Cooperative control method and device, robot, and electronic device

Abstract

Embodiments of the application disclose a cooperative control method and device, a robot and electronic equipment, and relate to the field of automatic control. In the method, during the process that the robot performs a gripping operation on a load, the uncertainty of the spatial inertia of the load and the micro-slip precursor probability of each support foot of the robot are determined; the effective friction cone and the pressure center allowable region are contracted to generate a feasible set sequence at multiple time points in a prediction time domain; the effective friction cone includes a constraint condition that the ratio of the tangential contact force to the normal contact force between the support foot and the ground is less than or equal to the effective friction coefficient of the foot bottom; the pressure center allowable region includes a region in which the foot bottom normal force distribution center point of the support foot is limited within a preset safety boundary in the foot bottom support surface; the feasible set sequence includes the feasible range of the whole body state vector of the robot; and the cooperative control robot is constrained according to the feasible set sequence; and stable operation under the conditions of load uncertainty and foot end friction degradation is realized.

Description

Technical Field

[0001] This application relates to the field of automatic control, and more particularly to a cooperative control method, device, robot, and electronic device. Background Technology

[0002] As humanoid and legged robots are increasingly used in various fields (such as industrial handling, logistics distribution, and emergency rescue), robots may face the problem of "limb" incoordination when grasping heavy objects, making it difficult to achieve full-body coordinated control of the robot. Summary of the Invention

[0003] This application provides a collaborative control method to address the problem of "limb" incoordination that robots may face when grasping heavy objects in related technologies.

[0004] Accordingly, embodiments of this application also provide a collaborative control device, an electronic device, a computer-readable storage medium, and a computer program product to ensure the implementation and application of the above methods.

[0005] On one hand, embodiments of this application provide a collaborative control method, the method comprising: During the process of the robot performing a gripping operation on the load, the uncertainty of the load's spatial inertia is determined, as well as the probability of micro-slippage precursors of each of the robot's supporting feet. Based on the uncertainty of the spatial inertia of the load and the probability of microslip precursors, the allowable region of the effective friction cone and pressure center is shrunk to generate a feasible set sequence of multiple times in the prediction time domain. The effective friction cone includes the constraint that the ratio of the tangential contact force to the normal contact force between the supporting foot and the ground is less than or equal to the effective friction coefficient of the foot; the allowable pressure center region includes the region where the center point of the normal force distribution of the supporting foot is restricted to a preset safety boundary within the foot support surface; the feasible set sequence includes the feasible range of the robot's whole-body state vector. Under the condition of satisfying the feasible set sequence constraint, the robot is subjected to whole-body coordinated control.

[0006] On the other hand, embodiments of this application provide a collaborative control device, which includes: The parameter determination module is used to determine the uncertainty of the spatial inertia of the load and the probability of micro-slippage precursors of each support foot of the robot during the robot's gripping operation. The parameter shrinkage module is used to shrink the allowable region of the effective friction cone and pressure center based on the uncertainty of the spatial inertia of the load and the probability of microslip precursors, and generate a feasible set sequence of multiple times in the prediction time domain. The effective friction cone includes the constraint that the ratio of the tangential contact force to the normal contact force between the supporting foot and the ground is less than or equal to the effective friction coefficient of the foot; the allowable pressure center region includes the region where the center point of the normal force distribution of the supporting foot is restricted to a preset safety boundary within the foot support surface; the feasible set sequence includes the feasible range of the robot's whole-body state vector. The control module is used to perform whole-body coordinated control of the robot under the condition of satisfying the feasible set sequence constraint.

[0007] On the other hand, embodiments of this application provide an electronic device, including a processor and a memory, which are interconnected; The aforementioned memory is used to store computer programs; The processor described above is configured to execute the control method provided in the embodiments of this application when the computer program described above is invoked.

[0008] On the other hand, embodiments of this application provide a computer-readable storage medium storing a computer program that is executed by a processor to implement the control method provided in embodiments of this application.

[0009] On the other hand, embodiments of this application provide a computer program product, which includes a computer program that, when executed by a processor, implements the control method provided in embodiments of this application.

[0010] In this embodiment, during the robot's gripping operation on the load, the uncertainty of the load space inertia and the probability of micro-slippage precursors of each supporting foot are determined separately. The uncertainty of the load space inertia and the probability of micro-slippage precursors of the foot end are jointly mapped to the contraction amount of the control constraint, driving the active contraction of the effective friction cone and the allowable area of ​​the pressure center. In the prediction time domain, the contracted constraint is constrained in the form of a feasible set sequence to constrain the robot's whole-body state vector, and the robot is subjected to whole-body coordinated control. In this way, the friction cone and the allowable area of ​​the pressure center can be contracted before the uncertainty of the load space inertia increases or the foot end is significantly displaced, avoiding the continued application of torque when the friction is insufficient. This provides more time for subsequent conservative actions (e.g., performing unloading operations in advance), achieving stable operation under the conditions of load uncertainty and foot end friction degradation. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating the collaborative control method provided in an embodiment of this application is shown; Figure 2 This illustration shows one of the scenario diagrams of the collaborative control method provided in an embodiment of this application; Figure 3 The second scenario diagram of the collaborative control method provided in the embodiments of this application is shown; Figure 4 The third scenario diagram of the collaborative control method provided in this application is shown; Figure 5 The fourth scenario diagram of the collaborative control method provided in this application is shown. Figure 6 A schematic diagram of the structure of the collaborative control device provided in an embodiment of this application is shown; Figure 7 A schematic diagram of the structure of the robot provided in an embodiment of this application is shown; Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] In some embodiments, during the process of a robot grasping a heavy object (e.g., gripping, lifting, transferring, and placing), the parameters of the object (e.g., mass, center of mass position, and moment of inertia) are unknown to the robot. These uncertainties directly affect the accuracy of the robot's dynamic model. Simultaneously, the contact state between the robot's feet and the ground may change due to variations in ground friction conditions and load distribution (e.g., gradually degenerating from adhesion to micro-slippage and then macro-slippage). In other words, based on these factors, how to stably control the robot becomes a pressing technical challenge in this field.

[0015] The collaborative control method provided in this application will be described below with reference to specific embodiments: See Figure 1 The collaborative control method provided in this application embodiment may specifically include the following steps: Step 101: During the process of the robot performing a gripping operation on the load, determine the uncertainty of the load's spatial inertia, and determine the probability of micro-slippage precursors of each support foot of the robot.

[0016] Optionally, the spatial inertia of the load can be determined by load parameters such as load mass, center of mass position, and moment of inertia, and uniformly represented by a spatial inertia matrix (e.g., with a dimension of 6×6): the influence of the load on linear momentum and angular momentum during the robot's gripping operation on the load.

[0017] Optionally, in this embodiment, "robot" can be a robot in a broad sense, for example, the present invention can be applied to bipedal humanoid, quadrupedal with arms, wheeled, or mobile operating platform with variable support contact; wherein, the robot's upper limb end can be a gripper, two hands, or a tray. Specifically, it can be applied to the following scenarios: Industrial material handling: Humanoid or legged robots grasp, lift, transfer, and place unknown heavy objects in factories, and cope with sudden changes in load parameters and ground friction. Logistics and delivery: Robots need to adjust their whole-body dynamic parameters in real time when handling various packages on different ground conditions (tiles, carpets, gravel, etc.). Emergency rescue: Robots need to maintain stability when moving heavy objects in uncertain environments such as ruins and under unknown loads and uncertain ground conditions; Service industry: Humanoid robots move various items in home or office environments, and need to adapt to different loads and ground conditions.

[0018] Optionally, for each support foot, the probability of microslippage precursors for that support foot can be determined by fusing multi-source observation data from each support foot. The multi-source observation data for the support foot may include, but is not limited to: the pressure center drift velocity of the support foot, the redistribution rate of the normal force between the left and right feet, the ratio of tangential force to normal force, tangential wavelet energy, plantar shear phase consistency, and relative velocity at the foot tip.

[0019] Step 102: Based on the uncertainty of the spatial inertia of the load and the probability of microslip precursors, the allowable region of the effective friction cone and pressure center is shrunk to generate a feasible set sequence of multiple times in the prediction time domain. The effective friction cone includes the constraint that the ratio of tangential contact force to normal contact force between the supporting foot and the ground is less than or equal to the effective friction coefficient of the foot. The allowable pressure center region includes the region where the center point of the normal force distribution of the supporting foot is restricted to a preset safety boundary within the foot support surface. The feasible set sequence includes the feasible range of the robot's whole-body state vector.

[0020] Optionally, the increased load uncertainty and / or the increased probability of microslip precursors characterize the increased operational stability risk of the current robot.

[0021] Optionally, when the risk to the robot's operational stability increases, at least one of the following operations can be performed: actively contracting the effective friction cone to reduce the upper limit of the tangential force that the sole can withstand, and asymmetrically contracting the allowable area of ​​the pressure center. Specifically, when asymmetrically contracting the allowable area of ​​the pressure center, the allowable area of ​​the pressure center can be tightened away from the predicted slip direction.

[0022] Optionally, the duration range of the prediction time domain can be determined according to the actual situation. For example, when the risk of robot operation stability increases, a longer duration can be set to ensure that the robot can "stablely" perform gripping operations on the load; when the risk of robot operation stability decreases, a shorter duration can be set to reduce the waste of data caused by long-term monitoring.

[0023] Optionally, at each moment in the prediction time domain, the feasible set sequence may include at least one of the following feasible ranges corresponding to the whole-body state vector: the robot's center of mass position, center of mass momentum, contact force of each supporting foot, pressure center position of each supporting foot, joint configuration, joint velocity, etc.

[0024] Alternatively, the predicted feasible set can be represented by a polyhedron, ellipsoid, reachability set, probabilistic chance constraint, or control barrier function.

[0025] Step 103: Under the condition of satisfying the feasible set sequence constraint, perform whole-body cooperative control on the robot.

[0026] Optionally, the total momentum can be divided into the following aspects based on the momentum requirements of the robot in performing the grasping operation on the load: the momentum required to perform upper limb operations, the momentum required to maintain trunk stability, and the momentum required to adjust the support force and recover the movement.

