Four-legged robot composite control method and system based on super-spiral sliding mode anti-interference model

Abstract

The application discloses a quadruped robot composite control method and system based on a super-helix sliding mode anti-interference model, wherein the method comprises the following steps: firstly, a single rigid body dynamics model containing unknown disturbance is established; secondly, a super-helix sliding mode disturbance observer is designed to perform high-precision and strong-robust estimation on the translational and rotational disturbances; in the support phase, the disturbance estimation value is fed forward to a model predictive controller to solve a constrained optimization problem to generate optimal foot end force; in the swing phase, a virtual model control is adopted to realize fast tracking of the foot end trajectory; and the smooth switching of the control strategies in the support phase and the swing phase is realized through contact state detection. The application can effectively improve the motion stability, trajectory tracking accuracy and anti-interference ability of the robot in an unstructured environment.

Description

Technical Field

[0001] This invention relates to the technical field of robot motion control, and in particular to a composite control method and system for a quadruped robot based on a super-helical sliding mode anti-interference model. Background Technology

[0002] As a biomimetic mobile platform, quadruped robots, with their redundant degrees of freedom in their legs enabling flexible gait generation and strong terrain adaptability, demonstrate irreplaceable application potential in a range of complex, unstructured scenarios, including emergency rescue, field exploration, material transportation, and security patrols. These task environments are often characterized by rugged terrain and dynamic changes, frequently featuring randomly distributed ground protrusions, soft soil, and sudden obstacles—a multitude of uncertainties. Therefore, achieving highly robust and precise autonomous movement of quadruped robots in such environments is a core prerequisite for their practical engineering applications, and it also places extremely stringent requirements on the design of motion controllers. From a kinematics perspective, the periodic gait of a quadruped robot can be decoupled into two alternating phases: a support phase and a swinging phase. These two phases differ fundamentally in terms of control objectives and physical constraints. During the support phase, the robot relies on the ground reaction force provided by its ground-touching feet to drive trunk movement and maintain overall balance. The core of control in this phase lies in achieving precise force distribution to each foot while meeting physical constraints such as ground friction cone constraints and no foot slippage, thereby completing stable tracking control of the fuselage attitude and center of mass trajectory. During the sway phase, the legs need to quickly and accurately track a preset trajectory in the air to achieve periodic stepping movements, obstacle avoidance, and stable landing. The control tasks in these two phases are fundamentally contradictory: the support phase is essentially a force control problem, focusing on optimizing global forces and simultaneously satisfying multiple constraints; while the sway phase is mainly a trajectory tracking problem, pursuing local dynamic response speed and anti-interference capabilities. Traditional solutions using a single controller architecture struggle to achieve optimal coordinated control throughout the entire gait cycle, often necessitating compromises in one phase to ensure performance in another.

[0003] To address these challenges, academia and industry have developed various control strategies; however, these methods still have significant limitations under the comprehensive challenges of unstructured environments. Model predictive control (MPC), due to its ability to explicitly handle system constraints and perform rolling time-domain optimization, has become one of the mainstream methods for dynamic gait control, especially in support phase or whole-body coordinated control. However, the control performance of MPC is highly dependent on the accuracy of its predictive model. When faced with unknown external disturbances introduced by unstructured environments and model parameter mismatches within the robot itself, model errors will cause the optimization objective to deviate from reality, leading to a significant decrease in control performance or even system instability. Although robust MPC or tube-MPC methods have been proposed to enhance system robustness, these methods usually come at the cost of nominal performance and have high computational complexity, making them difficult to widely apply in the real-time demanding control of quadruped robots.

[0004] Virtual model control, an intuitive control method based on virtual component models, generates virtual forces or torques by simulating spring-damping systems. It is commonly used for body posture adjustment or foot trajectory tracking during the swing phase. This method has a simple structure and high computational efficiency. However, when used in the full-time domain control of quadruped robots, it cannot guarantee the stability of the support force or compensate for disturbances, thus leading to instability of the center of mass.

[0005] For addressing disturbance issues in a system, common solutions include directly designing an anti-interference controller or introducing an observer for feedforward compensation. Linear disturbance observers or extended state observers can estimate and compensate for the "total disturbance," but their design typically relies on strong assumptions such as "bounded rate of change of disturbance" or "known dynamic model of disturbance." In the actual motion of quadruped robots, disturbances often exhibit high-order, rapidly time-varying characteristics, making these assumptions difficult to strictly adhere to, thus limiting the performance of the observer. While sliding mode observers theoretically possess strong robustness to matched disturbances, traditional first-order sliding mode observers suffer from significant high-frequency chattering and have insufficient ability to handle unmatched disturbances.

