force can be derived as a constraint force without the need for additional force sensors. However, when sliding occurs, the static contact assumption no longer holds true. At this time, the contact point of the supporting leg will deviate from its predetermined position, causing the supporting polygon to deform, causing the robot to lose balance and fall. In order to improve the motion stability of the sliding surface, some existing algorithms actively switch to a slower quadruped control gait, or directly switch the foot control to position control to curb the occurrence of sliding. Although these methods can prevent the robot from falling, they will also greatly damage the robot's motion performance. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a quadruped robot anti-slip control method to improve the stability of the quadruped robot moving on a sliding surface; the method accurately determines the occurrence of slip by designing a slip detector that estimates the contact condition of the foot end. At the same time, a slip compensation controller that is activated when slip occurs is designed to achieve contact force solution and hybrid control of sliding contact compensation position control through orthogonal decomposition and null space projection.

[0005] Slippery surfaces place high demands on the motion control algorithms of quadruped robots. Conventional leg control methods derive contact forces by assuming stationary foot contact, but this assumption no longer holds true under slippery conditions. This paper develops a slip detector that does not require any external sensors, and proposes a novel slip recovery controller using orthogonal decomposition to stabilize the sliding contact and prevent the robot from falling. This method aims to solve the problems of inaccurate contact traction and sliding displacement of the contact foot of quadruped robots on slippery surfaces, so as to improve the stability and safety of their motion.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A method for controlling a quadruped robot against sliding includes a sliding detector for estimating a foot end contact condition, and a sliding compensation controller activated after recognizing the occurrence of sliding; wherein:

[0008] The slip detector estimates the position and velocity of the quadruped robot's contact foot in the world coordinate system to determine whether slipping occurs;

[0009] The sliding compensation controller of the quadruped robot is realized based on model predictive control. The dynamic model of the quadruped robot is a single rigid body dynamic model that ignores the mass of the legs. The model predictive control problem is constructed according to the dynamic model of the quadruped robot; the model predictive control problem is solved to obtain the optimal ground control force.

[0010] Furthermore, the sliding detector estimates the position and speed of the contact foot of the quadruped robot in the world coordinate system to determine whether sliding occurs, specifically including:

[0011] Define the position error δd of the i-th contact foot i and velocity error δv i for:

[0012] δd i =||x c,i -x c0,i ||;

[0013]

[0014] Among them, x c,i is the position of the robot’s ith contact foot in the world coordinate system, x c0,i is the position where the robot’s ith contact foot first contacts the ground in the world coordinate system, J c,i is the Jacobian matrix corresponding to the ith contact foot of the robot, and q represents the generalized coordinates of the robot;

[0015] Based on the position error δd i and velocity error δv i, estimate the slip probability P of the i-th contact foot slip,i :

[0016]

[0017] Among them, the expression of erf function is k p and k d is the proportional coefficient of the sliding detector to adjust the sensitivity of the detection;

[0018] The judgment criteria of the sliding detector are as follows:

[0019]

[0020] Where X slip,i is the sliding judgment state of the i-th contact foot, True indicates sliding, and False indicates no sliding.

[0021] Furthermore, the generalized coordinate q of the robot is given by a Kalman filter algorithm, and the input of the Kalman filter algorithm is the proprioceptive information of the robot, including a foot contact sensor, an inertial measurement unit IMU and a joint angle encoder.

[0022] Furthermore, the sliding compensation controller of the quadruped robot is realized based on the model predictive control. The dynamic model of the quadruped robot is a single rigid body dynamic model ignoring the mass of the legs. The model predictive control problem is constructed according to the dynamic model of the quadruped robot, which specifically includes:

[0023] The dynamic model of the quadruped robot is:

[0024]

[0025] Where m is the mass of the robot, p is the position of the center of mass of the robot body in the world coordinate system, and λ i is the contact force between the ith contact foot and the ground, g is the acceleration due to gravity, I is the moment of inertia of the robot's center of mass, ω is the angular velocity of the body, and p f,i is the position of the i-th contact foot in the world coordinate system;

[0026] Define the robot's state x and control quantity u as:

[0027]

[0028] where Θ = [φ,θ,ψ] T It is the Euler angle representation of the robot's body roll angle, pitch angle, and yaw angle, and the mapping of the angular velocity ω to the Euler angle change R(θ) is:

[0029]

[0030] Combining equations (5) to (9), we get the linear state space equation:

[0031]

[0032] in,

[0033]