[0027] Based on the above classification, under the condition of satisfying the feasible set sequence constraint, the torque of each joint of the robot and the contact force between the robot and the load can be determined by allocating the momentum budget among the upper limb operation, trunk stability, support force adjustment and recovery action, thereby enabling the whole-body coordinated control of the robot.

[0028] In this embodiment, during the robot's gripping operation on the load, the uncertainty of the load space inertia and the probability of micro-slippage precursors of each supporting foot are determined separately. The uncertainty of the load space inertia and the probability of micro-slippage precursors of the foot end are jointly mapped to the contraction amount of the control constraint, driving the active contraction of the effective friction cone and the allowable area of ​​the pressure center. In the prediction time domain, the contracted constraint is constrained in the form of a feasible set sequence to constrain the robot's whole-body state vector, and the robot is subjected to whole-body coordinated control. In this way, the friction cone and the allowable area of ​​the pressure center can be contracted before the uncertainty of the load space inertia increases or the foot end is significantly displaced, avoiding the continued application of torque when the friction is insufficient. This provides more time for subsequent conservative actions (e.g., performing unloading operations in advance), achieving stable operation under the conditions of load uncertainty and foot end friction degradation.

[0029] In some feasible implementations, the spatial inertia of the load is used to characterize the effect of the load on linear momentum and angular momentum, and the spatial inertia of the load is obtained based on the parameter vector of the load; The load parameter vector includes:

[0030] in, Indicates the mass of the load; These represent the mass moments of the load, which indicate the offset of the load's center of mass relative to the origin of the gripping coordinate system. These represent the moments of inertia of the load about each axis of the gripping coordinate system; These represent the product of inertia of the load about each axis of the gripping coordinate system.

[0031] Optionally, the mass of the load Used to characterize the magnitude of the translational inertia of a load. For example, .

[0032] Optionally, the mass moment of the load These represent the offsets of the load's center of mass relative to the origin of the gripping coordinate system (e.g., specifically along the x, y, and z axes) (e.g., which can be related to h=m). c is obtained and used to determine the location of the point of application of the load gravity.

[0033] Optionally, the moment of inertia of the load about each axis of the gripping coordinate system (e.g., specifically about the x-axis, y-axis, and z-axis). It can be used to characterize the rotational inertia of a load about three orthogonal axes; Optionally, the product of inertia of the load These represent the inertia products of the loads about each axis of the gripping coordinate system (specifically: between the x-axis and y-axis, between the x-axis and z-axis, and between the y-axis and z-axis), and can be used to characterize the coupling effect of the loads between different rotation axes.

[0034] Alternatively, it can be based on the parameter vector of the load. The spatial inertia matrix of the load with dimension 6×6 The unified term refers to the influence of the load on linear momentum and angular momentum during the process of a robot performing a gripping operation on the load.

[0035] In some embodiments, the parameter vector of the load can be directly used as the spatial inertia of the load. That is... Equivalent to .

[0036] Alternatively, the inertia estimation period can be: It is used for recursive updates of space inertia.

[0037] In this embodiment of the application, by uniformly representing the load parameters as a 10-dimensional load parameter vector, and further determining the spatial inertia of the load through the load parameter vector, the influence of the load on the robot's linear momentum and angular momentum during the robot's gripping operation can be fully described. Furthermore, characterizing it in the form of a parameter vector is beneficial for subsequent recursive estimation.

[0038] In some feasible implementations, the uncertainty of the spatial inertia of the load is determined, including: Determine the current action phase when the robot performs a load-based heavy-duty task; wherein the current action phase includes at least one of the following: gripping, lifting, transferring, and placing; Based on the robot's first observation data in the current action phase, estimate the posterior distribution of the spatial inertia of the load in the current action phase; the posterior distribution includes: the posterior mean of the spatial inertia of the load in the current action phase and the posterior covariance of the spatial inertia of the load in the current action phase; the first observation data includes at least one of the following: wrist six-dimensional force, joint torque residual and end effector acceleration. The uncertainty of the spatial inertia of the load is characterized by the posterior covariance of the spatial inertia of the load in the current action phase.

[0039] Optionally, during the process of the robot performing a gripping operation on the load, the process can be divided into the following four stages according to different contact and dynamic conditions: gripping, lifting, transferring, and placing.

[0040] The term "placement" can also be referred to as "unloading," and this application does not limit the use of this term in its embodiments.

[0041] Taking the placement of a load from position a to position b by a robot as an example, the following stages are defined: the stage in which the robot stably grasps the load at position a is the grasping stage; the stage in which the robot stably grasps the load to the preset height corresponding to position a is the lifting stage; the stage in which the robot moves the load from the preset height at position a to the preset height corresponding to position b is the transfer stage; and the stage in which the robot places the load from the preset height corresponding to position b at position b is the placement stage.

[0042] Optionally, the robot's action phases in the above-mentioned stages can be defined as: grasping action phase, lifting action phase, transfer action phase, and placement action phase.

[0043] Optionally, the robot may have different contact conditions and dynamic constraints depending on its action phase. For example, during the gripping phase, a coarse prior is established regarding the transformation of mass and gripping; during the lifting phase, the mass and center of mass are identified primarily by utilizing changes in the direction of gravity and joint torque residuals; during the transfer phase, multi-axis acceleration and angular velocity are used to supplement the moment of inertia information; and during the placement / unloading phase, contact force transfer and weight changes are primarily detected.

[0044] Optionally, wrist six-dimensional force The force can be obtained by a six-dimensional force sensor installed on the robot's wrist, which is used to characterize the force exerted by the load on the robot's end effector.

[0045] Optionally, the actual joint torque detected by the current or torque sensors installed on the robot can be acquired sequentially. The torque under no-load conditions or the torque predicted by the current model. ,based on and The difference between them yields the joint torque residual. ,For example, .

[0046] Optionally, terminal acceleration The linear / angular acceleration used to characterize the hand gripping coordinate system can be detected by an inertial measurement unit (IMU) installed in the robot, or obtained through visual differential, or determined based on the robot's motion state in the current action phase.

[0047] Optionally, the robot's motion state in the current action phase may include, but is not limited to, joint positions. Generalized speed Generalized acceleration End attitude wait.

[0048] Optionally, the observations and migration information required to determine the robot's motion state under different action phases can be found in Table 1.

[0049] Table 1 Action phase Main Observations Migration Information Hold on Wrist strength, gripping force, visual posture Establish coarse priors for quality and gripping transformation Lift Changes in gravity direction, joint torque residuals Identify quality and centroid, and propagate covariance. transport Multiaxial acceleration and angular velocity Replenish rotational inertia, continuous small updates Place / Unload Contact force transfer, weight reduction Detect contact changes and reset load status Optionally, acceleration at the robot's end effector Given the robot's motion state in the current action phase, the robot's end effector acceleration... It can be represented as:

[0050] in, That is The first derivative.

[0051] Optionally, the smaller the posterior covariance of the load's spatial inertia in the current action phase, the more reliable the estimation of the load parameters is, and the lower the uncertainty of the load's spatial inertia is; conversely, the larger the posterior covariance of the load's spatial inertia in the current action phase, the less reliable the estimation of the load parameters is, and the higher the uncertainty of the load's spatial inertia is.

[0052] In this embodiment, by distinguishing different action phases, the efficiency of identifying load parameters can be improved; by further fusing three types of observation data—wrist force, torque residual, and end-effector acceleration—the posterior distribution of the load's spatial inertia under the current action phase can be estimated, making the estimated posterior distribution more consistent with the actual observation data; by characterizing the uncertainty of the load's spatial inertia through the posterior covariance of the load's spatial inertia under the current action phase, the uncertainty of the load's spatial inertia can be quantified, thereby quantifying the reliability of the load parameter estimation.

[0053] In some feasible implementations, the posterior distribution of the spatial inertia of the load in the current action phase is estimated based on the robot's first observation data in the current action phase, including: Based on the gripping coordinate transformation between the current action phase and the previous action phase, the posterior distribution of the load's spatial inertia under the previous action phase is transferred to the prior distribution of the load's spatial inertia under the current action phase. Based on the prior distribution, and using the robot's first observation data in the current action phase, the posterior distribution of the spatial inertia of the load in the current action phase is estimated.

[0054] Optionally, the "previous action phase" is the action phase preceding the current action phase.

[0055] As an example, suppose the current action phase is the transfer action phase, then the previous action phase is the lifting action phase, and the next action phase is the placement action phase.

[0056] Optionally, the gripping coordinate transformation between the current action phase and the previous action phase can be used to describe the spatial pose transformation relationship from the first gripping coordinate system (the robot's coordinate system in the previous action phase) to the second gripping coordinate system (the robot's coordinate system in the current action phase).

[0057] Optionally, by estimating the posterior distribution of the spatial inertia of the load in the current action phase based on the prior distribution from the previous action phase and the robot's first observation data in the current action phase, it is possible to transfer relevant information from the previous action phase to the next action phase.

[0058] In the embodiments of this application, through the cross-phase transfer mechanism, the load parameters identified in the previous action phase can be transferred to the next action phase without having to re-estimate the distribution information from zero for each action phase, nor with repeated large-scale excitation; at the same time, by propagating the covariance, large-scale misestimation caused by short-term measurement errors can be avoided, thereby improving the accuracy of the determined posterior distribution.

[0059] In some feasible implementations, based on the gripping coordinate transformation between the current action phase and the previous action phase, the posterior distribution of the load's spatial inertia in the previous action phase is transferred to the prior distribution of the load's spatial inertia in the current action phase, including: Based on the gripping coordinate transformation between the current action phase and the previous action phase, determine the coordinate transformation matrix between the first gripping coordinate system and the second gripping coordinate system; wherein, the first gripping coordinate system is the robot's coordinate system under the previous action phase; and the second gripping coordinate system is the robot's coordinate system under the current action phase. Multiply the posterior mean of the load's spatial inertia in the previous action phase by the coordinate transformation matrix to obtain the prior mean of the load's spatial inertia in the current action phase. Based on the coordinate transformation matrix, the posterior covariance of the load's spatial inertia in the previous action phase, and the process noise covariance, the prior covariance of the load's spatial inertia in the current action phase is determined; the calculation methods for the prior covariance include:

[0060] in, Represents the prior covariance; Represents the coordinate transformation matrix; This represents the posterior covariance of the spatial inertia of the load in the previous action phase. This represents the process noise covariance.