[0006] Existing quadruped robot control methods fail to achieve coordinated optimization of the constraint control requirements of the support phase and the trajectory tracking requirements of the swing phase within a unified framework when dealing with unstructured environments. Optimization-based controllers such as MPC lack real-time, robust estimation capabilities for complex disturbances. Traditional observers rely on overly idealistic dynamic assumptions about disturbances, making it difficult to adapt to the rapidly changing, high-order unknown disturbance characteristics in real-world environments. These limitations collectively result in poor overall motion stability, trajectory tracking accuracy, and environmental adaptability of the robot in complex terrain, failing to meet the requirements of high-reliability tasks. Therefore, current research focuses on developing a cooperative control scheme that effectively integrates the advantages of phased control and incorporates advanced disturbance rejection estimation and compensation techniques, thereby systematically improving the overall motion performance of quadruped robots in unstructured environments. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a composite control method and system for quadruped robots based on a super-helical sliding mode anti-interference model.

[0008] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0009] A composite control method for quadruped robots based on a superspiral sliding mode anti-interference model includes:

[0010] A single rigid body motion-dynamic model of a quadruped robot with unknown disturbances is constructed. The model includes the translational dynamic equation of the center of mass and the rotational dynamic equation of the center of mass. The equivalent external disturbance force and equivalent disturbance torque are explicitly introduced into the equations to provide a predictive model basis with disturbances for the model predictive controller.

[0011] A translational perturbation observer and a rotational perturbation observer are designed independently for the translational dynamic equation and the rotational dynamic equation of the center of mass, respectively. By constructing a sliding surface and solving the nonlinear dynamic update law, the estimated values ​​of translational perturbation acceleration and rotational perturbation angular acceleration are output in real time.

[0012] During the support phase, the output disturbance acceleration and disturbance angular acceleration estimates are used as feedforward compensation terms and injected into the single rigid body motion-dynamic model of the quadruped robot after linearization and discretization. A prediction model with disturbance compensation is constructed. The model prediction controller then solves the rolling time domain optimization problem based on the prediction model to generate the optimal foot contact force sequence, which is mapped to the support-related joint torque command through the foot Jacobian matrix.

[0013] During the swing phase, a virtual control force is generated by a spring-damped virtual model controller. The virtual control force is based on the foot position and velocity tracking error and mapped to the swing joint torque command through the leg Jacobian matrix to drive the swing leg to track the preset Bezier trajectory.

[0014] Based on the leg contact state detection results, a smooth switching logic is executed between the support phase and the swing phase to ensure that the generated support-related joint torque command and the swing-related joint torque command are seamlessly connected, thereby achieving stable motion control throughout the entire gait cycle.

[0015] Furthermore, the equations of motion for the translational movement of the center of mass are as follows:

[0016] ;

[0017] in, For the overall quality of the robot; Let be the linear acceleration vector of the center of mass in the world coordinate system; This represents the current number of supporting legs in contact with the ground. For the first The ground reaction force vector at the foot of the supporting leg; This is the vector of gravitational acceleration; The translational disturbance force characterizes the uncertainty in the translational direction, including the impact reaction force caused by ground protrusion, lateral wind load, and inertial force caused by sudden load changes;

[0018] The dynamic equation for the rotation of the center of mass is as follows:

[0019] ;

[0020] in, Let be the moment of inertia matrix of the fuselage about its center of mass; This is the angular velocity vector of the fuselage relative to the world coordinate system; Let be the position vector of the centroid in the world coordinate system; For the first The position vector of the foot end of the supporting leg; The rotational disturbance torque characterizes the uncertainty in the direction of rotation, including torque fluctuations caused by deviations in the foot contact position, fuselage attitude disturbances, and changes in the joint friction coefficient.

[0021] Furthermore, the process of designing a translational perturbation observer includes:

[0022] The nominal translational acceleration of the center of mass is:

[0023] ;

[0024] The sliding surface is constructed as follows:

[0025] ;

[0026] in, To translate and observe the sliding surface, The location of the center of mass. For the estimated location, For the velocity of the center of mass, It is an estimate of speed. This is the weighting coefficient for the position error of the sliding surface;

[0027] The derivative of the position estimate is:

[0028] ;

[0029] The derivative of the velocity estimate is:

[0030] ;

[0031] in, To approximate the internal state of the disturbance, they are designed as follows:

[0032] ;

[0033] ;

[0034] in, For the observer position channel gain, For the observer velocity channel gain, For component-wise sign functions;

[0035] Let the position and velocity estimation errors be:

[0036] ;

[0037] ;

[0038] Therefore, the error derivative and the sliding surface derivative are:

[0039] ;

[0040] ;

[0041] ;

[0042] Among them, translational interference force Bounded, that is ,in It is an unknown constant.

[0043] Furthermore, the process of designing a rotating perturbation observer includes:

[0044] The nominal angular acceleration of the center of mass is:

[0045] ;

[0046] Constructing a rotating sliding surface:

[0047] ;

[0048] in, To rotate the observation sliding surface, Angular velocity, For angular velocity estimation, For small-angle attitude vectors, For attitude estimation, This is the weighting coefficient for the position error of the rotating sliding surface;

[0049] The attitude estimation derivative is:

[0050] ;

[0051] The derivative of the estimated angular velocity is:

[0052] ;

[0053] The internal state is:

[0054] ;

[0055] ;

[0056] in, For observer gain;

[0057] Let the attitude estimation error be:

[0058] ;

[0059] The angular velocity estimation error is:

[0060] ;

[0061] The derivative of the error estimate is:

[0062] ;

[0063] ;

[0064] The derivative of the sliding surface is:

[0065] ;

[0066] Among them, the rotational disturbance torque Bounded, that is ,in This is an unknown constant.