[0034] Then, the state transfer matrices A(t) and B(t) are converted into discrete forms using a first-order retainer. k and B k ; Then we get the following linear model predictive control problem:

[0035]

[0036] Where N is the prediction step size, x k and is the current state and set state of the robot at time k; R and Q are two positive definite weight matrices; x s is the initial state of the robot; u min and u max Respectively represent the maximum and minimum limits of the robot's control force, u k represents the control quantity at the kth moment, and C represents the constraint coefficient matrix;

[0037] Constraint u min ≤Cu k ≤u max In addition to basic limiting, the content also includes friction cone limiting to prevent the robot from slipping:

[0038]

[0039] where μ i is the friction coefficient between the ith contact foot and the ground, c(i)∈{0,1} is used to indicate whether the ith contact foot is suitable for ground contact; λ i,x ,λ i,y ,λ i,z are the components of the contact force between the ith contact foot and the ground in the x-axis direction, y-axis direction and z-axis direction respectively; max Indicates the maximum allowable ground contact force in the z direction.

[0040] Furthermore, solving the model predictive control problem to obtain the optimal ground control force specifically includes:

[0041] When the i-th contact foot and the ground slide, that is, X slip,i When True, the corresponding μ iwill be set to 0; this means that the sliding contact foot is only allowed to generate contact force in the vertical direction with the ground, which can prevent sliding;

[0042] At the same time, in order to keep the sliding contact foot in the correct position, the sliding compensation controller will output a horizontal sliding restoring force

[0043]

[0044] where p s,i is the position of the i-th contact foot in the sliding state when it starts sliding, and is a positive definite matrix parameter; sliding restoring force Always parallel to the ground, used to pull the sliding contact foot to the correct contact point and maintain it;

[0045] The motor control torque τ of the robot's i-th leg i Obtained by the following formula:

[0046]

[0047] Among them, J i is the Jacobian matrix of the ith leg in the fuselage coordinate system, and R is the rotation transformation matrix from the fuselage to the world coordinate system.

[0048] Compared with the prior art, the beneficial technical effects of the present invention are:

[0049] The present invention develops a slip detector that does not require any external sensors, and proposes a novel slip recovery controller using orthogonal decomposition to stabilize the sliding contact and prevent the robot from falling. The present invention combines a hybrid position / force controller and integrates it into a model predictive control (MPC) motion strategy. By introducing the velocity and displacement errors of the end effector into the detection model, the present invention achieves accurate detection of foot-end slip in different gait patterns. At the same time, the hybrid controller proposed in the present invention decomposes the control task of stabilizing the sliding leg into horizontal plane position control and vertical force control.

[0050] Conventional quadruped robot control algorithms are mostly based on the static contact assumption to derive and control the foot contact force. Therefore, they have poor adaptability under sliding conditions and often have difficulty in responding quickly to real-time dynamic changes, resulting in the robot being unable to adjust its posture or gait in time when encountering sliding or ground friction changes. In contrast, the method proposed in the present invention can solve the problems of inaccurate contact traction and sliding displacement of the contact foot of a quadruped robot on slippery surfaces, thereby improving the stability and safety of its movement. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1It is a system block diagram of the present invention.

[0052] Figure 2 Schematic diagram of application scenario of the invention. DETAILED DESCRIPTION

[0053] A preferred embodiment of the present invention is described in detail below with reference to the accompanying drawings.

[0054] The present invention designs a quadruped robot anti-slip control method for slippery scenes. The system used is as follows Figure 1 As shown, it mainly includes a slip detector for estimating the contact situation of the foot end, and a slip compensation controller activated after identifying that slip occurs.

[0055] 1. Slide detector

[0056] The slip detector estimates the position and speed of the quadruped robot's contact foot in the world coordinate system to determine whether slip has occurred. The specific technical process is as follows:

[0057] First, for each contact foot, define the position error δd of the foot end i and velocity error δv i for:

[0058] δd i =||x c,i -x c0,i ||;

[0059]

[0060] Among them, x c,i is the position of the ith contact foot in the world coordinate system, x c0,i is the position where the ith contact foot first contacts the ground in the world coordinate system. c,i is the Jacobian matrix corresponding to the ith contact foot, and q corresponds to the generalized coordinates of the robot.

[0061] The robot's generalized coordinates q are given by the Kalman filter algorithm, whose input is the robot's proprioceptive information, including foot contact sensors, inertial measurement units (IMUs), and joint angle encoders.