[0061] Optionally, the prior mean of the spatial inertia of the load in the current action phase can be calculated using the following methods:

[0062] in, This represents the prior mean of the spatial inertia of the load in the current action phase. This represents the posterior mean of the spatial inertia of the load in the previous action phase.

[0063] Optionally, process noise covariance is used for quantification: additional variations caused by gripping point slippage, changes in hand constraints, internal mass movement, and unmodeled flexibility when the robot grasps a load. Specifically, when any of these variations is detected, the noise level can be increased from a preset value. When no change is detected, it can be Set to the default value.

[0064] Understandable. The corresponding preset values ​​can be set according to actual needs, and no specific restrictions are imposed here. Specifically, by only locally increasing process noise to address contact changes, the recovery speed to normal can be achieved more quickly.

[0065] In this embodiment, by performing cross-phase migration of the mean and covariance of the load parameters based on the gripping coordinate transformation between the current action phase and the previous action phase, it can be ensured that the load parameters are updated as the gripping coordinate system changes. Through the cross-phase migration mechanism, the load parameters identified in the previous action phase can be transferred to the next action phase without having to re-estimate the distribution information from zero for each action phase. At the same time, by propagating the covariance, large misestimations caused by short-term measurement errors can be avoided, improving the accuracy of the determined posterior distribution. Furthermore, the additional uncertainty brought about by contact changes is also considered when performing covariance migration, further improving the accuracy of the determined posterior distribution.

[0066] In some feasible implementations, based on the prior distribution, the posterior distribution of the spatial inertia of the load in the current action phase is estimated using the robot's first observation data in the current action phase, including: The observation residual of the current action phase is determined based on the difference between the first observation data and the second observation data; wherein the second observation data is the observation data predicted based on the prior mean. Based on the robot's motion state in the current action phase, determine the linear mapping matrix from the spatial inertia of the load to the observation residual; The recursive Bayesian estimation method is adopted to determine the posterior distribution of the load space inertia in the current action phase based on the prior distribution, the observation residual of the current action phase, the observation noise covariance of the current action phase, and the linear mapping matrix.

[0067] Optionally, the robot's first observation data in the current action phase is the actual observation data, the observation data predicted based on the prior mean is the predicted observation data, and the observation residual of the current action phase is... It can be expressed as: the absolute value of the difference between the actual observed data and the predicted observed data.

[0068] Optionally, the robot's motion state in the current action phase may include, but is not limited to, joint positions. Generalized speed Generalized acceleration End attitude wait.

[0069] Optionally, the linear mapping matrix from the spatial inertia of the load to the observation residual. When the parameter vector characterizing the load undergoes a unit change, the observed residuals The corresponding change.

[0070] Optionally, the recursive Bayesian estimation method may include, but is not limited to: recursive least squares, extended Kalman, Bayesian filtering, unscented Kalman filtering, moving window optimization, or learned residual models. Extended Kalman may include estimation based on Kalman gain.

[0071] In the embodiments of this application, the recursive Bayesian estimation framework can be used to fuse prior information (from migration) with the current observation, and the load parameters can be corrected by driving the observation residuals, so that the estimation result at each time step becomes the prior for the next time step, forming a continuous closed loop and improving the accuracy of the determined posterior distribution.

[0072] In some feasible implementations, the recursive Bayesian estimation method includes estimation based on Kalman gain, determining the posterior distribution of the load space inertia in the current action phase based on the prior distribution, the observation residuals of the current action phase, the observation noise covariance of the current action phase, and the linear mapping matrix, including: The Kalman gain is determined based on the prior covariance, the linear mapping matrix, and the observation noise covariance of the current action phase; wherein the Kalman gain includes:

[0073] in, Indicates Kalman gain; Represents the prior covariance; Represents a linear mapping matrix; Represents the observation noise covariance; Based on the prior mean, Kalman gain, linear mapping matrix, and observation residuals of the current action phase, the posterior mean of the current action phase is determined; wherein, the posterior mean of the current action phase includes:

[0074] in, This represents the posterior mean of the current action phase; This represents the prior mean; This represents the observation residual of the current action phase; Based on the prior covariance, Kalman gain, and linear mapping matrix, the posterior covariance for the current action phase is determined; wherein, the posterior covariance for the current action phase includes:

[0075] in, This represents the posterior covariance under the current action. Represents the identity matrix.

[0076] Optionally, Kalman gain It can be used to characterize the strength of the correction of actual observation data to predicted observation data.

[0077] Optionally, the prior mean of the spatial inertia of the load in the current action phase is used to characterize the uncertainty after the phase transition.

[0078] In this embodiment, Kalman filtering is used to assign weights between the prior distribution and the current actual observations, updating the posterior distribution under the current action phase. For example, the larger the prior covariance or the smaller the observation noise, the larger the Kalman gain, making it easier to adjust the prior distribution based on the actual observations to obtain the posterior distribution. Simultaneously, the updated covariance is strictly reduced (…). This allows the understanding of load parameters to continue to converge.

[0079] Among some feasible implementation methods, the approach also includes: Based on the load parameter information gain, the cost of micro-excitation amplitude, and the cost of micro-slip risk, the target micro-excitation is determined from the micro-excitation candidate set. Embedding target micro-excitations into the robot's task trajectory reduces the posterior covariance of the load's spatial inertia in the current action phase; The target micro-excitation includes at least one of short-term posture sway, velocity curvature change and hand-hand differential movement.

[0080] Optionally, short-term attitude oscillations may include minute attitude oscillations (e.g., small pitch, roll, or yaw amplitudes) generated by the robot's floating base (e.g., torso or waist) over a very short period of time; in some embodiments, the amplitude is typically 0.2°. 1.0°).

[0081] In some embodiments, the moment of inertia of the load can be determined based on short-term posture swaying. For example, during torso swaying, the load generates an inertial torque, and by measuring changes in wrist force or torque, the moment of inertia of the load about each axis of the gripping coordinate system (e.g., specifically about the x, y, and z axes) can be deduced. .

[0082] Optionally, the velocity curvature variation may include the square of the normal acceleration or jerk introduced by the robot's end effector (e.g., hand) as it moves along the planned trajectory, causing small fluctuations in the end effector trajectory.

[0083] In some embodiments, the position or mass of the load's center of mass can be determined by changes in velocity curvature. For example, when the end effector suddenly accelerates or turns, the load's inertial force increases significantly. At this time, centrifugal force or tangential inertial force can be detected by a wrist force sensor, and the position of the load's center of mass relative to the gripper can be determined based on the centrifugal force or tangential inertial force.

[0084] Optionally, dual-arm differential includes the inertial torque or force generated by the two arms applying opposite directions during the process of the robot's two arms grasping the same load (e.g., the left hand pushes forward and the right hand pulls back, or the left hand moves up and the right hand moves down, etc.).

[0085] In some embodiments, the center of mass or mass of a load can be determined by two-hand differential motion. For example, by two-hand differential motion, a "rotational excitation" that reflects the offset of the load's center of mass can be "artificially" created without moving the overall center of gravity of the load.

[0086] Optionally, in some embodiments, micro-excitation can be applied only in the range where there is sufficient gripping margin and bipedal stability, i.e., the target micro-excitation can be embedded in the robot's task trajectory.

[0087] In the embodiments of this application, by considering the information gain of load parameters, the cost of micro-excitation amplitude and the cost of micro-slip risk, a target micro-excitation is determined from the micro-excitation candidate set, and then the target micro-excitation is embedded in the robot's task trajectory. This can improve the identifiability of load parameters from multiple directions (at least one of short-term posture sway, velocity curvature change and hand differential movement), accelerate covariance convergence, and reduce the posterior covariance of the load's spatial inertia in the current action phase.

[0088] In some feasible implementations, the target micro-incentive satisfies the following condition:

[0089] in, Indicates the gain of load parameter information; Represents the Fischer information matrix; Represents the regularization term; Indicates the cost of micro-incentives; This represents the weighting coefficient corresponding to the cost of the micro-incentive. This indicates the cost of microslip risk; This represents the weighting coefficient corresponding to the cost of microslip risk.

[0090] Optionally, regularization terms Used to ensure numerical stability Used to measure the identifiable gain of a target micro-excitation on load parameters.

[0091] Optionally, embodiments of this application provide for and The specific value is not limited and can be set according to actual needs.

[0092] Optionally, the cost of micro-incentive amplitude Used to constrain the magnitude of the incentive.

[0093] Optionally, the cost of microslip risk Used to suppress the risk of micro-slippage induced by micro-stimulation candidates at the foot. This can also be called a preset risk weighting coefficient. This represents the probability of microslip precursor after the micro-excitation is executed.

[0094] In this embodiment, Fisher information-driven optimization is used to treat micro-slip risk as a penalty, thus preventing slippage from occurring simply to increase information gain and improving safety. By achieving a Pareto optimal balance between information gain, energy consumption, and safety risk, and selecting the excitation that minimizes the load parameter covariance as the target micro-excitation, information gain can be maximized and load parameter covariance can be effectively reduced.

[0095] In some feasible implementations, the probability of microslippage precursors for each foot of the robot is determined, including: During the robot's gripping operation on the load, at least two observational data points are acquired: the drift velocity of the center of pressure (CoP) of the supporting foot. , left and right foot normal force redistribution rate The ratio of tangential force to normal force Tangential wavelet energy Plantar shear phase consistency Relative speed to the foot ; Based on at least two observational data points, determine the microslippage precursor feature vector of the supporting foot; The microslip precursor eigenvectors are mapped to microslip precursor probabilities.

[0096] Optionally, the pressure center drift speed of the support foot It can be used to characterize the speed of normal load migration on the sole of the foot.