[0067] Furthermore, the process of obtaining the support-related torque command includes:

[0068] The single-rigid-body motion-dynamic model of the quadruped robot is linearized and discretized. The estimated values ​​of translational perturbation acceleration and rotational perturbation angular acceleration are used as inputs to the model, forming the discrete state-space equations for prediction.

[0069] ;

[0070] in, for The state vector at any given moment includes attitude angle, position, angular velocity, and linear velocity; for The ground reaction force calculated at all times; The system matrix after linearization and precise discretization; , This is the vector of gravitational acceleration; ; This is an estimate of the translational disturbance acceleration; This is an estimate of the angular acceleration of the rotational disturbance;

[0071] The rolling optimization problem is defined based on the discrete state-space equation. The deviation between the future state and the reference trajectory is minimized, while the control input energy is constrained. Ground friction cone constraint, foot slip-free constraint and torque constraint are applied. The optimal foot force sequence is obtained by solving the quadratic programming problem.

[0072] Take the first control variable in the sequence and map it to the support-related joint torque command through the Jacobian matrix at the foot of each support leg:

[0073] ;

[0074] in, For the first Leg joint torque, For the foot Jacobian matrix, This is the rotation matrix for the machine's attitude.

[0075] Furthermore, the process of obtaining the swing-related joint torque command includes:

[0076] Design virtual model control force :

[0077] ;

[0078] in, These are the position and velocity gain matrices, respectively. and These are the center of mass position and velocity tracking errors, respectively. By adjusting the diagonal elements, the tracking stiffness and damping characteristics in each direction can be independently controlled to ensure the smoothness of trajectory tracking.

[0079] The virtual model is controlled by the Jacobian matrix of the legs. Mapped to swing-related joint torque commands.

[0080] Furthermore, the smooth switching logic includes:

[0081] The system monitors the contact state of each leg of the quadruped robot in real time. When a leg switches from the swing phase to the support phase, the model predictive controller is activated for force control; when a leg switches from the support phase to the swing phase, the virtual model controller is activated for trajectory tracking control.

[0082] Design a smooth interpolation mechanism for the control quantity during the transition phase to avoid abrupt changes in the control quantity at the moment of switching, and ensure the continuity and stability of the control output throughout the entire gait cycle.

[0083] Furthermore, to achieve the above objectives, the present invention also provides a quadruped robot composite control system based on a superhelical sliding mode anti-interference model, used to implement the above-mentioned quadruped robot composite control method based on a superhelical sliding mode anti-interference model, which includes a modeling module, a disturbance observer design and observation module, a support phase control module, a swing phase control module, and a logic switching module.

[0084] The modeling module is used to construct a single rigid body motion-dynamic model of a quadruped robot with unknown perturbations;

[0085] The disturbance observer design and observation module is used to design translational disturbance observers and rotational disturbance observers, output estimated values ​​of translational disturbance acceleration and rotational disturbance angular acceleration, and connect with the modeling module to obtain model parameters;

[0086] The support phase control module is used to perform motion control during the support phase phase and is connected to the disturbance observer design module to obtain estimated values ​​of disturbance acceleration and disturbance angular acceleration.

[0087] The swing phase control module is used to perform motion control during the swing phase phase.

[0088] The logic switching module is used to execute smooth switching logic and coordinate the working states of the support phase control module and the swing phase control module.

[0089] Furthermore, the support phase control module includes:

[0090] The prediction model building unit is used to construct a prediction model by using the disturbance estimate as a feedforward compensation term.

[0091] The rolling optimization solver unit is used to define and solve quadratic programming optimization problems with physical constraints.

[0092] The force mapping unit is used to map the optimal foot force into the support-related joint torque command through the foot Jacobian matrix.

[0093] Furthermore, the disturbance observer design and observation module includes:

[0094] The disturbance observer design unit is used to design translational disturbance observers and rotational disturbance observers;

[0095] The observation unit is used to output estimates of translational perturbation acceleration and rotational perturbation angular acceleration using translational perturbation observers and rotational perturbation observers.

[0096] Compared with existing technologies, the principles and advantages of this technical solution are as follows:

[0097] 1. By proposing a collaborative framework of "superhelical sliding mode observation-staged optimization control," the global force optimization of the support phase and the local trajectory tracking task of the oscillating phase are decoupled, and a smooth switching logic based on contact state is designed. This framework integrates the constraint handling capability of model predictive control, the fast response characteristics of virtual model control, and the robust estimation capability of the superhelical sliding mode observer, enabling the quadruped robot system to achieve high-precision and high-stability motion control throughout the entire gait cycle. This method can significantly reduce the body posture error and foot tracking error under complex road conditions, and systematically improve the stability of the gait cycle.

