Real-time load estimation and adaptive control method and system for a biped robot

Abstract

The present application belongs to the technical field of robots, and discloses a real-time load estimation and adaptive control method and system for a double-wheel-foot robot. The method comprises the following steps: establishing a mapping relationship between joint output torque and equivalent contact force at the end of the leg in the sagittal plane of the robot, and using the mapping relationship to solve the vertical support force of the left and right legs of the double-wheel-foot robot; calculating the increments of the total vertical support force and the moment respectively; using the total vertical support force to calculate the mass and coordinates of the load, and using the coordinates of the load to update the length of the equivalent inverted pendulum and the moment of inertia of the center of mass of the robot. Through the present application, the problem of mismatching of the robot dynamics model, decline of control performance, and even instability caused by changes in the load is solved.

Description

Technical Field

[0001] This invention belongs to the field of robot-related technology, and more specifically, relates to a real-time load estimation and adaptive control method and system for a bipedal robot. Background Technology

[0002] Bicycle-legged robots combine the efficiency of wheeled mobility with the terrain adaptability of legged mechanisms, making them promising for applications in material handling, express delivery, indoor services, and human-robot collaboration. In these scenarios, the robot may bear external loads with constantly changing mass and position. These load variations cause changes in the robot system's mass distribution and inertial parameters, consequently affecting its posture, equilibrium point position, and closed-loop stability. Therefore, obtaining real-time load mass and position information and using it to update the control model is a crucial problem that needs to be solved for the stable operation of bicycle-legged robots.

[0003] In the fields of wheeled inverted pendulum robots, wheeled humanoid robots, and biwheeled legged robots, Jeong and Takahashi established the design and motion equations for a wheeled inverted pendulum robot based on the I-PENTAR assistant robot; Nasir and Mahmoud further conducted robust motion control research on similar platforms; Zafar et al. expressed the mass model error as the centroid error and combined robust control and online learning to update the mass model; Baek et al. estimated the new equilibrium point for a wheeled humanoid robot under unknown loads by using position and pitch angle time series; Yu et al., for biwheeled legged robots, incorporated the centroid state and equivalent disturbance into the model predictive control framework for compensation. These schemes can mitigate the impact of model mismatch on equilibrium control, but their main objectives are equilibrium point correction, centroid error compensation, or equivalent disturbance suppression. They typically do not directly output explicit physical parameters such as the external load mass and the load's longitudinal and lateral positions in the body coordinate system.

[0004] For bipedal robots used in applications such as material handling, simply performing balance point correction or equivalent disturbance compensation is insufficient to meet the requirements of load state recognition, load-bearing capacity assessment, and online updating of dynamic parameters. Furthermore, acquiring support force information through additional force / torque sensors or foot force sensors increases robot hardware costs, structural complexity, and maintenance difficulty. Therefore, current technology still lacks a load estimation method that can obtain the load mass and its position in the body coordinate system in real time using a closed-loop solution, without requiring additional foot force sensors or active posture excitation, and utilizing the robot's existing joint angle information, joint output torque information, and body posture information. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a real-time load estimation and adaptive control method and system for a bipedal robot, which solves the problems of robot dynamics model mismatch, control performance degradation and even instability caused by load changes.

[0006] To achieve the above objectives, according to one aspect of the present invention, a real-time load estimation and adaptive control method for a bipedal robot is provided, the method comprising the following steps: A mapping relationship between the joint output torque and the equivalent contact force at the end of the leg is established in the sagittal plane of the robot. Using this mapping relationship and the current posture of the robot, the components of the wheel-ground contact reaction forces acting on the left and right wheels of the bipedal robot in the vertical axis direction of the world coordinate system are solved, and these are used as the left vertical support force and the right vertical support force. The increments of the total vertical support force and the torque are calculated using the left vertical support force and the right vertical support force, respectively. The mass and coordinates of the load are calculated using the total vertical support force, and the robot's center of mass equivalent inverted pendulum length and moment of inertia are updated accordingly.

[0007] More preferably, the calculation formulas for the left vertical support force and the right vertical support force are as follows:

[0008] in, For the leg side index, the value is... or , respectively representing the left leg or the right leg; For the first The joint angle vector of the lateral leg; For the first The joint output torque vector of the lateral leg; For vertical support force, For the first lateral leg joint angle vector A defined Jacobian matrix for the legs.