[0097] Optionally, This can represent the COP drift warning threshold, which is one of the thresholds for determining micro-slippage precursors.

[0098] Optionally, the redistribution rate of normal force between the left and right feet It can be used to characterize the intensity of dynamic transfer of normal load on the sole of the foot.

[0099] Optionally, the "redistribution rate of normal force between the left and right feet" can also be referred to as "slight increase in relative speed of the soles of the feet". This application does not limit the name.

[0100] Optionally, the ratio of tangential force to normal force It can be used to characterize tangential force fluctuations. The "ratio of tangential force to normal force" can also be called "tangential force fluctuations". This application does not limit the name in the embodiments.

[0101] Optionally, tangential wavelet energy It can be used to characterize high-frequency shear oscillation energy, and thus characterize local friction / slippage changes at the foot. For example, the tangential wavelet frequency band can be 15-80 Hz. That is, the tangential wavelet energy can be specifically defined as the tangential wavelet energy of 15-80 Hz.

[0102] Optionally, plantar shear phase consistency It can be used to characterize the phase difference of shear stress in the plantar region.

[0103] Optionally, relative velocity of the feet It can be used to characterize the amplitude of micro-slip velocity between the sole of the foot and the ground.

[0104] Alternatively, the foot can be monitored using a pressure array, a six-dimensional force sensor, tactile skin, a plantar IMU, a visual odometry, or a motor-side observer.

[0105] Optionally, the observation data can be classified based on its role, for example, into multiple independent observation families, specifically, type A normal load transfer, type B tangential stress / vibration, and type C kinematic relative motion.

[0106] Optionally, the type A normal load transfer may include: pressure center drift velocity. The rate of redistribution of normal force between the left and right feet / the relative velocity of the soles of the feet increased slightly. Type B tangential stress / vibration may include: tangential force / normal force ratio / tangential force fluctuation. 15-80 Hz tangential wavelet energy Plantar shear phase consistency / local shear phase difference Category C kinematic relative motion may include visual or kinematic foot-to-foot relative velocity. .

[0107] In some embodiments, at least two independent observation families can be acquired during the robot's gripping operation on the load. Based on the at least two independent observation families, the micro-slippage precursor feature vector of the support foot can be determined to avoid a single pressure spike only increasing uncertainty without directly determining the slippage probability. That is, when a single observation data exceeds the threshold, it only increases the risk uncertainty and cannot directly determine macroscopic slippage.

[0108] Optionally, the micro-slip precursor feature vector of the supporting foot It can be represented as:

[0109] Optionally, at least one of the following can be used: a logical model with ground type conditions, a temporal network, or a Bayesian filter. Based on at least two independent observation families, the precursor probability of single-foot microslippage (i.e., the predicted probability of macroscopic slippage of the supporting foot) can be determined. This involves mapping the microslip precursor eigenvectors to microslip precursor probabilities.

[0110] Understandably, the probability of microslippage precursors for each supporting foot can be determined based on the above method.

[0111] In some embodiments, The value of is between 0 and 1, where Approaching 0 (i.e.) →0), indicating that the contact force depth falls inside the friction cone, the risk of slippage is low, and it is relatively safe; Approaching 1 (i.e.) →1), which indicates that the contact force approaches / penetrates the boundary of the friction cone, and the risk of slippage is high, making slippage very easy to occur.

[0112] In this embodiment, by fusing multi-source observation data, micro-slip precursors can be detected before macroscopic foot displacement, and the friction cone and stabilizing tube (i.e., the allowable area of ​​the pressure center) can be contracted before obvious displacement, thus gaining time for preventive unloading. At the same time, the probability of micro-slip precursors is determined by at least two physically independent observation families, avoiding misjudgment triggered by noise from a single sensor. Moreover, even if a certain type of sensor fails, other observation families can still provide micro-slip precursor information.

[0113] In some feasible implementations, the allowable area of ​​the effective friction cone and pressure center is reduced, including: The first shrinkage amount is determined based on the uncertainty of the spatial inertia of the load; wherein, the first shrinkage amount is positively correlated with the uncertainty of the spatial inertia of the load. The second contraction amount is determined based on the microslip precursor probability; wherein, the second contraction amount is positively correlated with the microslip precursor probability; Based on the first and second contraction amounts, the effective friction coefficient of the foot sole is monotonically reduced according to the probability of micro-slippage precursor, and the allowable area of ​​the pressure center is asymmetrically contracted toward the side away from the predicted slippage direction. The calculation methods for the effective coefficient of friction of the sole include:

[0114] in, Indicates the effective coefficient of friction of the sole of the foot; Indicates the nominal coefficient of friction; Indicates the coefficient of friction cone shrinkage; This indicates the probability of microslip precursors.

[0115] Optionally, the higher the uncertainty, the greater the initial contraction, i.e., the tighter the contraction at the stable tube boundary.

[0116] Optionally, the higher the probability of microslippage precursors, the more limited the plantar tolerance is, and correspondingly, the greater the second contraction, the tighter the stabilizing tube boundary contraction.

[0117] Optionally, The actual usable coefficient of friction after shrinkage, also known as the "nominal coefficient of friction" or "theoretical coulombic coefficient of friction of the material".

[0118] Optionally, the friction cone contraction coefficient can characterize the proportionality coefficient that maps the microslip precursor probability to the friction coefficient decay.

[0119] Optionally, when contracting the allowable area of ​​the pressure center, the boundary can be tightened in a preset sliding direction, while retaining an operating margin in the opposite direction.

[0120] In this embodiment, the uncertainty of the spatial inertia of the load and the risk of slippage can be transformed into a continuous constraint contraction amount through the dual contraction mechanism, which jointly drives the contraction of the stabilizing tube, so that the pressure center allowable area contracts asymmetrically away from the slippage direction, which tightens the constraint while retaining the operating margin in the safe direction.

[0121] See Figure 2 It can perform covariance and risk co-contraction stabilization based on the migration of the load's spatial inertia (grasping prior, lifting identification, transfer update, placement unloading) and foot contact risk information (micro-slip precursor probability determined based on pressure center, tangential wavelet, shear phase, relative foot speed, etc.).

[0122] In some feasible implementations, the allowable area of ​​the effective friction cone and pressure center is reduced, including: When the probability of microslip precursors is greater than or equal to the first threshold and less than the second threshold, the effective friction cone and the allowable area of ​​the pressure center are contracted. The second threshold is greater than the first threshold. The first threshold is used to trigger the contraction operation, and the second threshold is used to trigger the tangential impulse limit. The contraction operation includes contracting the effective friction cone and the allowable area of ​​the pressure center.

[0123] Optionally, the first threshold can be called the microslippage warning threshold or the stabilization tube contraction warning threshold, indicating that the slippage risk is controllable. The second threshold can be called the recovery action trigger threshold, indicating that the slippage risk is uncontrollable.

[0124] Optionally, the specific values ​​of the first and second thresholds can be set according to actual needs. As an example, the first threshold can be 0.60, and the second threshold can be 0.82.

[0125] Optionally, in In this case, it indicates that the probability of microslippage precursors is controllable and can be used as a basis for judging the contraction stabilization tube. For example, based on the first and second contraction amounts mentioned above, the effective friction cone and CoP boundary are contracted to ensure the stability of the robot.

[0126] In this embodiment, a progressive intervention is formed by two levels of thresholds to distinguish between the two states of "warning" and "danger". Different levels of control actions are triggered respectively, and the control actions are contracted or triggered to recover in stages. This can achieve early intervention in the warning stage, that is, tightening constraints, gaining reaction time, and avoiding robot instability.

[0127] Among some feasible implementations, the above methods also include: If the probability of microslip precursor is greater than or equal to the second threshold, it is prohibited to continue increasing the tangential impulse of the support foot in the same direction.

[0128] Optionally, in In this case, it indicates that the probability of micro-slippage precursor is too high, which can be used as the basis for initiating preventive unloading / step recovery. It is forbidden to continue to increase the tangential impulse of the support foot in the same direction to ensure the operational stability of the robot.

[0129] In this embodiment of the application, by directly limiting the growth of tangential force in the dangerous direction when the risk of slippage is extremely high, a safety net is provided to prevent the tangential force from continuing to increase when there is insufficient friction, which could lead to slippage of the robot's foot.

[0130] In some feasible implementations, a sequence of feasible sets of times within the prediction time domain is generated, including: Based on the shrinkage results of the effective friction cone and the allowable area of ​​the pressure center, determine the contact constraint after shrinkage; Under the condition of satisfying the contact constraints after contraction, based on the robot's center of mass position, robot's center of mass momentum, and robot joint position information... The acceleration of the robot's joints Contact force of the supporting foot The uncertainty of at least one of the following—the position of the pressure center of the supporting foot and the spatial inertia of the load—is used to determine the robot's whole-body state vector at each moment in a preset time domain. .

[0131] Alternatively, as mentioned above, the posterior covariance of the load's spatial inertia in the current action phase can be used as a metric. The uncertainty in the spatial inertia of the load.

[0132] Optionally, for a certain moment within a preset time domain, the whole-body state vector at that moment... It can be represented as: {Robot's center of mass position, robot's center of mass momentum, robot's joint position information} The acceleration of the robot's joints Contact force of the supporting foot The pressure center position of the supporting foot (CoP).

[0133] Optionally, the preset time domain, also known as the rolling prediction time domain, is the time window for the stabilizing tube to predict forward. The duration of the preset time domain can be set according to actual needs, for example, 0.4s≤H≤1.2s.

[0134] Optionally, the sequence of feasible sets at t times within the prediction time domain H can be represented as:

[0135] Optionally, the boundary of the feasible set sequence can vary with the uncertainty of the spatial inertia of the load. and microslip precursor probability Dynamic contraction occurs when the uncertainty of the spatial inertia of the load or the probability of microslippage is greater, leading to a more pronounced contraction of the feasible set sequence. In other words:

[0136] in, The original unconstrained feasible region inequality represents the fundamental constraints corresponding to the static supporting polygon and the ideal friction cone. This represents the load instability contraction term (i.e., the first contraction), which is caused by the contraction of the load inertia covariance to stabilize the tube boundary. This represents the slip risk contraction term (i.e., the second contraction), which is caused by the micro-slip probability contraction of the stable tube boundary.