[0098] 2. To address the complex time-varying perturbation problem faced by quadruped robots in unstructured environments, a super-helical sliding mode perturbation observer was designed. This observer does not require precise knowledge of the boundary of the perturbation derivative, only an assumption that it is bounded, thus relaxing the restrictions on the dynamic characteristics of the perturbation and enhancing its adaptability and estimation robustness to rapidly changing perturbations such as ground impacts and sudden load changes. Furthermore, the perturbation estimate output by the observer is used as a feedforward compensation term and directly injected into the prediction model of the support phase model predictive controller, realizing active and accurate compensation for perturbations and improving the performance degradation problem of traditional MPC under model mismatch and unknown disturbances.

[0099] 3. The control architecture embodies a task-oriented, complementary controller design. In the support phase, a model predictive controller (MPC) is used to centrally handle complex optimization problems such as ground reaction force constraints and friction cone constraints, focusing on fuselage attitude stability. In the oscillation phase, a simple and computationally efficient PD-type virtual model controller is used to handle rapid foot trajectory tracking. The two are coordinated through upper-level planning and lower-level contact detection, with smooth switching logic ensuring a seamless transition. This complementary mode of "MPC handling complex constraints and VMC handling rapid tracking" effectively balances the system's computational complexity and real-time requirements while ensuring control performance. Attached Figure Description

[0100] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0101] Figure 1 This is a flowchart illustrating the principle of the quadruped robot composite control method based on the super-helical sliding mode anti-interference model of the present invention.

[0102] Figure 2 This is a schematic diagram of the coordinate system of the left foreleg joint of the quadruped robot used in this invention;

[0103] Figure 3 This is a planar projection of the left foreleg of a quadruped robot.

[0104] Figure 4 This is a ZX-plane projection of the left front leg of a quadruped robot.

[0105] Figure 5 To track the desired height of the quadruped robot's center of mass;

[0106] Figure 6 The graph shows the variation of the estimated translational disturbance acceleration obtained from the translational disturbance observer.

[0107] Figure 7 This is a graph showing the variation of the estimated angular acceleration of the rotational disturbance obtained from the rotational disturbance observer. Detailed Implementation

[0108] The present invention will be further described below with reference to specific embodiments:

[0109] like Figure 1 As shown in this embodiment, the quadruped robot composite control method based on the superspiral sliding mode anti-interference model includes the following steps:

[0110] S1. Construct a single rigid body motion-dynamic model of a quadruped robot with unknown disturbances. The model includes the translational dynamic equation of the center of mass and the rotational dynamic equation of the center of mass. The equivalent external disturbance force and equivalent disturbance torque are explicitly introduced into the equations to provide a predictive model basis with disturbances for the model predictive controller.

[0111] The process of constructing a single rigid body motion-dynamic model of a quadruped robot with unknown perturbations is as follows:

[0112] 1) Perform kinematic modeling on the quadruped robot:

[0113] First, it is necessary to perform single-leg forward kinematics calculations on the quadruped robot. For example... Figure 2 As shown in the leg joint coordinate system, the hip joint lateral movement coordinate system... can be The coordinate system is obtained by rotating it around the x-axis. Let... The hip joint lateral swing angle, The hip flexion-extension angle, For the knee flexion-extension angle, then Compared to The rotation matrix is:

[0114] ;

[0115] Hip flexion-extension joint coordinate system Can be determined by coordinate system Obtained by rotating about the y-axis, then Compared to The rotation matrix is:

[0116] ;

[0117] Knee flexion-extension joint coordinate system Can be determined by coordinate system Obtained by rotating about the y-axis, then Compared to The rotation matrix is:

[0118] ;

[0119] in, , for and The length of the connecting rod between them. The position of the foot relative to... If the position is constant, then the homogeneous coordinates of the foot position are:

[0120] ;

[0121] in, The length of the link from the knee joint to the foot. The y-axis is offset in the negative direction. .

[0122] Secondly, inverse kinematics calculations are performed on the quadruped robot. The purpose of inverse kinematics is to convert the known foot positions in three-dimensional space into rotation angles of each joint. Taking the left foreleg as an example, for instance... Figure 3 As shown, when the hip joint is laterally rotated, it is projected onto the zy plane. When the foot position is set as ,but:

[0123] ;

[0124] in, We can obtain:

[0125] ;

[0126] like Figure 4 As shown, if the left foreleg is projected onto the zx plane, then:

[0127] ;

[0128] in ,but:

[0129] ;

[0130] 2) Perform dynamic modeling of the quadruped robot:

[0131] To accurately characterize the dynamics of a quadruped robot in unstructured environments while meeting the complexity requirements of real-time control algorithms, a dynamic model suitable for uncertain scenarios is constructed based on single-rigid-body dynamics and by introducing unknown disturbance terms. First, the translational equations of the robot's center of mass are established. The translational dynamics of the robot's center of mass follow Newton's second law, representing the balance between the center of mass acceleration and external forces, as shown in the following equation:

