Virtual force field balancing lever system and method based on intent recognition

Abstract

The application provides a virtual force field balance support rod system and method based on intention recognition, and relates to the field of motor control. The system comprises a multi-source perception fusion module, a base environment solving module, an orthogonal intention extraction module and a predicted virtual impedance control engine. By acquiring system posture and cross-axis current signals, the gravity and friction base torque are solved. The generalized momentum observer is used to integrate and calculate the residual of the base torque and the actual electromagnetic torque, and the external interaction torque estimate value is orthogonally stripped. The external interaction torque estimate value is input into the rotary virtual admittance model to generate the expected angle trajectory, and the explicit model predictive control is combined to position in the polyhedral state space and output the optimal compensation torque command. The application eliminates the crosstalk of environmental resistance to intention recognition and eliminates the implicit optimization delay under the condition of no force sensor, and realizes the high dynamic smooth human-computer interaction response decoupled from the disturbance and compliance.

Description

Technical Field

[0001] This application relates to the fields of motor control and human-computer interaction technology, and in particular to a virtual force field balancing strut system based on intent recognition. Background Technology

[0002] Existing electric strut systems, when operating in complex, unstructured ramp environments, typically rely on directly sampling the total motor current fluctuations to infer the user's operational intent. This mechanism results in severe aliasing between the low-frequency torque component representing the motor's ability to overcome macroscopic gravitational potential energy and the high-frequency torque component representing the user's microscopic transient operational intent, all within a single current signal dimension. Based on this aliasing, a fundamental and irreconcilable dynamic contradiction exists between the control system's ability to provide high stiffness to resist gravitational sliding disturbances and to provide low stiffness to accommodate user push-pull operations. Conventional current differential control architectures or low-pass filtering methods not only fail to physically decouple this crosstalk but also introduce significant control phase delays, ultimately leading to frequent misjudgments and severe mechanical response stuttering during human-machine interaction. Summary of the Invention

[0003] Firstly, this application provides a virtual force field balancing strut system based on intent recognition. The system includes: a multi-source sensing fusion module for acquiring multi-source physical state signals of the system, the multi-source physical state signals including the pitch angle, pitch angular velocity, and quadrature-axis component of the motor stator current relative to an absolute reference plane; a base environment calculation module for acquiring fixed physical parameters of the system, and constructing and outputting a base torque characterizing gravity and inherent resistance under the current spatial pose based on the pitch angle, the pitch angular velocity, and the fixed physical parameters; and an orthogonal intent extraction module for... The actual electromagnetic torque of the motor is obtained based on the quadrature-axis component of the motor stator current. Then, through a preset generalized momentum observer, the estimated value of the external interactive torque after orthogonally stripping environmental interference is extracted based on the residual integral of the base torque and the actual electromagnetic torque of the motor. A predictive virtual impedance control engine is used to input the estimated value of the external interactive torque into a preset rotating virtual admittance model to generate a desired angle trajectory. Then, through an explicit model predictive controller, the desired angle trajectory is located in a pre-divided polyhedral state space based on the difference between the desired angle trajectory and the current pitch angle, and a compensation torque command is output.

[0004] Secondly, this application provides a virtual force field balancing strut method based on intent recognition, applied to the system described in the first aspect above. The method includes: acquiring multi-source physical state signals of the system, the multi-source physical state signals including the pitch angle, pitch angular velocity, and cross-axis component of the motor stator current relative to an absolute reference plane; acquiring fixed physical parameters of the system, and constructing and outputting a base torque characterizing gravity and inherent resistance under the current spatial pose based on the pitch angle, the pitch angular velocity, and the fixed physical parameters of the system; acquiring the actual electromagnetic torque of the motor based on the cross-axis component of the motor stator current, and extracting an estimated value of the external interactive torque after orthogonal stripping of environmental interference based on the residual integral of the base torque and the actual electromagnetic torque of the motor through a preset generalized momentum observer; inputting the estimated value of the external interactive torque into a preset rotating virtual admittance model to generate a desired angle trajectory, and locating and outputting a compensation torque command in a pre-divided polyhedral state space based on the difference between the desired angle trajectory and the current pitch angle through an explicit model predictive controller.

[0005] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the second aspect above.