[0009] More preferably, the formulas for calculating the increments of the total vertical support force and moment are as follows:

[0010]

[0011]

[0012]

[0013] in, This represents the total vertical support force increment caused by external loads. and These represent the vertical support force increments for the left and right legs, respectively. and These represent the vertical support forces of the left and right wheels, respectively, in the robot's current state. and These are the vertical support forces of the left and right wheels under no-load conditions, corresponding to the robot's current leg configuration and body posture. This represents the increase in pitching moment caused by external loads. For the robot's current state around the body coordinate system The resultant torque of the shaft; To determine the robot's position in a no-load state relative to its current leg configuration and body posture, orbiting the body coordinate system. The resultant torque of the shaft.

[0014] More preferably, the formulas for calculating the mass and coordinates of the load are as follows:

[0015]

[0016]

[0017] in, For the quality of the load, subscript Indicates load; This represents the total vertical support force increment caused by external loads. This represents the magnitude of gravitational acceleration. For the load in the body coordinate system of The longitudinal coordinate along the axis; For the coordinate system around the machine body caused by external load of The pitching moment increment of the shaft; For the load in the body coordinate system of The horizontal coordinate along the axis; The equivalent distance between the lines of action of the ground contact reaction forces of the left and right wheels in the transverse direction of the aircraft body; and These represent the vertical support force increments for the right and left wheel sections, respectively.

[0018] More preferably, the formula for calculating the total centroid is as follows:

[0019] in, It is the center of mass of the robot under load. For the robot's no-load mass, The position of the robot's center of mass without load. These are the three-dimensional coordinates of the load. It is the quality of the load.

[0020] More preferably, the equivalent inverted pendulum length The calculation formula is as follows:

[0021] in, The equivalent length of the inverted pendulum, The overall center of mass of the system consisting of the robot and the load in the body coordinate system after loading is... The position vector below, The midpoint of the line connecting the centers of the left and right drive wheel axles of the bipedal robot in the body coordinate system The position vector below, To determine the body coordinate system based on the robot's current body posture. The displacement vector represented in the figure Convert to world coordinate system The rotation matrix of the displacement vector is represented in the figure, where, Indicates the body coordinate system. This represents a world coordinate system that is fixed relative to the ground.

[0022] More preferably, the formula for calculating the system's moment of inertia is as follows:

[0023] in, The total centroid of the system consisting of the robot and the load after the load is applied, relative to the updated system. The moment of inertia matrix; The unloaded robot body relative to its unloaded center of mass, corresponding to the robot's current leg configuration. The moment of inertia matrix; The unloaded mass of the robot body; For load quality; To determine the unloaded center of mass of the robot body in the body coordinate system corresponding to the current leg configuration of the robot. The position vector below; The equivalent center of mass of the load in the body coordinate system The position vector below; The overall center of mass of the system consisting of the robot and the load in the body coordinate system after loading is... The position vector below; where, Indicates the body coordinate system; It is a 3×3 identity matrix.

[0024] According to another aspect of the present invention, a real-time load estimation and adaptive control system for a bipedal robot is provided, the system including an actuator for implementing the aforementioned real-time load estimation and adaptive control method for a bipedal robot.

[0025] According to another aspect of the present invention, a computer storage medium is provided having a computer program stored thereon, the program being used to implement the above-described method for real-time load estimation and adaptive control of a bipedal robot.

[0026] According to another aspect of the present invention, a computer program product is provided, which includes a computer program that, when executed by a processor, implements the above-described method for real-time load estimation and adaptive control of a bipedal robot.

[0027] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art: 1. The method provided by this invention, when a bipedal robot is in a state of effective support on both sides, calculates the joint output torque as the vertical support force of the left and right legs based on the leg geometry and Jacobian mapping relationship, calculates the increment of the total support force and torque, and finally updates the model parameters or control gain of the balance controller. This ensures that the model parameters used by the balance controller match the robot's current load state, or that the control gain is adjusted accordingly with load changes, thereby reducing the impact of model mismatch on closed-loop control and effectively solving the problems of robot dynamic model mismatch, control performance degradation, and even instability caused by load changes.

[0028] 2. This invention first determines the corresponding unloaded support force and pitch moment benchmarks based on the current leg configuration and body posture. It then calculates the current left and right vertical support forces using joint angles, joint output torque, and the leg Jacobian relationship. The increments in support force and torque caused by external loads are obtained through subtraction. When the quasi-static estimation conditions are met, the load mass and its position are estimated in real time based on the increments, thereby updating the system's total mass, center of mass position, equivalent inverted pendulum length, and moment of inertia, among other dynamic parameters. The balance controller then adjusts the control model or feedback gain based on the updated dynamic parameters. For example, based on the updated system total mass and equivalent inverted pendulum length, an LQR feedback gain matrix matching the current load state is interpolated. This reduces the parameter deviation between the dynamic model used by the controller and the robot's current load state, allowing the control action to adjust with load changes, thus reducing the risk of attitude adjustment performance degradation and closed-loop instability caused by model mismatch. When the quasi-static estimation conditions are not met, the reliable estimate from the previous moment is maintained or the current estimate weight is reduced to avoid unreliable estimates causing erroneous updates to control parameters.