[0132] ;

[0133] in, For the overall quality of the robot; Let be the linear acceleration vector of the center of mass in the world coordinate system; This represents the current number of supporting legs in contact with the ground. For the first The ground reaction force vector at the foot of the supporting leg; Let be the vector of gravitational acceleration. The rotational dynamics of the fuselage about its center of mass follows the law of angular momentum, describing the balance between the change in angular velocity and the external torque. The mathematical expression is:

[0134] ;

[0135] in, Let be the moment of inertia matrix of the fuselage about its center of mass; This is the angular velocity vector of the fuselage relative to the world coordinate system; Let be the position vector of the centroid in the world coordinate system; For the first The position vector of the foot end of the supporting leg.

[0136] For the two core dimensions of robot motion, interference terms are defined respectively:

[0137] Translational interference force : Characterizes the uncertainty in the translational direction, including impact reaction forces caused by ground protrusion, lateral wind loads, and inertial forces caused by sudden load changes; rotational disturbance torque, etc. This characterizes the uncertainty in the direction of rotation, including torque fluctuations caused by deviations in foot contact position, fuselage attitude disturbances, and changes in joint friction coefficients. Considering typical motion scenarios... If the influence of the disturbance term is minimal, then the translational and rotational dynamic equations of the center of mass with the disturbance term added are as follows:

[0138] ;

[0139] ;

[0140] S2. Design translational perturbation observers and rotational perturbation observers independently for the translational dynamic equation and the rotational dynamic equation of the center of mass, respectively. By constructing a sliding mode surface and solving the nonlinear dynamic update law, the estimated values ​​of translational perturbation acceleration and rotational perturbation angular acceleration are output in real time.

[0141] In this step,

[0142] 1) The process of designing a translational perturbation observer includes:

[0143] The nominal translational acceleration of the center of mass is:

[0144] ;

[0145] The sliding surface is constructed as follows:

[0146] ;

[0147] in, To translate and observe the sliding surface, The location of the center of mass. For the estimated location, For the velocity of the center of mass, It is an estimate of speed. This is the weighting coefficient for the position error of the sliding surface;

[0148] The derivative of the position estimate is:

[0149] ;

[0150] The derivative of the velocity estimate is:

[0151] ;

[0152] in, To approximate the internal state of the disturbance, they are designed as follows:

[0153] ;

[0154] ;

[0155] in, For the observer position channel gain, For the observer velocity channel gain, For component-wise sign functions;

[0156] Let the position and velocity estimation errors be:

[0157] ;

[0158] ;

[0159] Therefore, the error derivative and the sliding surface derivative are:

[0160] ;

[0161] ;

[0162] ;

[0163] Among them, translational interference force Bounded, that is ,in It is an unknown constant.

[0164] 2) The process of designing a rotating perturbation observer includes:

[0165] The nominal angular acceleration of the center of mass is:

[0166] ;

[0167] Constructing a rotating sliding surface:

[0168] ;

[0169] in, To rotate the observation sliding surface, Angular velocity, For angular velocity estimation, For small-angle attitude vectors, For attitude estimation, This is the weighting coefficient for the position error of the rotating sliding surface;

[0170] The attitude estimation derivative is:

[0171] ;

[0172] The derivative of the estimated angular velocity is:

[0173] ;

[0174] The internal state is:

[0175] ;

[0176] ;

[0177] in, For observer gain;

[0178] Let the attitude estimation error be:

[0179] ;

[0180] The angular velocity estimation error is:

[0181] ;

[0182] The derivative of the error estimate is:

[0183] ;

[0184] ;

[0185] The derivative of the sliding surface is:

[0186] ;

[0187] Among them, the rotational disturbance torque Bounded, that is ,in This is an unknown constant.

[0188] S3. In the support phase, the output disturbance acceleration and disturbance angular acceleration estimates are used as feedforward compensation terms and injected into the single rigid body motion-dynamic model of the quadruped robot after linearization and discretization. A prediction model with disturbance compensation is constructed. The model prediction controller then solves the rolling time domain optimization problem based on the prediction model to generate the optimal foot contact force sequence, which is mapped to the support-related joint torque command through the foot Jacobian matrix.

[0189] During the swing phase, a virtual control force is generated by a spring-damped virtual model controller. The virtual control force is based on the foot position and velocity tracking error and mapped to the swing joint torque command through the leg Jacobian matrix to drive the swing leg to track the preset Bezier trajectory.

[0190] In this step, the process of obtaining the support-related joint torque command includes:

[0191] The single-rigid-body motion-dynamic model of the quadruped robot is linearized and discretized. The estimated values ​​of translational perturbation acceleration and rotational perturbation angular acceleration are used as inputs to the model, forming the discrete state-space equations for prediction.