[0062] In order to curb the noise of the sensor, a low-pass filter is used to filter the foot end velocity error to reduce the occurrence of sliding misjudgment.

[0063] Based on the above two errors, the slip probability P of the i-th contact foot can be estimated slip,i , the specific calculation formula is:

[0064]

[0065] Among them, the expression of erf function is k p and k d It is the proportional coefficient of the sliding detector to adjust the sensitivity of the detection.

[0066] Finally, the judgment criteria for the slip detector are as follows:

[0067]

[0068] Where X slip,i is the sliding judgment state of the i-th contact foot.

[0069] 2. Slip compensation controller

[0070] The quadruped robot and its sliding compensation controller are implemented based on an MPC (model predictive control), which is introduced in detail below.

[0071] The dynamic model of the quadruped robot in MPC is a single rigid body dynamic model that ignores the mass of the legs:

[0072]

[0073] Where m is the mass of the robot, p is the position of the center of mass of the robot body in the world coordinate system, and λ i is the contact force between the ith contact foot and the ground, g is the acceleration due to gravity, I is the moment of inertia of the robot's center of mass, ω is the angular velocity of the body, and p f,i is the position of the ith contact foot in the world coordinate system.

[0074] Then, the robot's state x and control quantity u are defined as:

[0075]

[0076]

[0077] where Θ = [φ,θ,ψ] T is the Euler angle representation of the robot's body roll, pitch, and yaw, and the mapping of the angular velocity ω to the change in this Euler angle for

[0078] The dynamic equations can be combined to obtain a linear state space equation:

[0079]

[0080] in,

[0081]

[0082] Then, the state transfer matrices A(t) and B(t) are converted into discrete forms using a first-order retainer. k and B k .

[0083] According to the above dynamics, the following linear MPC problem can be obtained:

[0084]

[0085] Where N is the prediction step size, x k and is the current state and set state of the robot at time k. R and Q are two positive definite weight matrices. s is the initial state of the robot.

[0086] Constraint u min ≤Cu k ≤u max In addition to basic limiting, the content also includes friction cone limiting to prevent the robot from slipping:

[0087]

[0088] where μ i is the friction coefficient between the ith contact foot and the ground, and c(i)∈{0,1} is used to indicate whether the ith contact foot is suitable for ground contact.

[0089] By solving this MPC problem, we can obtain the control amount u required for tracking the trajectory, that is, the optimal ground control force.

[0090] When the i-th contact foot and the ground slide, that is, X slip,i When True, the corresponding μ i will be set to 0. This means that the sliding leg is only allowed to generate vertical contact force with the ground, which will prevent sliding.

[0091] At the same time, in order to stabilize the sliding leg to keep it in the correct position, the sliding compensation controller will output a horizontal sliding restoring force

[0092]

[0093] where p s,i is the position of the i-th contact foot in the sliding state when it starts sliding. and is a positive definite matrix parameter. Since the foot always slides against the ground, the sliding restoring force The restoring force will pull the sliding foot to its correct contact point and keep it there, thus curbing the sliding of the foot end.

[0094] Finally, the motor control torque τ of the i-th leg of the legged robot i Obtained by the following formula:

[0095]

[0096] Among them, J i is the Jacobian matrix of the ith leg in the fuselage coordinate system, and R is the rotation transformation matrix from the fuselage to the world coordinate system.

[0097] Figure 1 The left side corresponds to its reference state generation; the middle part is the control strategy, including the slip compensator and MPC controller; the right side is the perception part, including the state estimator and slip estimator.

[0098] Figure 2 An example of an invention application scenario: a standing quadruped robot with its left front leg sliding from the initial contact position (blue) to the current position (red). The sliding detector detects the occurrence of sliding by measuring the subsequent lateral velocity and displacement of the foot. Subsequently, the activated sliding compensation controller generates the desired sliding restoring force for position control to stabilize the sliding leg; at the same time, it provides vertical contact force to support the robot torso.

[0099] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0100] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A quadruped robot anti-slip control method, characterized in that: A slip detector is provided for estimating a foot end contact condition, and a slip compensation controller is activated after detecting that a slip has occurred; wherein: The slip detector estimates the position and velocity of the quadruped robot's contact foot in the world coordinate system to determine whether slipping occurs; The sliding compensation controller of the quadruped robot is realized based on model predictive control. The dynamic model of the quadruped robot is a single rigid body dynamic model that ignores the mass of the legs. The model predictive control problem is constructed according to the dynamic model of the quadruped robot; the model predictive control problem is solved to obtain the optimal ground control force.