[0029] 3. The support force in the calculation method of this invention is obtained by solving the joint output torque and leg geometry, rather than by direct measurement by an additional foot force sensor. At the same time, it does not require active posture excitation, which can reduce the impact of model mismatch caused by load changes on the balance control of the bi-wheeled legged robot.

[0030] 4. Based on the increments of total vertical support force, pitching moment, and left and right vertical support forces, this invention estimates the load mass, longitudinal position, and lateral position, respectively. The resulting load parameters have clear physical meanings and can provide parameter basis for load state identification and online updates of the system's total mass, total center of mass position, equivalent inverted pendulum length, and moment of inertia.

[0031] 5. This invention determines the corresponding no-load support force and pitch moment benchmarks based on the current leg configuration and body posture. Combined with quasi-static condition determination, load threshold determination, signal filtering, and reliable estimation value preservation mechanisms, it helps to reduce load misjudgment or parameter erroneous updates caused by posture changes, rapid movement, wheel slippage, and measurement disturbances, thereby improving the reliability and continuity of load estimation results. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the mechanical structure model of a single leg of a robot constructed according to a preferred embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram of the overall structural model of a bipedal robot constructed according to a preferred embodiment of the present invention.

[0034] Figure 3 This is a force analysis diagram of the robot body constructed according to a preferred embodiment of the present invention under static equilibrium conditions.

[0035] Figure 4 This is a schematic diagram of force analysis in the sagittal plane of a robot constructed according to a preferred embodiment of the present invention.

[0036] Figure 5 This is a schematic diagram of the forces acting on the body in the sagittal plane according to a preferred embodiment of the present invention.

[0037] Figure 6 This is a schematic diagram of the forces acting on the body in the coronal plane according to a preferred embodiment of the present invention.

[0038] Figure 7 This is a block diagram of a hierarchical control system for a bipedal robot constructed according to a preferred embodiment of the present invention.

[0039] Figure 8 This is a flowchart of a load estimation and adaptive control method constructed according to a preferred embodiment of the present invention.

[0040] Figure 9 This is a schematic diagram comparing the no-load state and the loaded state constructed according to a preferred embodiment of the present invention, wherein (a) shows that the left and right support forces of the robot are approximately symmetrical in the no-load state, and (b) shows the left and right support force increment caused by the load and the load position relationship in the loaded state. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0042] A preferred embodiment of the biwheeled legged robot single-leg structure of the present invention is as follows: Figure 1 As shown, the overall structure of the robot is as follows: Figure 2 As shown, the bipedal robot includes a body, a left leg, a right leg, and wheels respectively located at the ends of the left and right legs. In this embodiment, the body refers to the main frame or shell structure of the bipedal robot located between the left and right legs, used to connect the left and right legs and install the inertial measurement unit, excluding the left and right legs and the wheels at the ends of the two legs. In this embodiment, the robot body refers to the entire robot consisting of the body, left leg, right leg, and wheels at the ends of the two legs, excluding external loads; the external load refers to an object additionally carried by the robot, not included in the robot's no-load body parameters; the no-load state refers to the state where the normal components of the robot body are installed but not carrying the external load. The left and right legs are respectively connected to the left and right sides of the body, and the structures of the left and right legs can adopt the same or substantially the same configuration so as to provide support force to the body under the effective support state on both sides. In this embodiment, the effective support state on both sides means that both the left and right wheels maintain effective contact with the ground and can transmit support force to the robot respectively. Each leg includes a thigh link, a lower leg link, a hip joint, a knee joint, and a wheel. One end of the thigh link is connected to the body via the hip joint, and the other end of the thigh link is connected to the lower leg link via the knee joint. The distal end of the lower leg link is connected to the wheel.

[0043] In this embodiment, the hip joint has flexion-extension degrees of freedom in the sagittal plane, used to adjust the posture of the leg relative to the body; the knee joint has flexion-extension degrees of freedom in the sagittal plane, used to adjust the angle between the thigh link and the lower leg link, thereby changing the equivalent length of the leg and the position of the wheels relative to the body. In this embodiment, the leg configuration refers to the geometric posture of the leg determined by the hip joint angle and knee joint angle of the left and right legs. The wheels may include hub motor-driven drive wheels for realizing the robot's rolling movement and forward and backward motion control on the ground. Through the coordinated action of the hip joints, knee joints, and wheels of the left and right legs, the robot can achieve height adjustment, posture balance, and movement control while maintaining effective support on both sides.