[0006] This application has the following beneficial effects:

[0007] This application establishes a generalized momentum observer architecture without sensor constraints, strictly orthogonally separating the system load into a slowly varying environmental fundamental field and a transient user intent field in the Lagrange dynamic space. This mechanism reduces the dimensionality of traditional impedance control, enabling explicit model predictive control algorithms to drive a pure intent field in real time under limited microcontroller computing power. This transformation not only eliminates crosstalk and false alarms caused by environmental disturbances such as slope to intent recognition, but also eliminates online optimization latency through offline multi-parameter planning, compressing the response time of intent inference and trajectory planning to the microsecond level, reconstructing the dynamic response boundary of the system, and achieving decoupling of high disturbance immunity and high compliance. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0009] Figure 1 A structural block diagram of a virtual force field balancing strut system based on intent recognition provided in an embodiment of this application;

[0010] Figure 2 A flowchart illustrating a virtual force field balancing strut method based on intent recognition, provided for embodiments of this application;

[0011] Figure 3 This is a schematic diagram illustrating the dynamic simulation results of a system under external interactive torque, provided in an embodiment of this application. In the diagram, a) is the simulation diagram for observing the intended torque, and b) is the simulation diagram for tracking the desired trajectory. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0014] This embodiment provides a virtual force field balancing strut system based on intent recognition. In a specific implementation, this method employs an architecture that combines a generalized momentum observer with an explicit model predictive controller, thereby achieving strict dimensionality reduction and decoupling of the physical torque space and delay-free motion tracking. This method solves the problem in existing technologies where the total motor current is difficult to distinguish between static environmental resistance and transient human intent, leading to control jerks. It achieves the beneficial effect of constructing a smooth and error-free elastic virtual force field without adding physical sensors.

[0015] refer to Figure 1 As shown, the virtual force field balancing strut system based on intent recognition in this embodiment includes multiple independent hardware and logic execution units. These multiple independent hardware and logic execution units include a multi-source perception fusion module, a base environment calculation module, an orthogonal intent extraction module, a predictive virtual impedance control engine, and an adaptive calibration and parameter calibration module.

[0016] The multi-source sensing fusion module is configured to acquire, filter, and reconstruct the state signals of the underlying physical hardware. Deployed at the front end of the strut drive control board, the module consists of a microcontroller with direct memory access channels, a six-axis microelectromechanical system (MEMS) inertial measurement unit (IMU) chip, and a three-channel phase current sampling circuit. Internally, the module includes a high-frequency attitude calculation subunit. This subunit executes Kalman filtering code to remove high-frequency noise from the accelerometer signals, fusing a smooth tilt angle sequence. It also includes a field-oriented control current reconstruction subunit. This subunit configures the analog-to-digital converter (ADC) timer and pulse width modulation (PWM) timer to be aligned, triggering sampling and performing coordinate transformation matrix operations at pulse troughs. The module maintains a sensor snapshot frame structure in its internal static random access memory. For example, the sensor snapshot frame structure includes an unsigned 64-bit integer timestamp field for recording absolute time, a double-precision floating-point field for storing pitch angle data, a double-precision floating-point field for storing angular velocity data, and a double-precision floating-point field for storing stator quadrature-axis current data.

[0017] The base environment calculation module is configured to reconstruct the objective physical impedance experienced by the system at its current pose based on spatial geometric relationships. The base environment calculation module operates independently using the hardware floating-point unit of the microcontroller. The base environment calculation module includes a gravity gradient mapping subunit. This subunit calculates the gravitational torque component corresponding to the current strut angle using a trigonometric function library. The base environment calculation module also includes a dynamic Coulomb drag subunit. This subunit extracts the polarity sign of the angular velocity and calculates the dry friction drag with a constant opposite direction based on the friction model. The base environment calculation module maintains a system parameter dictionary structure in the memory address space. For example, the system parameter dictionary structure contains values... Total physical mass of the system (dimensions are) ), value physical distance of the centroid (dimensions are) ) and the value Coulomb friction torque constant (dimensions are) ).

[0018] The orthogonal intent extraction module is configured to strip away environmental effects and quantify purely external interaction torques. The orthogonal intent extraction module includes a discretized momentum integrator subunit. This subunit performs cumulative iterative calculations of the one-dimensional difference equation. The orthogonal intent extraction module also includes a drift compensation subunit. This subunit tracks and eliminates thermal drift zero-point deviations in the current sensing circuit through closed-loop feedback. The orthogonal intent extraction module maintains a dynamic residual tracking tensor. Exemplarily, this dynamic residual tracking tensor is a 128-byte double-precision floating-point one-dimensional circular buffer array. Each element in the array stores a joint state snapshot containing the system's generalized momentum value, the motor's actual electromagnetic torque value, and the basis torque value within a specific control cycle, to support data backtracking within the integration window.