[0192] ;

[0193] The discrete state-space equations are obtained by injecting the perturbation observations into the quadruped robot's dynamic equations, which are as follows:

[0194] ;

[0195] in,

[0196] ;

[0197] ;

[0198] ;

[0199] ;

[0200] ;

[0201] ;

[0202] The location of the center of mass. For the velocity of the center of mass, For the ZYX Euler angles of the machine, For roll angle, The pitch angle, Yaw angle For volume coordinates, angular velocity For the first The ground reaction force at the foot of the supporting leg, The vector of gravitational acceleration. For the estimation of translational disturbance acceleration output by the observer, For estimating the rotational perturbation angular acceleration output by the observer, For the body mass, Let be the matrix of rotational inertia of the fuselage about its center of mass. For the first The end of the leg and foot, The location of the center of mass. It is a 3×3 identity matrix. To discretize the sampling time interval, Let be the transformation matrix from angular velocity to Euler angle derivative. Let ZYX be the Euler angles of the aircraft's attitude. The specific form is as follows:

[0203] ;

[0204] The design cost function is as follows:

[0205] ;

[0206] The constraints are:

[0207] ;

[0208] in, Indicates the predicted length. This represents the system state at time k+1. This represents the desired state at time k+1. Let Q represent the ground reaction force calculated at time k, Q be the diagonal positive semi-definite matrix of the state tracking error weights, and S be the diagonal positive semi-definite matrix of the ground reaction force regularization weights. , , This represents the components of the ground reaction force in the x, y, and z directions. These are the lower and upper limits of the joint torque vectors, respectively. is the coefficient of friction. Here is the torque constraint matrix. This is the force constraint matrix for the swinging leg.

[0209] The rolling optimization problem is defined based on the discrete state-space equation. The deviation between the future state and the reference trajectory is minimized, while the control input energy is constrained. Ground friction cone constraint, foot slip-free constraint and torque constraint are applied. The optimal foot force sequence is obtained by solving the quadratic programming problem.

[0210] Take the first control variable in the sequence and map it to the support-related joint torque command through the Jacobian matrix at the foot of each support leg:

[0211] ;

[0212] in, For the first Leg joint torque, For the foot Jacobian matrix, This is the rotation matrix for the machine's attitude.

[0213] In this step, the process of obtaining the swing-related joint torque command includes:

[0214] Design virtual model control force :

[0215] ;

[0216] in, These are the position and velocity gain matrices, respectively. and These are the center of mass position and velocity tracking errors, respectively. By adjusting the diagonal elements, the tracking stiffness and damping characteristics in each direction can be independently controlled to ensure the smoothness of trajectory tracking.

[0217] The virtual model is controlled by the Jacobian matrix of the legs. Mapped to swing-related joint torque commands.

[0218] S4. Based on the leg contact state detection results, a smooth switching logic is executed between the support phase and the swing phase to ensure that the generated support-related joint torque command and swing-related joint torque command are seamlessly connected, thereby achieving stable motion control throughout the entire gait cycle.

[0219] In this step, the smooth switching logic includes:

[0220] The system monitors the contact state of each leg of the quadruped robot in real time. When a leg switches from the swing phase to the support phase, the model predictive controller is activated for force control; when a leg switches from the support phase to the swing phase, the virtual model controller is activated for trajectory tracking control.

[0221] Design a smooth interpolation mechanism for the control quantity during the transition phase to avoid abrupt changes in the control quantity at the moment of switching, and ensure the continuity and stability of the control output throughout the entire gait cycle.

[0222] To demonstrate the effectiveness of the method described in this invention, the following simulation experiments were conducted:

[0223] Consider the walking control of a quadruped robot in a rugged gravel road environment. The simulation uses a distance walking length... The system state is defined as follows: The center of mass dynamics satisfies , The system at discrete times state vector ,in The location of the center of mass. For the velocity of the center of mass, The ZYX Euler angles represent the aircraft's attitude. Let be the angular velocity of the center of mass. The initial parameters are selected as follows: m, m / s, rad, rad / s. This represents the joint angles and angular velocities of the four legs. Since the initial joint angles of all legs are the same, we choose... , In addition, the initial contact mode Indicates whether the four legs are supported; external load. As an equivalent disturbance This is added to the centroid translation equations and takes effect when the system starts running; the perturbation observer begins operating at 6.5 s. The expected centroid trajectory is given by a polynomial: The trajectory of the swinging foot is a cubic Bézier curve: The following results can be obtained from the simulation results: Figure 5 This demonstrates that the centroid height tracks the desired height of the upper centroid within 1 second after the translational disturbance observer is applied, and remains stable while traversing complex gravel roads. This signifies the successful observation and compensation of disturbances from a 2kg load applied to the fuselage and disturbances from complex road surfaces. Furthermore, Figure 6 , Figure 7 The estimated changes in the observed translational perturbation acceleration and the estimated changes in the rotational perturbation angular acceleration are described.

[0224] Furthermore, this embodiment also includes a quadruped robot composite control system based on a superspiral sliding mode anti-interference model, used to implement the above-mentioned quadruped robot composite control method based on a superspiral sliding mode anti-interference model, which includes a modeling module, a disturbance observer design and observation module, a support phase control module, a swing phase control module, and a logic switching module.