[0044] To reduce the distal rotational inertia of the leg and improve its dynamic response, in one embodiment, a knee joint drive module is positioned near the proximal thigh, and this module can be arranged adjacent to or coaxially with the hip joint drive module. The output torque of the knee joint drive module is transmitted to the lower leg link via a parallelogram transmission mechanism, causing the lower leg link to flex and extend relative to the thigh link. This parallelogram transmission mechanism enables the long-range transmission of the knee joint drive torque without altering the fundamental principle of calculating the support force based on the joint output torque and leg geometry in this invention. The wheel joint is located at the distal end of the lower leg.

[0045] An inertial measurement unit (IMU) is installed on the fuselage to acquire attitude and motion information such as pitch angle, roll angle, angular velocity, and / or linear acceleration. Joint encoders are installed at the hip and knee joints to acquire the corresponding joint angles. The current information of each joint drive motor can be combined with the motor torque constant to calculate the corresponding joint output torque. In this embodiment, the joint output torque includes the output torque of the left hip joint, the left knee joint, the right hip joint, and the right knee joint. The aforementioned joint angles, joint output torques, and fuselage attitude information together constitute the input information for the present invention to calculate support force, compare no-load states, and estimate load parameters in a closed-loop manner.

[0046] It should be noted that, Figure 1 and Figure 2 The structure shown is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. In other embodiments, the specific length of the leg link, the installation position of the hip and knee joint drive modules, the form of the transmission mechanism, the wheel drive method, and the sensor arrangement can all be adjusted according to the specific robot platform. As long as the robot can obtain the joint angles of the left and right legs, the joint output torque, and the body posture information under effective support on both sides, and can calculate the support force of the left and right legs according to the leg geometry or Jacobian mapping relationship, it can be applied to the real-time load estimation and adaptive control method in this embodiment.

[0047] In one embodiment of the present invention, the so-called vertical support force refers to the wheel-ground contact reaction force exerted by the ground on the left and right wheels of the robot at the wheel-ground contact point when the robot wheels maintain effective contact with the ground, as defined in the world coordinate system. vertical axis Components in direction, where The positive direction of the axis is opposite to the direction of gravity; the reaction force from the wheel-to-ground contact is transmitted to the robot body through the corresponding wheels and legs. The vertical support forces on the left and right sides are denoted as follows: and The total vertical support force is denoted as ,and . Figure 3 The supporting forces exerted by the left and right legs on the fuselage represent the equivalent effect of the aforementioned wheel-to-ground contact reaction forces transmitted through the wheels and legs to the fuselage interface. When calculating pitching moments, all forces and moments are converted to the same coordinate system and the same reference point. The wheel-to-ground contact reaction force is the resultant contact force exerted by the ground through the wheel-to-ground contact area on the corresponding wheel, and its direction adopts the direction of the ground's action on the wheel.

[0048] like Figure 3 As shown, when the robot is effectively supported on both sides and meets the quasi-static conditions, it can be positioned in the body coordinate system. The following is a force analysis of the robot body. The robot's own weight and the external load gravity act vertically along the world coordinate system. The left and right legs exert supporting forces on the robot body and around the robot's coordinate system, respectively. The equivalent torque of the shaft; the arrows in the diagram are used to indicate the direction of the force and torque, not their magnitude. Figure 3 The mass symbol representing the robot's own weight corresponds to the robot's unloaded mass. The equivalent torques of the left and right legs correspond to respectively and Under quasi-static conditions, the resultant force and resultant moment on the body satisfy the equilibrium relationship. Therefore, the difference between the current left and right support forces and pitching moments and the unloaded reference quantities under the same leg configuration and body posture can be obtained to obtain the increase in support force and moment caused by external load.