[0137] In this embodiment, the boundary of the feasible set sequence is tightened by driving the uncertainty of the spatial inertia of the load and the slip risk. If the predicted trajectory exceeds the boundary of the feasible set sequence in the prediction time domain, the system can select a recovery action in advance (e.g., reduce upper limb acceleration, retract the load, adjust the torso, redistribute the support force or take a step) without waiting for the zero moment point (ZMP) to cross the boundary.

[0138] In some feasible implementations, under the condition of satisfying the feasible set sequence constraint, the robot is subjected to whole-body cooperative control, including: Under the conditions of satisfying robot dynamics, contraction contact constraints, and feasible set sequence constraints, adjust the allocation ratio between at least one of the following momentum budgets: Momentum budget required for a robot to perform upper limb manipulation The momentum budget required for the robot to maintain torso stability And the momentum budget required for the robot to adjust its support force and recover its motion. .

[0139] Alternatively, robot dynamics can be characterized as .

[0140] in, This represents the robot's coordinate information obtained through encoder or state estimation (e.g., 6-DOF pose of the base + all joint angles). express The first derivative of , i.e., the rate; express The second derivative of , i.e., the acceleration vector; Represents the inertia matrix; That is, to merge nonlinear force vectors; Indicates the driving force of the joint. This represents the joint torque vector, obtained through detection by a current or torque sensor. This indicates a selection matrix; the base has no driver. Indicates the contact force of the foot. This represents the helical force of the foot contact force, obtained through plantar pressure or a six-dimensional force sensor. The Jacobian matrix of the foot space is represented by the Jacobian transpose, which maps the spatial contact force to... The force applied; This represents the interaction force at the end of the hand (wrist strength). This indicates the six-dimensional wrist force detected by a wrist force sensor. This represents the generalized Jacobian matrix of the hand's end.

[0141] Optionally, the contraction contact constraint may include, but is not limited to, at least one of the following: , , And it does not penetrate upon contact.

[0142] in, This represents the normal force exerted by the sole of the foot in contact with the ground. This means that the feet must be firmly planted on the ground to avoid lifting off the ground and losing support; This represents the tangential force vector on the sole of the foot. This indicates the magnitude of the resultant tangential force. This indicates that the sole of the foot should not exceed the friction limit to avoid slippage. This means the robot's "center of gravity" must not be "off-center" to prevent the robot from tipping over; "non-penetrating contact" means that the displacement of each contact point / surface of the supporting foot (sole) in the normal (vertical) direction must be greater than or equal to the ground height.

[0143] Optionally, the momentum budget can characterize the rate of change of center of mass momentum and the share of contact force margin that allow for upper limb manipulation, trunk compensation, gait adjustment, and emergency recovery.

[0144] Optionally, under the conditions of satisfying robot dynamics, contraction contact constraints, and stabilizing tube constraints (i.e., a sequence of feasible sets of multiple moments in the prediction time domain), upper limb manipulation can be performed. Trunk stability Support adjustment and recovery actions Momentum budget allocation between (i.e.) ), and solve for the control commands of the joint torques and contact forces of the whole body.

[0145] Optionally, when allocating momentum budget, allocation priorities can be set. For example, the first allocation priority can be higher than the second allocation priority, and the second allocation priority can be higher than the third allocation priority. The first allocation priority includes the allocation priority of the momentum budget required for the robot to adjust its support force and recover its movements. The second allocation priority includes the allocation priority of the momentum budget required for the robot to maintain trunk stability. The third allocation priority includes the allocation priority of the momentum budget required for the robot to perform upper limb operations. In this way, operational safety can be prioritized over operational efficiency.

[0146] Optionally, when solving for the joint torques and contact force control commands, they can be solved sequentially in descending order of priority, and low-priority tasks must not violate the upper-level safety constraints: Level 1 (Hard Safety Constraints): Satisfies the floating base dynamics equation unilateral contact Effective friction cone after contraction Plantar pressure center And joint rigidity limit (joint torque limit); Level 2: Maintain the predicted state within the stable tube; Level 3: Track end effector and torso posture within the remaining feasible space; Level 4: Minimize torque variation and energy consumption.

[0147] If the predicted trajectory is about to "break out" of the stabilization tube (the predicted state will leave the range of the stabilization tube constraint within a preset time), then calculate the predicted out-of-bounds time. Based on the most dangerous constraints, remaining friction / CoP margins for each foot, available footing areas, and load vulnerability levels, recovery schemes are selected in order of increasing severity, and hysteresis is employed to avoid frequent switching. (1) Preventative unloading: In cases of mild slippage risk and In prolonged cases, reduce the upper limb load acceleration, retract the load, or transfer the support force to restore friction and stability margin; (2) Intrafoot pressure redistribution: When the danger mainly comes from the edge of the CoP of a single foot and the foot still has friction margin, the normal force in the foot is redistributed to widen the CoP margin. (3) Transfer of support force between the two feet: When there is a bearing margin on the opposite foot, the support load is transferred between the two feet to balance the friction margin on both sides; (4) Step recovery: If the above actions still cannot restore the stable tube margin within the specified time, In cases where the contact is very short or about to be lost, adjust the landing point by taking a step to expand the support polygon.

[0148] Optionally, in the above control process, it can be based on quadratic programming (QP), model predictive control, sequential quadratic programming, or policy networks with safe projection; contact hard constraints, stability priority, and recovery budget lower bounds cannot be covered by low priority tasks.

[0149] In other words, see Figure 3 The process of predicting the stability of the tube and hierarchical control is as follows: A posteriori of load space inertia: mass, center of mass, moment of inertia and covariance propagation across phases; Determination of early signs of foot slippage: Output the slippage probability of each supporting foot from multi-source data; Predicting a stable tube: The uncertainties of the friction cone, pressure center, and inertia form a time-varying feasible set (i.e., a sequence of feasible sets); Momentum budgeting and tiered QP: Prioritize safety, stability and recovery, and redistribute upper limb operational margins.

[0150] In the embodiments of this application, momentum budget allocation is performed under the conditions of satisfying robot dynamics, contraction contact constraints and feasible set sequence constraints. The momentum share of each task can be dynamically adjusted according to the risk level to perform whole-body optimization of upper limb operation, trunk stability and foot support force.

[0151] In some feasible implementations, while satisfying robot dynamics and feasible set sequence constraints, the allocation ratio between at least one of the following momentum budgets is adjusted, including at least one of the following: If the uncertainty of the spatial inertia of the load meets the inertia confidence threshold and the probability of microslip precursor is less than or equal to the preset value, increase the momentum budget required for the robot to perform upper limb operations. If the uncertainty of the spatial inertia of the load does not meet the inertia confidence threshold, and the probability of microslippage precursors is greater than the preset value, the momentum budget required for the robot to maintain trunk stability and the momentum budget required for the robot to adjust the support force and recover the action are increased.

[0152] Optionally, the momentum budget required for the robot to adjust its support force and recover its motion. The lower bound is determined by the maximum probability of microslippage precursors at each support foot. Trace of the covariance of the load space inertia Synchronous rise:

[0153] in, This represents the slip probability weighting coefficient, i.e., the coefficient by which the recovery momentum budget increases with the slip probability. This represents the load uncertainty weighting coefficient, which is the proportionality coefficient of the recovery momentum budget as it increases with the inertial covariance.

[0154] Optionally, the inertia confidence threshold can be expressed as ,in, The preferred value is 0.15.

[0155] Optionally, the preset value here can be determined according to the actual situation, and there is no restriction here.

[0156] The uncertainty of the spatial inertia of the load satisfies the inertia confidence threshold. If the probability of microslippage precursors is less than or equal to a preset value, it indicates that the current operation risk is low, the load identification is sufficient, the foot risk is low, high acceleration transfer is allowed, and the momentum budget required for the robot to perform upper limb operations can be increased. This improves operational efficiency.

[0157] The uncertainty of the spatial inertia of the load does not satisfy the inertia reliability threshold. If the probability of microslippage precursors is greater than a preset value, it indicates that the current operation is at high risk, and the momentum budget required for the robot to maintain trunk stability should be increased. And the momentum budget required for the robot to adjust its support force and recover its motion. This ensures the safety of the robot's operation.

[0158] In the embodiments of this application, a dynamic allocation strategy is used to automatically reduce the operating budget and increase the stabilization and recovery budget when the load uncertainty is large or the slip risk is high; when the load uncertainty is small or the slip risk is low, the operating budget is increased to improve operating efficiency, thereby achieving dynamic control efficiency that ensures safety and improves operating efficiency.

[0159] Referring to the foregoing, the cooperative control method provided in this application is applied to a load gripping scenario to perform cooperative control of the robot's entire body. Therefore, this cooperative control method can be called the "Load-Based Whole-Body Control (LoadTube-WBC; WBC is an abbreviation for Whole-Body Control)" method. In this LoadTube-WBC method: First, the posterior inertia of the load space is recursively derived according to the action phase, and the posterior of the previous action phase is propagated to the next action phase through gripping transformation, avoiding the need to re-estimate from zero at each stage and shortening the transition time from conservative to normal operation of the unknown load. Meanwhile, constrained micro-excitations are injected into the natural operating trajectory to improve the identifiability of load parameters; By integrating plantar pressure, six-dimensional force, inertia, and relative foot motion to extract micro-slip precursors, a predictive stabilization tube containing uncertainty is constructed. Multi-source micro-slip precursors are used to predict slip risk in advance, and load parameter uncertainty and slip risk are uniformly incorporated into the predictive stabilization tube, enabling the controller to automatically take more conservative actions when the center of gravity position is not yet determined or when ground friction changes. Finally, dynamic allocation of operations, stabilization and recovery of momentum budgets are achieved in the whole-body optimization, forming a complete closed loop of "load identification - micro-slip prediction - stabilization tube constraint - momentum budget allocation". This enables the simultaneous reduction of load acceleration, adjustment of the torso and redistribution of support force when the risk increases, avoiding secondary risks caused by "only applying foot end force".