[0019] The predictive virtual impedance control engine is configured to formulate a dynamic feedforward strategy in response to the external interactive torque. The predictive virtual impedance control engine includes an admittance trajectory generation subunit. This subunit is used to generate a reference motion trajectory based on the discrete solution of a second-order ordinary differential equation. The predictive virtual impedance control engine also includes a polyhedral addressing query subunit. This subunit is used to perform a point location search algorithm in a multidimensional state-space geometry. The predictive virtual impedance control engine accesses an explicit control law addressing tree embedded in non-volatile flash memory. Exemplarily, the explicit control law addressing tree consists of 2048 interconnected binary tree nodes, each leaf node storing the coefficients of a specific polyhedral constraint hyperplane matrix equation and the associated optimal affine feedback gain vector matrix.

[0020] To achieve consistent declarations of system variables, this application establishes a dimensional system encompassing multiple dimensions. This dimensional system defines physical quantities characterizing rotational inertia, using kilograms per square meter. As a basic unit, it is used to accurately describe the resistance characteristics of torque when it induces angular acceleration. The dimensional system defines physical quantities characterizing rotational kinematic states, using radians. radians per second and radians per square second As a basic unit, it is used to eliminate the conversion coefficients in the Taylor expansion calculation of trigonometric functions. The dimensional system defines physical quantities used to characterize the rate of energy transfer and the intensity of action, using the Newton-meter unit. As the basic unit, the system is used in all dynamic discrete-difference calculations through a defined time step. (dimensions are) Perform integration to ensure that the dimensions on both sides of the equation are absolutely balanced after algebraic operations.

[0021] Combination Figure 2 As shown in the figure, this application provides a virtual force field balancing strut method based on intent recognition, applied to the above system. The method includes the following steps.

[0022] S000. Initialization calibration is performed during the system's first power-on or self-test. Current and tilt angle data for the entire stroke under no-load conditions are collected, and parameter identification is performed using the least squares method to generate a system dynamics reference parameter dictionary stored in memory.

[0023] In one specific embodiment, S000 is automatically executed by the adaptive calibration and parameter calibration module. The microcontroller sends a constant angular velocity reference command to the underlying motor driver. The strut performs a uniform motion from a fully closed state to a fully extended state without external human intervention. During this uniform motion, the multi-source sensing fusion module records an array of data at high frequency at 10-millisecond intervals. The array of data contains... The sequence of stator current quadrature-axis components and pitch angle sequence at discrete time points. The adaptive calibration and parameter calibration module constructs an overdetermined linear equation matrix. Among them, the observation vector Each element is the total electromagnetic torque data at the corresponding time point, and the regression matrix is... The column vector is composed of the gravity geometric projection coefficients and angular velocity signs calculated based on the pitch angle sequence. The adaptive calibration and parameter calibration module calls the built-in matrix inversion function to calculate the pseudo-inverse matrix. and the results with Multiplying them together separates the parameter vector to be identified. For example, the parameter vector Includes the identified total system mass (dimensions are) and Coulomb friction torque constant (dimensions are) The identification results are serialized and saved to the memory to form the system parameter dictionary structure, eliminating model deviations caused by assembly mechanical tolerances.

[0024] S100. Acquire multi-source physical state signals of the system, wherein the multi-source physical state signals include the pitch angle, pitch angular velocity and cross-axis component of the motor stator current relative to the absolute reference plane.

[0025] The multi-source sensing fusion module periodically performs physical signal conversion and data extraction operations under the drive of a determined clock interrupt.

[0026] S110: Read the three-axis acceleration signal and three-axis angular velocity signal through the serial peripheral interface bus, and use the attitude calculation filtering algorithm to generate the pitch angle and pitch angular velocity sequence at the current moment and store it in the circular buffer.

[0027] The multi-source sensing fusion module sends a read request command to the register address of the six-axis inertial measurement unit via a hardware serial peripheral interface. The multi-source sensing fusion module receives the returned 16-bit signed integer format raw three-dimensional acceleration and raw three-dimensional angular velocity data. The high-frequency attitude calculation subunit multiplies the raw data by the factory-defined range resolution coefficient, converting it into a floating-point signal with standard physical dimensions.

[0028] S111. Construct a predictive step model of the system's state-space equations and perform prior estimation calculations.

[0029] The high-frequency attitude calculation subunit instantiates a discrete Kalman filter model in the microcontroller. The high-frequency attitude calculation subunit defines a two-dimensional state vector. ,in Represents the absolute pitch angle (dimensions: 1 / 2π). ), This indicates the zero-bias drift of the gyroscope (dimension: 1). The high-frequency attitude calculation subunit defines the state transition matrix. Control input matrix The control input is the rotational rate of the measured three-dimensional angular velocity signal mapped to the pitch axis. The high-frequency attitude calculation subunit performs prior state estimation calculations. Simultaneously, the high-frequency attitude calculation subunit reads the set process noise covariance matrix. And perform prior covariance matrix prediction. .