[0049] like Figure 8 As shown, Figure 8This describes the data acquisition, no-load benchmark comparison, confidence condition determination, load and dynamic parameter updates, and controller update process within an online estimation and control cycle. Before online operation, the system establishes a no-load model or lookup table corresponding to the leg configuration and body posture according to step S1. Within each online cycle, the data acquisition and preprocessing module first acquires joint angles, joint output torques, and body posture information, and performs filtering, amplitude limiting, or outlier removal. The support force and no-load state comparison and calculation module then obtains the current left and right support forces and torques, and calculates the difference with the corresponding no-load benchmark quantities. The quasi-static estimation conditions are met, and... When the load exceeds a preset load threshold, the system performs closed-form estimation of the load parameters in step S2 and dynamic parameter update in step S3; if the quasi-static estimation condition is not met, the system retains the reliable estimation result from the previous moment or reduces the weight of the current result in the filtering fusion; if the quasi-static condition is met but... When the load is less than or equal to the preset load threshold, the load position estimation result is not updated; the previous reliable estimate is maintained or the state is determined to be without effective load. Subsequently, the controller uses the current reliable dynamic parameters to update the model parameters or control gain according to step S4, and the balance control module generates the control quantity for the current control cycle. Figure 8 The word "end" in the text indicates the end of a control cycle, and the data acquisition and processing will be executed again in the next control cycle.

[0050] S1: Establishing the Correspondence between No-Load States. In the no-load state of the robot, the correspondence between leg configurations and body postures and the left-right vertical support forces and pitch moments is established through calculations using a no-load dynamics model or through offline calibration. In one implementation, the bipedal robot traverses multiple combinations of leg heights, leg configurations, and body postures in the no-load state, recording the corresponding joint angles, joint output torques, body postures, left-right vertical support forces, and pitch moments, forming a lookup table of support forces and moments in the no-load state indexed by leg configurations and body postures.

[0051] The system currently acquires load status information. It reads the feedback signals from the left hip joint encoder, left knee joint encoder, right hip joint encoder, and right knee joint encoder at preset frequencies to obtain the left hip joint angle, left knee joint angle, right hip joint angle, and right knee joint angle. Simultaneously, the system reads the current feedback from the left hip joint motor, left knee joint motor, right hip joint motor, and right knee joint motor, and calculates the output torque of the left hip joint, left knee joint, right hip joint, and right knee joint based on the corresponding motor torque constants. Furthermore, the system reads the angular velocity, acceleration, and / or attitude data output by the inertial measurement unit, and obtains the body pitch angle, roll angle, and / or angular velocity information through attitude calculation. To suppress high-frequency noise, the system can perform low-pass filtering, amplitude limiting, and outlier removal processing on the joint angles, joint output torques, and body attitude information. In this embodiment, the preset frequency can be 1000Hz.

[0052] The system calculates the quantities corresponding to the no-load state, the current support force, and the increment. First, based on the current leg configuration and body posture, the system obtains the left and right vertical support forces and pitch moments corresponding to the current leg configuration and body posture under no-load conditions from the no-load model or no-load lookup table established in step S1 through nearest neighbor lookup, linear interpolation, or multidimensional interpolation. Then, based on the joint angles and joint output torques under the current loaded state, the system calculates the left and right leg vertical support forces and the current pitch moment. Both the no-load state quantities and the current state quantities are projected onto the vertical axis of the world coordinate system according to the same posture conversion rules. Direction, and adopt The positive direction of the axis is positive; finally, subtract the corresponding left and right vertical support forces and pitching moments under the unloaded state from the current load-bearing state's left and right vertical support forces and pitching moments, respectively, to calculate the support force increment and moment increment caused by the external load.

[0053] Furthermore, since the left and right vertical support forces and pitch moments corresponding to the no-load state are determined based on the current leg configuration and the current body posture obtained by the inertial measurement unit, rather than using the no-load result under a fixed upright posture as a unified benchmark, when the body tilt angle changes when the external load causes the robot to be balanced, this tilt angle change has already been used as an input to determine the no-load state quantities, which can avoid misjudging the body posture change itself as an increase in support force or torque caused by the external load.

[0054] like Figure 9 As shown, Figure 9Figure (a) shows a no-load reference example when the left and right structures, leg configurations and body postures are approximately symmetrical. At this time, the left and right no-load vertical support forces are approximately symmetrical. This approximately symmetrical aspect is only an illustrated embodiment and is not a necessary condition for the validity of this method. The actual no-load reference quantity used is still determined by the model or lookup table established in step S1 based on the current leg configuration and body posture. Figure 9 Figure (b) shows the state of the robot after it has borne an external load, the load mass of which is... The equivalent centroid of the load is located at... ,in , and These represent the equivalent centroid of the load in the body coordinate system. The corresponding axial components. The current left and right vertical support forces and pitch moments are compared with the unloaded reference values ​​under the same configuration and attitude to obtain... , , and And based on this, estimate , and .