[0160] In this embodiment, by injecting constrained micro-excitations into the natural operating trajectory, the posterior inertia of the load space is recursively deduced according to the action phase; at the same time, micro-slip precursors are extracted by integrating plantar pressure, six-dimensional force, inertia and relative foot motion, a predictive stabilizing tube containing uncertainty is constructed, and the operating, stabilizing and restoring momentum budgets are dynamically allocated in the whole-body optimization.

[0161] The same spatial inertia state, stabilization tube, and momentum budget can cover bipedal standing, short-step adjustment, and continuous walking, without the need to exchange inconsistent load models between the upper limb maneuvering controller and the gait recovery controller.

[0162] In this embodiment, the spatial inertia dimension is fixed, and the recursive estimation complexity is linearly or quadratically related to the robot's degrees of freedom; the stabilizing tube uses a fixed prediction time domain and a polyhedral approximation, which can be incorporated into real-time QP. Precursor fusion operates locally on each foot, without requiring high-resolution visual ground reconstruction.

[0163] In the embodiments of this application, during the verification phase, the effectiveness of the collaborative control method provided in the embodiments of this application can be verified in the following scenarios: lifting of an unknown box, transfer of eccentric loads, hand-to-hand switching, placement on a low-friction surface, and sudden internal mass movement.

[0164] The effectiveness of the collaborative control method provided in this application embodiment can be verified by estimating the accuracy of at least one of the following indicators: inertia estimation error, macroscopic slip occurrence rate, minimum stable tube margin, peak joint torque, task time, and recovery success rate.

[0165] Furthermore, the effectiveness of the cooperative control method provided in the embodiments of this application can be verified by removing at least one of the following: cross-phase migration, microslip precursors, covariance contraction, and momentum budget.

[0166] In some embodiments, see Figure 4 This application also provides a collaborative control system, which may include: a manipulation phase identification module, a spatial inertia estimation module, a natural micro-excitation planning module, a foot micro-slippage precursor fusion module, a predictive stabilization tube generation module, and a whole-body hierarchical optimization control module. Among these, The Manipulation Phase Recognition Module is used to divide the current stage into multiple manipulation phases, such as gripping, lifting, transporting, and placing / unloading, based on the heavy-load task. Different manipulation phases have different contact conditions and different dynamic constraints. The Spatial Inertia estimation module is used to estimate the period based on the inertia. = 20-50ms, integrate wrist force, torque residual, and end-effector acceleration to update the load posterior; use grip transformation to transfer the inertia covariance of the previous stage to the prior of the current stage; The natural micro-incentive planning module is used to embed micro-incentives to accelerate convergence; The foot microslippage precursor fusion module is used to calculate features such as CoP drift, tangential wavelet energy, and shear phase based on plantar pressure, six-dimensional force, and IMU, and output the slippage probability of each foot. ; The predictive stable tube generation module is used to combine load covariance and slip probability, shrink the effective friction cone and CoP boundary, and construct a time-varying feasible stable tube (i.e., feasible set sequence) in the predictive time domain. The whole-body layered optimization control module is used to redistribute the operational / stabilization / recovery momentum budget shares according to risk; solve for the optimal joint torque and plantar contact force according to safety priority; and issue control commands. The whole-body hierarchical optimization control module may include a task planner, which provides the end-effector trajectory and motion phase labels, embeds micro-excitations in segments that do not change the key points of the task, and recursively updates the posterior load. The control cycle can be: .

[0167] When the stabilizer tube contracts or the inertia uncertainty increases, the controller reduces the operating momentum budget and transfers the margin to the trunk, support force adjustment, or step recovery.

[0168] The data required by the above modules can be acquired through sensors. Specifically, the sensors may include current or torque sensors, wrist six-dimensional force sensors, plantar pressure or six-dimensional force sensors, IMUs, and optional vision sensors. The current or torque sensors are used to output joint torque. A six-dimensional force sensor in the wrist is used to output wrist force. Plantar pressure or six-dimensional force is used to output the contact force at the foot end. And solvable pressure center The IMU is used to acquire base attitude and acceleration to calculate short-term attitude sway; vision is used to provide end-effector pose to calculate velocity curvature changes and plantar relative velocity.

[0169] See Figure 5 As shown in the figure, experiments have demonstrated that, compared to independent load or balanced baseline methods, the embodiments of this application achieve superior performance in terms of inertia convergence speed (i.e., inertia estimation error), slip prevention (i.e., determination of micro-slip precursor probability), stabilization margin (i.e., prediction of stabilization tube generation), and recovery success rate (i.e., whole-body hierarchical optimization control). Specifically, the embodiments of this application improve inertia convergence speed by 30%; slip prevention by 40%; stabilization margin by 23%; and recovery success rate by 34%.

[0170] Based on the same principles as the methods and apparatus provided in the embodiments of this application, the embodiments of this application provide a collaborative control device 600, see [link to relevant documentation]. Figure 6 The device 600 includes: The parameter determination module 601 is used to determine the uncertainty of the spatial inertia of the load and the probability of micro-slippage precursors of each support foot of the robot during the process of the robot performing a gripping operation on the load. The parameter shrinkage module 602 is used to shrink the allowable region of the effective friction cone and pressure center based on the uncertainty of the spatial inertia of the load and the probability of microslip precursors, and generate a feasible set sequence of multiple times in the prediction time domain. The effective friction cone includes the constraint that the ratio of the tangential contact force to the normal contact force between the supporting foot and the ground is less than or equal to the effective friction coefficient of the foot; the allowable pressure center region includes the region where the center point of the normal force distribution of the supporting foot is restricted to a preset safety boundary within the foot support surface; the feasible set sequence includes the feasible range of the robot's whole-body state vector. The control module 603 is used to perform whole-body coordinated control of the robot under the condition of satisfying the feasible set sequence constraint.

[0171] In some feasible implementations, the spatial inertia of the load is used to characterize the effect of the load on linear momentum and angular momentum, and the spatial inertia of the load is obtained based on the parameter vector of the load; The load parameter vector includes:

[0172] in, Indicates the mass of the load; These represent the mass moments of the load, which indicate the offset of the load's center of mass relative to the origin of the gripping coordinate system. These represent the moments of inertia of the load about each axis of the gripping coordinate system; These represent the product of inertia of the load about each axis of the gripping coordinate system.

[0173] In some feasible implementations, the parameter determination module 601, when determining the uncertainty of the spatial inertia of the load, is used for: Determine the current action phase when the robot performs a load-based heavy-duty task; wherein the current action phase includes at least one of the following: gripping, lifting, transferring, and placing; Based on the robot's first observation data in the current action phase, estimate the posterior distribution of the spatial inertia of the load in the current action phase; the posterior distribution includes: the posterior mean of the spatial inertia of the load in the current action phase and the posterior covariance of the spatial inertia of the load in the current action phase; the first observation data includes at least one of the following: wrist six-dimensional force, joint torque residual and end effector acceleration. The uncertainty of the spatial inertia of the load is characterized by the posterior covariance of the spatial inertia of the load in the current action phase.

[0174] In some feasible implementations, when the parameter determination module 601 estimates the posterior distribution of the spatial inertia of the load in the current action phase based on the robot's first observation data in the current action phase, it is used to: transfer the posterior distribution of the spatial inertia of the load in the previous action phase to the prior distribution of the spatial inertia of the load in the current action phase based on the gripping coordinate transformation between the current action phase and the previous action phase. Based on the prior distribution, and using the robot's first observation data in the current action phase, the posterior distribution of the spatial inertia of the load in the current action phase is estimated.

[0175] In some feasible implementations, the parameter determination module 601, when transferring the posterior distribution of the load's spatial inertia in the previous action phase to the prior distribution of the load's spatial inertia in the current action phase based on the gripping coordinate transformation between the current action phase and the previous action phase, is used for: Based on the gripping coordinate transformation between the current action phase and the previous action phase, determine the coordinate transformation matrix between the first gripping coordinate system and the second gripping coordinate system; wherein, the first gripping coordinate system is the robot's coordinate system under the previous action phase; and the second gripping coordinate system is the robot's coordinate system under the current action phase. Multiply the posterior mean of the load's spatial inertia in the previous action phase by the coordinate transformation matrix to obtain the prior mean of the load's spatial inertia in the current action phase. Based on the coordinate transformation matrix, the posterior covariance of the load's spatial inertia in the previous action phase, and the process noise covariance, the prior covariance of the load's spatial inertia in the current action phase is determined; the calculation methods for the prior covariance include:

[0176] in, Represents the prior covariance; Represents the coordinate transformation matrix; This represents the posterior covariance of the spatial inertia of the load in the previous action phase. This represents the process noise covariance.

[0177] In some feasible implementations, the parameter determination module 601, when estimating the posterior distribution of the spatial inertia of the load in the current action phase based on the prior distribution and the robot's first observation data in the current action phase, is used for: The observation residual of the current action phase is determined based on the difference between the first observation data and the second observation data; wherein the second observation data is the observation data predicted based on the prior mean. Based on the robot's motion state in the current action phase, determine the linear mapping matrix from the spatial inertia of the load to the observation residual; The recursive Bayesian estimation method is adopted to determine the posterior distribution of the load space inertia in the current action phase based on the prior distribution, the observation residual of the current action phase, the observation noise covariance of the current action phase, and the linear mapping matrix.

[0178] In some feasible implementations, the recursive Bayesian estimation method includes estimation based on Kalman gain. The parameter determination module 601, when determining the posterior distribution of the load space inertia in the current action phase based on the prior distribution, the observation residual of the current action phase, the observation noise covariance of the current action phase, and the linear mapping matrix, is used for: The Kalman gain is determined based on the prior covariance, the linear mapping matrix, and the observation noise covariance of the current action phase; wherein the Kalman gain includes:

[0179] in, Indicates Kalman gain; Represents the prior covariance; Represents a linear mapping matrix; Represents the observation noise covariance; Based on the prior mean, Kalman gain, linear mapping matrix, and observation residuals of the current action phase, the posterior mean of the current action phase is determined; wherein, the posterior mean of the current action phase includes:

[0180] in, This represents the posterior mean of the current action phase; This represents the prior mean; This represents the observation residual of the current action phase; Based on the prior covariance, Kalman gain, and linear mapping matrix, the posterior covariance for the current action phase is determined; wherein, the posterior covariance for the current action phase includes:

[0181] in, This represents the posterior covariance under the current action. Represents the identity matrix.