2. The anti-slip control method of a quadruped robot according to claim 1, characterized in that: The sliding detector estimates the position and speed of the contact foot of the quadruped robot in the world coordinate system to determine whether sliding occurs, specifically including: Define the position error δd of the i-th contact foot i and velocity error δv i for: δd i =||x c，i -x c0，i ||; Among them, x c,i is the position of the robot’s ith contact foot in the world coordinate system, x c0，i is the position where the robot’s ith contact foot first contacts the ground in the world coordinate system, J c，i is the Jacobian matrix corresponding to the ith contact foot of the robot, and q represents the generalized coordinates of the robot; Based on the position error δd i and velocity error δv i , estimate the slip probability P of the i-th contact foot slip,i : Among them, the expression of erf function is k p and k d is the proportional coefficient of the sliding detector to adjust the sensitivity of the detection; The judgment criteria of the sliding detector are as follows: Where X slip,i is the sliding judgment state of the i-th contact foot, True indicates sliding, and False indicates no sliding.

3. The anti-slip control method of a quadruped robot according to claim 2, characterized in that: The generalized coordinate q of the robot is given by a Kalman filter algorithm, and the input of the Kalman filter algorithm is the proprioceptive information of the robot, including a foot contact sensor, an inertial measurement unit IMU, and a joint angle encoder.

4. The anti-slip control method of a quadruped robot according to claim 1, characterized in that: The sliding compensation controller of the quadruped robot is realized based on the model predictive control. The dynamic model of the quadruped robot is a single rigid body dynamic model ignoring the mass of the legs. The model predictive control problem is constructed according to the dynamic model of the quadruped robot, which specifically includes: The dynamic model of the quadruped robot is: Where m is the mass of the robot, p is the position of the center of mass of the robot body in the world coordinate system, and λ i is the contact force between the ith contact foot and the ground, g is the acceleration due to gravity, I is the moment of inertia of the robot's center of mass, ω is the angular velocity of the body, and p f，i is the position of the i-th contact foot in the world coordinate system; Define the robot's state x and control quantity u as: x=[Θ T ,p T Oh, oh T ,p T ] T (7) where Θ = [φ, θ, ψ] T It is the Euler angle representation of the robot's body roll angle, pitch angle, and yaw angle, and the mapping of the angular velocity ω to the Euler angle change R(θ) is: Combining equations (5) to (9), we get the linear state space equation: in, Then, the state transfer matrices A(t) and B(t) are converted into discrete forms using a first-order retainer. k and B k ; Then we get the following linear model predictive control problem: Where N is the prediction step size, x k and is the current state and set state of the robot at time k; R and Q are two positive definite weight matrices; x s is the initial state of the robot; u min and u max Respectively represent the maximum and minimum limits of the robot's control force, u k represents the control quantity at the kth moment, and C represents the constraint coefficient matrix; Constraint u min ≤Cu k ≤u max In addition to basic limiting, the content also includes friction cone limiting to prevent the robot from slipping: where μ i is the friction coefficient between the ith contact foot and the ground, c(i)∈{0,1} is used to indicate whether the ith contact foot is suitable for ground contact; λ i，x ,λ i，y ,λ i，z are the components of the contact force between the ith contact foot and the ground in the x-axis direction, y-axis direction and z-axis direction respectively; max Indicates the maximum allowable ground contact force in the z direction.

5. The anti-slip control method of a quadruped robot according to claim 4, characterized in that: The method of solving the model predictive control problem to obtain the optimal ground control force specifically includes: When the i-th contact foot and the ground slide, that is, X slip,i When True, the corresponding μ i will be set to 0; this means that the sliding contact foot is only allowed to generate contact force in the vertical direction with the ground, which can prevent sliding; At the same time, in order to keep the sliding contact foot in the correct position, the sliding compensation controller will output a horizontal sliding restoring force where p s，i is the position of the i-th contact foot in the sliding state when it starts sliding, and is a positive definite matrix parameter; sliding restoring force Always parallel to the ground, used to pull the sliding contact foot to the correct contact point and maintain it; The motor control torque τ of the robot's i-th leg i Obtained by the following formula: Among them, J i is the Jacobian matrix of the ith leg in the fuselage coordinate system, and R is the rotation transformation matrix from the fuselage to the world coordinate system.

Images