[0055] Specifically, such as Figure 4 As shown, in the robot's sagittal plane, based on the principle of virtual work or the Jacobian mapping relationship of the legs, a mapping relationship between the joint output torque and the equivalent contact force at the leg end is established. For the left and right legs, the corresponding equivalent contact force vectors at the leg end are calculated based on the corresponding hip joint output torque, knee joint output torque, and joint angles, respectively. When the force vector is in the leg sagittal plane or the body coordinate system... When representing the orientation, the direction is converted to the world coordinate system based on the current body attitude. And take the vertical axis The components in the direction yield the vertical support force on the left side. and right vertical support force The direction of the ground's action on the wheel is taken as the positive direction of the wheel-ground contact reaction force. The positive direction of the axis is opposite to the direction of gravity. Under the condition of satisfying the quasi-static estimation conditions and neglecting or compensating for the inertial effects of the legs and wheels, the global vertical component of the equivalent contact force at the leg end corresponds to the global vertical component of the corresponding wheel-to-ground contact reaction force. In one embodiment, the mapping relationship can be expressed as:

[0056] in, It indicates the left or right leg.

[0057] Pitch moment in the current state It can be determined based on the supporting force exerted by the left and right legs on the body and their relationship to the origin of the body coordinate system. The point of application is calculated, or converted based on the output torque of the left and right hip joints and the geometric relationship of the legs. In one embodiment, Represents the coordinate system around the body The resultant torque of the axis has its positive direction consistent with the definition of the body coordinate system. Specifically, it can be expressed as:

[0058] in, , The position vector of the point of application of the equivalent forces of the left and right legs at the interface of the body relative to the origin of the body coordinate system; , To convert to the action point of the body interface and in the body coordinate system The following represents the equivalent forces acting on the left and right legs; , Equivalent torques for the left and right hips or legs at the same reference point. Here... and Used for calculating the resultant moment of the machine body, it is not equivalent to directly using the world coordinate system. Scalar vertical support force below and When performing cross products and summations, all position vectors, forces, and moments are in the body coordinate system. And represented under the same reference point.

[0059] Therefore, the increments of support force and torque are calculated as follows:

[0060]

[0061]

[0062]

[0063] S2: Closed-form estimation of load parameters. Before performing closed-form load estimation, the system determines whether the robot meets the quasi-static estimation conditions. The quasi-static estimation conditions include at least one of the following: body linear acceleration is less than a first threshold, body pitch rate is less than a second threshold, body roll rate is less than a third threshold, both left and right wheels are in effective contact, wheel-to-ground slippage is less than a fourth threshold, and joint output torque does not exceed a saturation threshold. When the robot meets the quasi-static estimation conditions, the system further determines whether the increase in total vertical support force is greater than a preset load threshold. When it is greater than the preset load threshold, the system calculates the load mass and position in real time through a closed-form relationship; when it is less than or equal to the threshold, the system does not update the load position estimation result, but maintains the previous reliable estimate value, or determines that the robot is in a state without effective load. When the quasi-static estimation conditions are not met, the system maintains the previous reliable load estimation result, or reduces the weight of the current estimation result in the filtering fusion.

[0064] like Figure 5 As shown, external load gravity The load acts on its equivalent center of mass and generates a pitch moment relative to the reference position in the body coordinate system. When the robot meets the quasi-static condition and has already completed coordinate and moment direction compensation based on the current body attitude, or when the body pitch angle meets the small-angle approximation condition, the increment of the pitch moment caused by the load satisfies its longitudinal position. ; combination ,available Therefore, the longitudinal position of the load can be determined by using the increase in pitching moment caused by the load and the increase in total vertical support force.

[0065] like Figure 6 As shown, the equivalent distance between the lines of action of the left and right wheel contact reaction forces on the transverse side of the aircraft is... The horizontal coordinate of the load equivalent centroid is After subtracting the unloaded baseline support force under the same leg configuration and body posture from the left and right vertical support forces under load, the corresponding term for the robot's body weight is eliminated, and there is... The vertical support forces and their increments used are all based on the vertical axis of the world coordinate system. The lateral component. When the roll angle of the aircraft meets the small-angle approximation condition, or when lateral geometric projection compensation has been completed according to the current attitude, the lateral load coordinate can be determined by the degree of asymmetry of the increment of the left and right vertical support forces, based on the moment balance relationship between the lines of action of the wheel-to-ground contact reaction forces on both sides and the mid-surface of the aircraft, specifically according to the following formula for calculating the lateral load coordinate. The positive direction of the lateral coordinate is based on the aforementioned aircraft coordinate system. The positive direction of the axis is defined by convention; the arrows in the diagram are only used to indicate the direction of force.