[0182] In some feasible implementations, the aforementioned device 600 is also used for: Based on the load parameter information gain, the cost of micro-excitation amplitude, and the cost of micro-slip risk, the target micro-excitation is determined from the micro-excitation candidate set. Embedding target micro-excitations into the robot's task trajectory reduces the posterior covariance of the load's spatial inertia in the current action phase; The target micro-excitation includes at least one of short-term posture sway, velocity curvature change and hand-hand differential movement.

[0183] In some feasible implementations, the target micro-incentive satisfies the following condition:

[0184] in, Indicates the gain of load parameter information; Represents the Fischer information matrix; Represents the regularization term; Indicates the cost of micro-incentives; This represents the weighting coefficient corresponding to the cost of the micro-incentive. This indicates the cost of microslip risk; This represents the weighting coefficient corresponding to the cost of microslip risk.

[0185] In some feasible implementations, the parameter determination module 601, when determining the micro-slippage precursor probability of each foot of the robot, is used for: During the process of the robot performing the gripping operation on the load, at least two observation data are acquired: the pressure center drift velocity of the supporting foot, the redistribution rate of the normal force between the left and right feet, the ratio of tangential force to normal force, tangential wavelet energy, plantar shear phase consistency, and relative velocity of the foot tip. Based on at least two observational data points, determine the microslippage precursor feature vector of the supporting foot; The microslip precursor eigenvectors are mapped to microslip precursor probabilities.

[0186] In some feasible implementations, the parameter shrinkage module 602, when shrinking the allowable area of ​​the effective friction cone and the pressure center, is used to: The first shrinkage amount is determined based on the uncertainty of the spatial inertia of the load; wherein, the first shrinkage amount is positively correlated with the uncertainty of the spatial inertia of the load. The second contraction amount is determined based on the microslip precursor probability; wherein, the second contraction amount is positively correlated with the microslip precursor probability; Based on the first and second contraction amounts, the effective friction coefficient of the foot sole is monotonically reduced according to the probability of micro-slippage precursor, and the allowable area of ​​the pressure center is asymmetrically contracted toward the side away from the predicted slippage direction. The calculation methods for the effective coefficient of friction of the sole include:

[0187] in, Indicates the effective coefficient of friction of the sole of the foot; Indicates the nominal coefficient of friction; Indicates the coefficient of friction cone shrinkage; This indicates the probability of microslip precursors.

[0188] In some feasible implementations, the parameter shrinkage module 602, when shrinking the allowable area of ​​the effective friction cone and the pressure center, is used to: When the probability of microslip precursors is greater than or equal to the first threshold and less than the second threshold, the effective friction cone and the allowable area of ​​the pressure center are contracted. The second threshold is greater than the first threshold. The first threshold is used to trigger the contraction operation, and the second threshold is used to trigger the tangential impulse limit. The contraction operation includes contracting the effective friction cone and the allowable area of ​​the pressure center.

[0189] In some feasible implementations, the aforementioned device 600 is also used for: If the probability of microslip precursor is greater than or equal to the second threshold, it is prohibited to continue increasing the tangential impulse of the support foot in the same direction.

[0190] In some feasible implementations, the parameter shrinking module 602, when generating a feasible set sequence of multiple times within the prediction time domain, is used for: Based on the shrinkage results of the effective friction cone and the allowable area of ​​the pressure center, determine the contact constraint after shrinkage; Under the condition of satisfying the contact constraints after contraction, the whole-body state vector of the robot at each moment in the preset time domain is determined based on at least one of the following: the position of the robot's center of mass, the robot's center of mass momentum, the position information of the robot's joints, the acceleration of the robot's joints, the contact force of the supporting foot, the position of the pressure center of the supporting foot, and the uncertainty of the spatial inertia of the load.

[0191] In some feasible implementations, when the control module 603 performs whole-body coordinated control of the robot under the condition of satisfying the feasible set sequence constraint, it is used for: Under the conditions of satisfying robot dynamics, contraction contact constraints, and feasible set sequence constraints, adjust the allocation ratio between at least one of the following momentum budgets: The momentum budget required for the robot to perform upper limb operations, the momentum budget required for the robot to maintain trunk stability, and the momentum budget required for the robot to adjust support force and recover movement.

[0192] In some feasible implementations, when the control module 603 adjusts the distribution ratio between at least one of the following momentum budgets under the conditions of satisfying robot dynamics and feasible set sequence constraints, it is used to execute at least one of the following: If the uncertainty of the spatial inertia of the load meets the inertia confidence threshold and the probability of microslip precursor is less than or equal to the preset value, increase the momentum budget required for the robot to perform upper limb operations. If the uncertainty of the spatial inertia of the load does not meet the inertia confidence threshold, and the probability of microslippage precursors is greater than the preset value, the momentum budget required for the robot to maintain trunk stability and the momentum budget required for the robot to adjust the support force and recover the action are increased.

[0193] The apparatus in this application embodiment can execute the method provided in this application embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0194] Based on the same principles as the methods and apparatus provided in the embodiments of this application, this application also provides a robot 700, see [link to relevant documentation]. Figure 7 The robot includes: Control module 701 is used to execute the above-described cooperative control method.

[0195] The control module 701 of this application embodiment can execute the method provided in this application embodiment. Its implementation principle is similar. For a detailed functional description of the control module 701, please refer to the description in the corresponding method shown above. It will not be repeated here.

[0196] Based on the same principles as the methods and apparatus provided in the embodiments of this application, an electronic device (such as a server) is also provided in the embodiments of this application. The electronic device may include a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the methods provided in any optional embodiment of this application.

[0197] See Figure 8 , Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Figure 8As shown, the electronic device 800 in this embodiment may include: a processor 801, a network interface 804, and a memory 805. Furthermore, the electronic device 800 may also include: an object interface 803, and at least one bus 802. The bus 802 is used to implement communication between these components. The object interface 803 may include a display screen and a keyboard; optionally, the object interface 803 may also include a standard wired interface or a wireless interface. The network interface 804 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 805 may be a high-speed RAM or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory 805 may also be at least one storage device located remotely from the aforementioned processor 801. Figure 8 As shown, the memory 805, which is a computer-readable storage medium, may include an operating system, a network communication module, an object interface module, and a device control application.

[0198] exist Figure 8 In the electronic device 800 shown, the network interface 804 provides network communication functions; the object interface 803 is mainly used to provide an interface for input to objects; and the processor 801 can be used to call the device control application stored in the memory 805 to implement the above methods.

[0199] It should be understood that in some feasible implementations, the processor 801 described above may be a central processing unit (CPU), which may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory. For example, the memory may also store device type information.

[0200] In some feasible implementations, bus 802 can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 802 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0201] The memory 805 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation herein.

[0202] The memory 805 stores computer programs that execute embodiments of this application, and its execution is controlled by the processor 801. The processor 801 executes the computer programs stored in the memory 805 to implement the steps shown in the foregoing method embodiments.

[0203] Electronic devices include, but are not limited to: displays, image acquisition devices, cameras, antennas, etc.

[0204] In specific implementation, the aforementioned electronic device 800 can perform the above-described actions through its built-in functional modules. Figure 1 The implementation methods provided for each step are detailed in the above-mentioned implementation methods, and will not be repeated here.

[0205] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments. For details, please refer to the implementation methods provided in the above steps, which will not be repeated here.

[0206] The aforementioned computer-readable storage medium can be an internal storage unit of the apparatus or electronic device provided in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device. The aforementioned computer-readable storage medium can also include magnetic disks, optical disks, read-only memory (ROM), or random access memory (RAM), etc. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of the electronic device. The computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0207] This application also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0208] The terms "first," "second," etc., in the claims, description, and drawings of this application are used to distinguish different objects, rather than to describe a specific order.

[0209] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0210] Furthermore, those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. The terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or electronic device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or electronic device.

[0211] The reference to "embodiment" herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The inclusion of this phrase in various locations throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments. The term "and / or" as used in this specification and the appended claims means, and includes, any combination of one or more of the associated listed items and all possible combinations.

[0212] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0213] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Those skilled in the art can implement the described functions using different methods for each specific application, but such implementations should not be considered beyond the scope of this application.

[0214] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.

Claims

1. A cooperative control method, characterized in that, include: During the process of the robot performing a gripping operation on the load, the uncertainty of the spatial inertia of the load is determined, and the probability of micro-slippage precursors of each support foot of the robot is determined. Based on the uncertainty of the spatial inertia of the load and the probability of microslip precursor, the allowable region of the effective friction cone and pressure center is shrunk to generate a feasible set sequence of multiple moments in the prediction time domain; The effective friction cone includes the constraint that the ratio of the tangential contact force to the normal contact force between the supporting foot and the ground is less than or equal to the effective friction coefficient of the foot; the allowable pressure center region includes the region where the center point of the normal force distribution of the supporting foot is restricted to a preset safety boundary within the foot support surface; the feasible set sequence includes the feasible range of the robot's whole-body state vector. Under the condition of satisfying the feasible set sequence constraint, the robot is subjected to whole-body coordinated control.

2. The method according to claim 1, characterized in that, The spatial inertia of the load is used to characterize the influence of the load on linear momentum and angular momentum, and the spatial inertia of the load is obtained based on the parameter vector of the load; The parameter vector of the load includes: in, Indicates the mass of the load; These represent the mass moments of the load, where the mass moment of the load represents the offset of the load's center of mass relative to the origin of the gripping coordinate system. These represent the moments of inertia of the load about each axis of the gripping coordinate system; These represent the product of inertia of the load about each axis of the gripping coordinate system.