[0030] For example, setting the clock period (dimensions are) Assuming at the previous moment... posterior state vector posterior covariance matrix The gyroscope measurement value acquired at the current moment. (dimensions are) Process noise covariance matrix The high-frequency attitude calculation subunit performs matrix multiplication and addition calculations to obtain the prior state vector. The high-frequency attitude calculation subunit continues to calculate the prior covariance matrix. .

[0031] S112. Construct the update step model of the system measurement equations and perform Kalman gain calculation and posterior state correction.

[0032] The high-frequency attitude calculation subunit defines a measurement matrix. And from the three-dimensional acceleration signal through inverse trigonometric functions Extracting the measured pitch angle containing high-frequency noise The high-frequency attitude calculation subunit reads the set measurement noise covariance matrix. The high-frequency attitude calculation subunit performs Kalman gain matrix calculation. Subsequently, the high-frequency attitude calculation subunit performs a posteriori state update. And update the posterior covariance matrix. .

[0033] For example, assume the extracted measured pitch angle (dimensions are) ), measuring noise constant Calculate the gain denominator Calculate the Kalman gain matrix. Calculate the measurement residuals. The updated state vector The first element in the state vector (dimensions are) That is, the extracted smooth pitch angle after noise reduction. Meanwhile, the corrected angular velocity (dimensions are) These two precise values ​​are stored in memory for use by downstream systems.

[0034] To ensure the reproducibility and optimal performance of the discrete Kalman filter model, its core parameter is the process noise covariance matrix. Covariance matrix of measurement noise The value is determined based on a clear physical calibration method, rather than being subjectively set.

[0035] Specifically, measuring the noise covariance constant. The calibration process is as follows: With the system stationary and free from external interference, continuous data is collected. (For example, The pitch angle measurement value directly calculated from the accelerometer Since the system is stationary, the fluctuations in these measurements can be considered pure sensor measurement noise. Calculate this... Statistical variance of a sample This variance value is used as the measurement noise covariance constant. The final value of . For example, if the calculated variance of the collected samples is . Then set .

[0036] Process noise covariance matrix The calibration reflects the degree of confidence in the uncertainty of the prediction model. In a simplified calibration process, it can be set based on an assessment of the gyroscope's zero-bias drift characteristics. By recording the integral drift of the gyroscope's angular velocity output during a long period of system stillness, the strength of the angular velocity random walk can be evaluated, and adjustments can be made accordingly. The values ​​of elements in the matrix related to angular velocity offset. In practical engineering debugging, iterative optimization methods are often used: initially given a small value... If the filter output angle is found to be too sensitive to the gyroscope's integral drift (i.e., sluggish dynamic response), then the value should be increased appropriately. The value is adjusted to increase the confidence weight of the accelerometer measurements, and vice versa. This systematic calibration process ensures that those skilled in the art can reproduce attitude calculation systems with similar filtering performance without excessive experimentation.

[0037] S120: The analog voltage of the three-phase current sampling of the drive circuit is converted by the analog-to-digital converter, and the coordinate transformation is performed to extract the stator current quadrature-axis component related to the actual output electromagnetic torque of the motor.

[0038] The analog-to-digital conversion peripheral inside the multi-source sensing fusion module is started and maintained under the control of the timer synchronous trigger signal.

[0039] S121. Configure hardware trigger source to extract instantaneous current ampere values ​​for each phase.

[0040] The multi-source sensing fusion module is configured with a center-aligned mode counter of an advanced control timer. At the counter's overflow peak, i.e., the trough of the pulse width modulation (PWM) wave when the lower arm of the three-phase inverter is fully conducting, a hardware trigger event is sent to the analog-to-digital converter (ADC). The ADC quantizes the amplified lower arm sampling resistor voltage signal into a digital code. The field-oriented control current reconstruction subunit reads the digital code, subtracts the pre-calibrated reference bias voltage code, and divides it by the hardware gain coefficient and the sampling resistor value to calculate the physical phase current value of the motor. For example, the instantaneous phase current obtained after conversion is: (dimensions are) ), (dimensions are) ), (dimensions are) It satisfies Kirchhoff's current law. .

[0041] S122. Perform Clarke coordinate transformation.

[0042] The magnetic field-oriented control current reconstruction subunit performs a Clark matrix multiplication transformation on the physical phase current value, transforming it from the three-phase stationary coordinate system (…). Mapped to a two-phase orthogonal stationary coordinate system ( The constant power transformation matrix coefficients used are: The calculation formula is: as well as For example, substituting the above current values, the calculation yields... (dimensions are) ). Calculated (dimensions are) ).