[0066] Accordingly, the load mass, longitudinal position, and lateral position can be calculated using the following formula:

[0067]

[0068]

[0069] in, The equivalent distance between the lines of action of the ground contact reaction forces of the left and right wheels in the transverse direction of the aircraft; aircraft coordinate system of righteousness The direction is pointing forward of the robot, and... The direction points to the right of the robot, which is... Direction along the height of the robot body; world coordinate system Relative to the ground, its positive The direction is vertically upward and opposite to the direction of gravity. Load height The equivalent center of mass of the load in the body coordinate system of Given the pre-defined coordinates along the axis, the load position vector is: .

[0070] S3: Dynamic parameters updated. The processor estimates the load mass in real time. and load location Update the overall system dynamic parameters: Total system mass:

[0071] Location of the system's overall centroid:

[0072] in, For the robot's no-load mass, This represents the position of the robot's center of mass without load.

[0073] Equivalent inverted pendulum length The updated position of the system's total centroid was calculated. The midpoint of the line connecting the centers of the left and right drive wheels of the bipedal robot. The distance is obtained. Among them, and All are in body coordinate system The position vector below, To determine the body coordinate system based on the robot's current body posture. The displacement vector represented in the figure Convert to world coordinate system The rotation matrix of the displacement vector is represented in the figure. Indicates the body coordinate system. This represents a world coordinate system fixed relative to the ground. The specific calculation formula is:

[0074] System rotational inertia Updated using the parallel axis theorem:

[0075]

[0076] in, The total centroid of the system consisting of the robot and the load after the load is applied, relative to the updated system. The moment of inertia matrix; The unloaded robot body relative to its unloaded center of mass, corresponding to the robot's current leg configuration. The moment of inertia matrix; The unloaded mass of the robot body; For load quality; To determine the unloaded center of mass of the robot body in the body coordinate system corresponding to the current leg configuration of the robot. The position vector below; The equivalent center of mass of the load in the body coordinate system The position vector below; The overall center of mass of the system consisting of the robot and the load in the body coordinate system after loading is... The position vector below; where, Indicates the body coordinate system; It is a 3×3 identity matrix; Denotes the Euclidean norm of a vector; superscript Indicates the transpose operation; and These are the offset vectors pointing from the updated system's total centroid to the robot's unloaded centroid and the loaded equivalent centroid, respectively. and These are the outer products of the corresponding offset vectors; and All in body coordinate system The following is indicated. In a simplified implementation, when When the offset is less than the preset offset threshold, the axis shift term of the robot's unloaded centroid relative to the updated system's total centroid can be ignored, and a simplified inertia compensation form caused by point mass load can be adopted.

[0077] S4: Controller Parameter Update. This embodiment uses a linear quadratic regulator (LQR) as the core balance controller. To adapt to different loads, the corresponding LQR feedback gain matrix is ​​pre-calculated offline for a series of combinations of total system mass and equivalent pendulum length, and stored as a lookup table. During online operation, the system updates the LQR parameters according to the current parameters. and The system generates a feedback gain matrix that matches the current load state in real time using bilinear interpolation, thereby dynamically adjusting the controller's response characteristics to compensate for the impact of load changes. It should be noted that this embodiment uses an LQR controller, but the invention is not limited to this. In other embodiments, model predictive controllers, PID controllers, robust controllers, or other balanced controllers can also be used, and the corresponding model parameters or control gains can be adjusted according to the updated dynamic parameters.

[0078] like Figure 7 As shown, the hierarchical closed-loop control system of this embodiment includes command generation, state estimation, adaptive estimation of load and dynamic parameters, and control of altitude, yaw angle, pitch balance, and roll angle. The adaptive estimation of load and dynamic parameters obtains the load mass and position according to steps S1 to S3 based on the joint angles, joint output torque, and body attitude information fed back by the robot, and updates the total system mass accordingly. System overall centroid Equivalent inverted pendulum length And dynamic parameters such as rotational inertia. Figure 7 The parameter pathways for estimating the mass and the equivalent pendulum length are used to characterize the updated total system mass. and equivalent inverted pendulum length The dashed parameter paths represent adaptive parameter updates. The balance controller corrects the control model based on the updated parameters and obtains the corresponding feedback gain matrix through querying or interpolation, so that the control model and control gain are adjusted with load changes; each controller generates control signals based on the current state estimate and applies them to the bipedal robot.