3. The method according to claim 1 or 2, characterized in that, The uncertainty in determining the spatial inertia of the load includes: Determine the current action phase when the robot performs a heavy-duty task based on the load; wherein the current action phase includes at least one of the following: gripping, lifting, transferring, and placing; Based on the first observation data of the robot under the current action phase, the posterior distribution of the spatial inertia of the load under the current action phase is estimated; the posterior distribution includes: the posterior mean of the spatial inertia of the load under the current action phase and the posterior covariance of the spatial inertia of the load under the current action phase; the first observation data includes at least one of the following: wrist six-dimensional force, joint torque residual and end effector acceleration. The uncertainty of the spatial inertia of the load is characterized by the posterior covariance of the spatial inertia of the load in the current action phase.

4. The method according to claim 3, characterized in that, The step of estimating the posterior distribution of the spatial inertia of the load in the current action phase based on the robot's first observation data in the current action phase includes: Based on the gripping coordinate transformation between the current action phase and the previous action phase, the posterior distribution of the spatial inertia of the load under the previous action phase is transferred to the prior distribution of the spatial inertia of the load under the current action phase. In the current action phase, based on the prior distribution, and according to the first observation data of the robot in the current action phase, the posterior distribution of the spatial inertia of the load in the current action phase is estimated.

5. The method according to claim 4, characterized in that, The step of transferring the posterior distribution of the spatial inertia of the load in the previous action phase to the prior distribution of the spatial inertia of the load in the current action phase based on the gripping coordinate transformation between the current action phase and the previous action phase includes: Based on the gripping coordinate transformation between the current action phase and the previous action phase, a coordinate transformation matrix between the first gripping coordinate system and the second gripping coordinate system is determined; wherein, the first gripping coordinate system is the coordinate system of the robot in the previous action phase; and the second gripping coordinate system is the coordinate system of the robot in the current action phase; Multiply the posterior mean of the spatial inertia of the load in the previous action phase by the coordinate transformation matrix to obtain the prior mean of the spatial inertia of the load in the current action phase. Based on the coordinate transformation matrix, the posterior covariance of the load's spatial inertia in the previous action phase, and the process noise covariance, the prior covariance of the load's spatial inertia in the current action phase is determined; wherein, the calculation method of the prior covariance includes: in, This represents the prior covariance; Represents the coordinate transformation matrix; This represents the posterior covariance of the spatial inertia of the load in the preceding action phase; The process noise covariance is represented by .

6. The method according to claim 5, characterized in that, The step of estimating the posterior distribution of the spatial inertia of the load in the current action phase based on the prior distribution and the first observation data of the robot in the current action phase includes: The observation residual of the current action phase is determined based on the difference between the first observation data and the second observation data; wherein the second observation data is the observation data predicted based on the prior mean. Based on the robot's motion state in the current action phase, determine the linear mapping matrix from the spatial inertia of the load to the observation residual; The recursive Bayesian estimation method is used to determine the posterior distribution of the load spatial inertia in the current action phase based on the prior distribution, the observation residual of the current action phase, the observation noise covariance of the current action phase, and the linear mapping matrix.

7. The method according to claim 6, characterized in that, The recursive Bayesian estimation method includes estimation based on Kalman gain. Determining the posterior distribution of the load space inertia in the current action phase based on the prior distribution, the observation residual of the current action phase, the observation noise covariance of the current action phase, and the linear mapping matrix includes: The Kalman gain is determined based on the prior covariance, the linear mapping matrix, and the observation noise covariance of the current action phase; wherein the Kalman gain includes: in, Indicates Kalman gain; This represents the prior covariance; Represents the linear mapping matrix; Represents the observation noise covariance; Based on the prior mean, the Kalman gain, the linear mapping matrix, and the observation residuals of the current action phase, the posterior mean of the current action phase is determined; wherein, the posterior mean of the current action phase includes: in, This represents the posterior mean under the current action phase; This represents the prior mean; This represents the observation residual of the current action phase; Based on the prior covariance, the Kalman gain, and the linear mapping matrix, the posterior covariance for the current action phase is determined; wherein, the posterior covariance for the current action phase includes: in, This represents the posterior covariance under the current action phase; Represents the identity matrix.

8. The method according to claim 2, characterized in that, The method further includes: Based on the load parameter information gain, the cost of micro-excitation amplitude, and the cost of micro-slip risk, the target micro-excitation is determined from the micro-excitation candidate set. Embedding the target micro-excitation into the robot's task trajectory reduces the posterior covariance of the spatial inertia of the load in the current action phase. The target micro-excitation includes at least one of short-term posture sway, velocity curvature change and hand-hand differential movement.

9. The method according to claim 8, characterized in that, The target micro-excitation satisfies the following conditions: in, Indicates the gain of load parameter information; Represents the Fischer information matrix; Represents the regularization term; Indicates the cost of micro-incentives; This represents the weighting coefficient corresponding to the cost of the micro-incentive. This indicates the cost of microslip risk; This represents the weighting coefficient corresponding to the cost of microslip risk.

10. The method according to claim 1, characterized in that, The determination of the micro-slippage precursor probability of each supporting foot of the robot includes: During the process of the robot performing the gripping operation on the load, at least two observation data are acquired: the pressure center drift velocity of the supporting foot, the redistribution rate of the normal force between the left and right feet, the ratio of tangential force to normal force, tangential wavelet energy, plantar shear phase consistency, and relative velocity of the foot tip. Based on the at least two observation data, determine the micro-slippage precursor feature vector of the supporting foot; The microslip precursor feature vector is mapped to the microslip precursor probability.

11. The method according to claim 1, characterized in that, The shrinkage of the effective friction cone and the allowable area of ​​the pressure center includes: A first shrinkage amount is determined based on the uncertainty of the spatial inertia of the load; wherein, the first shrinkage amount is positively correlated with the uncertainty of the spatial inertia of the load; A second contraction amount is determined based on the microslip precursor probability; wherein the second contraction amount is positively correlated with the microslip precursor probability; Based on the first contraction amount and the second contraction amount, the effective friction coefficient of the foot of the supporting foot is monotonically reduced according to the micro-slippage precursor probability, and the pressure center allowable area is asymmetrically contracted toward the side away from the predicted slippage direction; The calculation method for the effective coefficient of friction of the sole includes: in, This indicates the effective coefficient of friction of the foot. Indicates the nominal coefficient of friction; Indicates the coefficient of friction cone shrinkage; This represents the probability of the microslip precursor.

12. The method according to claim 1 or 11, characterized in that, The shrinkage of the effective friction cone and the allowable area of ​​the pressure center includes: If the probability of microslippage precursor is greater than or equal to the first threshold and less than the second threshold, the effective friction cone and the allowable pressure center region are contracted. Wherein, the second threshold is greater than the first threshold, the first threshold is used to trigger a contraction operation, and the second threshold is used to trigger a tangential impulse limit; the contraction operation includes contracting the effective friction cone and the pressure center allowable region.

13. The method according to claim 1, characterized in that, The method further includes: If the probability of microslippage precursor is greater than or equal to the second threshold, it is prohibited to continue increasing the tangential impulse of the support foot.

14. The method according to claim 1, characterized in that, The generation of a feasible set sequence of multiple time points within the prediction time domain includes: The contact constraint after contraction is determined based on the contraction results of the effective friction cone and the allowable area of ​​the pressure center; Under the condition of satisfying the contact constraint after contraction, the whole-body state vector of the robot at each moment in the preset time domain is determined according to at least one of the following: the position of the robot's center of mass, the momentum of the robot's center of mass, the position information of the robot's joints, the acceleration of the robot's joints, the contact force of the supporting foot, the position of the pressure center of the supporting foot, and the uncertainty of the spatial inertia of the load.

15. The method according to claim 1, characterized in that, The step of performing full-body coordinated control of the robot under the condition of satisfying the feasible set sequence constraint includes: Under the conditions of satisfying robot dynamics, contraction contact constraints, and the feasible set sequence constraints, the allocation ratio between at least one of the following momentum budgets shall be adjusted: The momentum budget required for the robot to perform upper limb operations, the momentum budget required for the robot to maintain trunk stability, and the momentum budget required for the robot to adjust support force and recover movement.

16. The method according to claim 15, characterized in that, Under the conditions of satisfying robot dynamics and the feasible set sequence constraints, the allocation ratio between at least one of the following momentum budgets is adjusted, including at least one of the following: If the uncertainty of the spatial inertia of the load meets the inertia confidence threshold and the probability of the microslip precursor is less than or equal to a preset value, the momentum budget required for the robot to perform upper limb operations is increased. If the uncertainty of the spatial inertia of the load does not meet the inertia confidence threshold, and the probability of microslippage precursor is greater than the preset value, the momentum budget required for the robot to maintain trunk stability and the momentum budget required for the robot to adjust support force and recover movement are increased.

17. A cooperative control device, characterized in that, include: The parameter determination module is used to determine the uncertainty of the spatial inertia of the load and the probability of micro-slippage precursors of each support foot of the robot during the process of the robot performing a gripping operation on the load. The parameter shrinkage module is used to shrink the allowable region of the effective friction cone and the pressure center based on the uncertainty of the spatial inertia of the load and the probability of microslip precursor, and generate a feasible set sequence of multiple times in the prediction time domain. The effective friction cone includes the constraint that the ratio of the tangential contact force to the normal contact force between the supporting foot and the ground is less than or equal to the effective friction coefficient of the foot; the allowable pressure center region includes the region where the center point of the normal force distribution of the supporting foot is restricted to a preset safety boundary within the foot support surface; the feasible set sequence includes the feasible range of the robot's whole-body state vector. The control module is used to perform whole-body coordinated control of the robot under the condition of satisfying the feasible set sequence constraints.

18. A robot, characterized in that, include: A control module for performing the method according to any one of claims 1 to 16.

19. An electronic device, characterized in that, It includes a processor and a memory, which are interconnected; The memory is used to store computer programs; The processor is configured to perform the method according to any one of claims 1 to 16 when the computer program is invoked.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 16.

21. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 16.

Images