[0225] The system includes the following modules: a modeling module for constructing a single-rigid-body motion-dynamic model of a quadruped robot with unknown perturbations; a perturbation observer design and observation module for designing translational and rotational perturbation observers, outputting estimates of translational perturbation acceleration and rotational perturbation angular acceleration, and connecting to the modeling module to obtain model parameters; a support phase control module for executing motion control during the support phase phase, and connecting to the perturbation observer design module to obtain estimates of perturbation acceleration and angular acceleration; a swing phase control module for executing motion control during the swing phase phase; and a logic switching module for executing smooth switching logic to coordinate the working states of the support phase control module and the swing phase control module.

[0226] More specifically, the supporting phase control module includes:

[0227] The prediction model building unit is used to build a prediction model by using the disturbance estimate as a feedforward compensation term; the rolling optimization solution unit is used to define and solve a quadratic programming optimization problem with physical constraints; and the force mapping unit is used to map the optimal foot force to the support-related joint torque command through the foot Jacobian matrix.

[0228] More specifically, the disturbance observer design and observation module include:

[0229] The disturbance observer design unit is used to design translational disturbance observers and rotational disturbance observers; the observation unit is used to output estimates of translational disturbance acceleration and rotational disturbance angular acceleration using translational disturbance observers and rotational disturbance observers.

[0230] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, any changes made in accordance with the shape and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A composite control method for a quadruped robot based on a superspiral sliding mode anti-interference model, characterized in that, include: A single rigid body motion-dynamic model of a quadruped robot with unknown disturbances is constructed. The model includes the translational dynamic equation of the center of mass and the rotational dynamic equation of the center of mass. The equivalent external disturbance force and equivalent disturbance torque are explicitly introduced into the equations to provide a predictive model basis with disturbances for the model predictive controller. A translational perturbation observer and a rotational perturbation observer are designed independently for the translational dynamic equation and the rotational dynamic equation of the center of mass, respectively. By constructing a sliding surface and solving the nonlinear dynamic update law, the estimated values ​​of translational perturbation acceleration and rotational perturbation angular acceleration are output in real time. During the support phase, the output disturbance acceleration and disturbance angular acceleration estimates are used as feedforward compensation terms and injected into the single rigid body motion-dynamic model of the quadruped robot after linearization and discretization. A prediction model with disturbance compensation is constructed. The model prediction controller then solves the rolling time domain optimization problem based on the prediction model to generate the optimal foot contact force sequence, which is mapped to the support-related joint torque command through the foot Jacobian matrix. During the swing phase, a virtual control force is generated by a spring-damped virtual model controller. The virtual control force is based on the foot position and velocity tracking error and mapped to the swing joint torque command through the leg Jacobian matrix to drive the swing leg to track the preset Bezier trajectory. Based on the leg contact state detection results, a smooth switching logic is executed between the support phase and the swing phase to ensure that the generated support-related joint torque command and the swing-related joint torque command are seamlessly connected, thereby achieving stable motion control throughout the entire gait cycle.

2. The quadruped robot composite control method based on the superspiral sliding mode anti-interference model according to claim 1, characterized in that, The equations of motion for the center of mass translation are as follows: ； in, For the overall quality of the robot; Let be the linear acceleration vector of the center of mass in the world coordinate system; This represents the current number of supporting legs in contact with the ground. For the first The ground reaction force vector at the foot of the supporting leg; This is the vector of gravitational acceleration; The translational disturbance force characterizes the uncertainty in the translational direction, including the impact reaction force caused by ground protrusion, lateral wind load, and inertial force caused by sudden load changes; The dynamic equation for the rotation of the center of mass is as follows: ； in, Let be the moment of inertia matrix of the fuselage about its center of mass; This is the angular velocity vector of the fuselage relative to the world coordinate system; Let be the position vector of the centroid in the world coordinate system; For the first The position vector of the foot end of the supporting leg; The rotational disturbance torque characterizes the uncertainty in the direction of rotation, including torque fluctuations caused by deviations in the foot contact position, fuselage attitude disturbances, and changes in the joint friction coefficient.

3. The quadruped robot composite control method based on the super-helical sliding mode anti-interference model according to claim 2, characterized in that, The process of designing a translational perturbation observer includes: The nominal translational acceleration of the center of mass is: ； The sliding surface is constructed as follows: ； in, To translate and observe the sliding surface, The location of the center of mass. For the estimated location, For the velocity of the center of mass, To estimate the speed, This is the weighting coefficient for the position error of the sliding surface; The derivative of the position estimate is: ； The derivative of the velocity estimate is: ； in, To approximate the internal state of the disturbance, they are designed as follows: ； ； in, For the observer position channel gain, For the observer velocity channel gain, For component-wise sign functions; Let the position and velocity estimation errors be: ； ； Therefore, the error derivative and the sliding surface derivative are: ； ； ； Among them, translational interference force Bounded, that is ,in It is an unknown constant.