[0079] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A real-time load estimation and adaptive control method for a bipedal robot, characterized in that, The method includes the following steps: A mapping relationship between the joint output torque and the equivalent contact force at the end of the leg is established in the sagittal plane of the robot. Using this mapping relationship and the current posture of the robot, the components of the wheel-ground contact reaction forces acting on the left and right wheels of the bipedal robot in the vertical axis direction of the world coordinate system are solved, and these are used as the left vertical support force and the right vertical support force. The increments of the total vertical support force and the torque are calculated using the left vertical support force and the right vertical support force, respectively. The mass and coordinates of the load are calculated using the total vertical support force, and the robot's center of mass equivalent inverted pendulum length and moment of inertia are updated accordingly.

2. The real-time load estimation and adaptive control method for a bipedal robot as described in claim 1, characterized in that, The calculation formulas for the left and right vertical support forces are as follows: in, For the leg side index, the value is... or , respectively representing the left leg or the right leg; For the first The joint angle vector of the lateral leg; For the first The joint output torque vector of the lateral leg; For vertical support force, For the first lateral leg joint angle vector A defined Jacobian matrix for the legs.

3. A real-time load estimation and adaptive control method for a bipedal robot as described in claim 1 or 2, characterized in that, The formulas for calculating the increment of the total vertical support force and moment are as follows: in, This represents the total vertical support force increment caused by external loads. and These represent the vertical support force increments for the left and right legs, respectively. and These represent the vertical support forces of the left and right wheels, respectively, in the robot's current state. and These are the vertical support forces of the left and right wheels under no-load conditions, corresponding to the robot's current leg configuration and body posture. This represents the increase in pitching moment caused by external loads. For the robot's current state around the body coordinate system The resultant torque of the shaft; To determine the robot's position in a no-load state relative to its current leg configuration and body posture, orbiting the body coordinate system. The resultant torque of the shaft.

4. The real-time load estimation and adaptive control method for a bipedal robot as described in claim 3, characterized in that, The formulas for calculating the mass and coordinates of the load are as follows: in, For the quality of the load, subscript Indicates load; This represents the total vertical support force increment caused by external loads. This represents the magnitude of gravitational acceleration. For the load in the body coordinate system of The longitudinal coordinate along the axis; For the coordinate system around the machine body caused by external load of The pitching moment increment of the shaft; For the load in the body coordinate system of The horizontal coordinate along the axis; The equivalent distance between the lines of action of the ground contact reaction forces of the left and right wheels in the transverse direction of the aircraft body; and These represent the vertical support force increments for the right and left wheel sections, respectively.

5. The real-time load estimation and adaptive control method for a bipedal robot as described in claim 1, characterized in that, The formula for calculating the total centroid is as follows: in, It is the center of mass of the robot under load. For the robot's no-load mass, The position of the robot's center of mass without load. These are the three-dimensional coordinates of the load. It is the quality of the load.

6. The real-time load estimation and adaptive control method for a bipedal robot as described in claim 1, characterized in that, The equivalent inverted pendulum length The calculation formula is as follows: in, The equivalent length of the inverted pendulum, The overall center of mass of the system consisting of the robot and the load in the body coordinate system after loading is... The position vector below, The midpoint of the line connecting the centers of the left and right drive wheel axles of the bipedal robot in the body coordinate system The position vector below, To determine the body coordinate system based on the robot's current body posture. The displacement vector represented in the figure Convert to world coordinate system The rotation matrix of the displacement vector is represented in the figure, where, Indicates the body coordinate system. This represents a world coordinate system that is fixed relative to the ground.

7. The real-time load estimation and adaptive control method for a bipedal robot as described in claim 1, characterized in that, The formula for calculating the system's moment of inertia is as follows: in, The total centroid of the system consisting of the robot and the load after the load is applied, relative to the updated system. The moment of inertia matrix; The unloaded robot body relative to its unloaded center of mass, corresponding to the robot's current leg configuration. The moment of inertia matrix; The unloaded mass of the robot body; For load quality; To determine the unloaded center of mass of the robot body in the body coordinate system corresponding to the current leg configuration of the robot. The position vector below; The equivalent center of mass of the load in the body coordinate system The position vector below; The overall center of mass of the system consisting of the robot and the load in the body coordinate system after loading is... The position vector below; where, Indicates the body coordinate system; It is a 3×3 identity matrix.

8. A real-time load estimation and adaptive control system for a bipedal robot, characterized in that, The system includes an actuator for implementing a real-time load estimation and adaptive control method for a bipedal robot as described in any one of claims 1-7.

9. A computer storage medium having a computer program stored thereon, characterized in that, The program is used to implement the real-time load estimation and adaptive control method for a bipedal robot as described in any one of claims 1-7.

10. A computer program product, characterized in that, The system includes a computer program that, when executed by a processor, implements the real-time load estimation and adaptive control method for a bipedal robot as described in any one of claims 1-7.

Images