4. The quadruped robot composite control method based on the superspiral sliding mode anti-interference model according to claim 2, characterized in that, The process of designing a rotating perturbation observer includes: The nominal angular acceleration of the center of mass is: ； Constructing a rotating sliding surface: ； in, To rotate the observation sliding surface, Angular velocity, For angular velocity estimation, For small-angle attitude vectors, For attitude estimation, This is the weighting coefficient for the position error of the rotating sliding surface; The attitude estimation derivative is: ； The derivative of the estimated angular velocity is: ； The internal state is: ； ； in, For observer gain; Let the attitude estimation error be: ； The angular velocity estimation error is: ； The derivative of the error estimate is: ； ； The derivative of the sliding surface is: ； Among them, the rotational disturbance torque Bounded, that is ,in This is an unknown constant.

5. The quadruped robot composite control method based on the superspiral sliding mode anti-interference model according to claim 1, characterized in that, The process of obtaining the support-related joint torque command includes: The single-rigid-body motion-dynamic model of the quadruped robot is linearized and discretized. The estimated values ​​of translational perturbation acceleration and rotational perturbation angular acceleration are used as inputs to the model, forming the discrete state-space equations for prediction. ； in, for The state vector at any given moment includes attitude angle, position, angular velocity, and linear velocity; for The ground reaction force calculated at all times; The system matrix after linearization and precise discretization; , This is the vector of gravitational acceleration; ; This is an estimate of the translational disturbance acceleration; This is an estimate of the angular acceleration of the rotational disturbance; The rolling optimization problem is defined based on the discrete state-space equation. The deviation between the future state and the reference trajectory is minimized, while the control input energy is constrained. Ground friction cone constraint, foot slip-free constraint and torque constraint are applied. The optimal foot force sequence is obtained by solving the quadratic programming problem. Take the first control variable in the sequence and map it to the support-related joint torque command through the Jacobian matrix at the foot of each support leg: ； in, For the first Leg joint torque, For the foot Jacobian matrix, This is the rotation matrix for the machine's attitude.

6. The quadruped robot composite control method based on the superspiral sliding mode anti-interference model according to claim 1, characterized in that, The process of obtaining the swing-related joint torque command includes: Design virtual model control force : ； in, These are the position and velocity gain matrices, respectively. and These are the center of mass position and velocity tracking errors, respectively. By adjusting the diagonal elements, the tracking stiffness and damping characteristics in each direction can be independently controlled to ensure the smoothness of trajectory tracking. The virtual model is controlled by the Jacobian matrix of the legs. Mapped to swing-related joint torque commands.

7. The composite control method for a quadruped robot based on a super-helical sliding mode anti-interference model according to claim 1, characterized in that, The smooth switching logic includes: The system monitors the contact state of each leg of the quadruped robot in real time. When a leg switches from the swing phase to the support phase, the model predictive controller is activated for force control; when a leg switches from the support phase to the swing phase, the virtual model controller is activated for trajectory tracking control. Design a smooth interpolation mechanism for the control quantity during the transition phase to avoid abrupt changes in the control quantity at the moment of switching, and ensure the continuity and stability of the control output throughout the entire gait cycle.

8. A composite control system for a quadruped robot based on a super-helical sliding mode anti-interference model, characterized in that, The method for implementing the quadruped robot composite control method based on the super-helical sliding mode anti-interference model as described in any one of claims 1-7 includes a modeling module, a disturbance observer design and observation module, a support phase control module, a swing phase control module, and a logic switching module. The modeling module is used to construct a single rigid body motion-dynamic model of a quadruped robot with unknown perturbations; The disturbance observer design and observation module is used to design translational disturbance observers and rotational disturbance observers, output estimated values ​​of translational disturbance acceleration and rotational disturbance angular acceleration, and connect with the modeling module to obtain model parameters; The support phase control module is used to perform motion control during the support phase phase and is connected to the disturbance observer design module to obtain estimated values ​​of disturbance acceleration and disturbance angular acceleration. The swing phase control module is used to perform motion control during the swing phase phase. The logic switching module is used to execute smooth switching logic and coordinate the working states of the support phase control module and the swing phase control module.

9. The quadruped robot composite control system based on the super-spiral sliding mode anti-interference model according to claim 8, characterized in that, The supporting phase control module includes: The prediction model building unit is used to construct a prediction model by using the disturbance estimate as a feedforward compensation term. The rolling optimization solver unit is used to define and solve quadratic programming optimization problems with physical constraints. The force mapping unit is used to map the optimal foot force into the support-related joint torque command through the foot Jacobian matrix.

10. The quadruped robot composite control system based on the super-helical sliding mode anti-interference model according to claim 8, characterized in that, The disturbance observer design and observation module include: The disturbance observer design unit is used to design translational disturbance observers and rotational disturbance observers; The observation unit is used to output estimates of translational perturbation acceleration and rotational perturbation angular acceleration using translational perturbation observers and rotational perturbation observers.

Images