[0043] S123. Extract the absolute electrical angle and perform Park coordinate transformation to strip the stator current quadrature component.

[0044] The field-oriented control current reconstruction subunit obtains the absolute mechanical position of the motor rotor through a dedicated encoder interface and multiplies it by the number of pole pairs of the motor. Obtaining electrical angle The magnetic field orientation control current reconstruction subunit is based on the formula... and Perform a rotational coordinate transformation. For example, assume the electrical angle currently read and calculated... (dimensions are) (i.e., 60 degrees). Corresponding to Perform floating-point multiplication and addition operations to obtain the excitation direct-axis current. (dimensions are) Perform floating-point multiplication and addition operations to obtain the torque quadrature-axis current. (dimensions are) The stator current quadrature axis component. It was then encapsulated in the cache of the sensor snapshot frame structure.

[0045] S200. Obtain the system's fixed physical parameters, and based on the pitch angle, the pitch angular velocity, and the system's fixed physical parameters, construct and output the base torque characterizing gravity and inherent drag under the current spatial pose; obtain the motor's actual electromagnetic torque based on the motor stator current quadrature-axis component, and extract the estimated external interactive torque after orthogonally stripping environmental interference based on the residual integral of the base torque and the motor's actual electromagnetic torque using a preset generalized momentum observer.

[0046] The underlying environment calculation module and the orthogonal intent extraction module logically constitute a pre-processing dynamics data pipeline. This pipeline operates at a determined downsampling frequency, achieving physical-layer reconstruction of pure state information by eliminating the risk of digital aliasing introduced by high-frequency sampling.

[0047] S210. Based on the explicit equation of gravity and friction, perform algebraic calculations to obtain the base torque characterizing the environmental impedance.

[0048] Within a defined main control cycle, the base environment calculation module extracts the continuous historical snapshot frames stored in the sensor snapshot frame structure.

[0049] S211, Perform clock domain alignment and arithmetic mean filtering.

[0050] The execution cycle of the base environment calculation module is set to 2 milliseconds (corresponding to a 500Hz frequency), while the sampling refresh cycle of the upstream multi-source sensing fusion module is 1 millisecond (corresponding to a 1000Hz frequency). To avoid frequency aliasing caused by Shannon's sampling theorem during downsampling, the base environment calculation module continuously extracts the two most recently written adjacent moments from the memory direct access controller's circular buffer each time it is woken up by the operating system scheduler. and The sensor snapshot frames. The base environment calculation module calculates the pitch angle field within these two snapshot frames. and A constant that performs double-precision floating-point addition and divides by 2.

[0051] For example, if the extracted continuous angles are and (All dimensions are...) The calculated smooth pitch angle is then obtained. for (dimensions are) The same smoothing operation is applied to the pitch angular velocity field to generate... This operation achieves first-order low-pass anti-aliasing filtering with extremely low computational overhead, ensuring the smoothness requirements of the input signal for downstream dynamics calculations.

[0052] S212. Construct a base environment damping reconstruction matrix containing triangular mapping and output the base torque.

[0053] The gravity gradient mapping subunit calls the approximate polynomial expansion instruction set built into the hardware floating-point arithmetic unit for calculation. The floating-point truth value. The underlying environment calculation module reads the fixed physical constants from the static memory address corresponding to the system parameter dictionary structure through pointer offset. The underlying environment calculation module executes the formula. .

[0054] For example, the physical gravitational acceleration constant set by the system Fixed as (dimensions are) ) Memory addressing reads system quality (dimensions are) Centroid Distance (dimensions are) ), and Coulomb friction constant (dimensions are) When the angle is smoothed (dimensions are) (corresponding to approximately 30 degrees), its sine value is When the smoothed angular velocity (dimensions are) When the direction of motion is positive, the sign function... Returns an integer value The base environment calculation module executes the gravitational torque calculation component: (dimensions are) Then, the friction torque calculation is performed: (dimensions are) Finally, the two components are added using floating-point addition to output the base torque. (dimensions are) ).

[0055] S220: Calculate the actual electromagnetic torque of the motor through multiplication.

[0056] The orthogonal intention extraction module receives the stator current quadrature-axis component contained in the sensor snapshot frame and also subjected to double-buffered average filtering. The orthogonal intent extraction module locates and reads the motor torque constant, which characterizes the motor's output torque and current conversion efficiency. Based on the linear proportional relationship between the motor torque constant and the stator current quadrature-axis component, the orthogonal intention extraction module executes a single-cycle floating-point multiplication instruction in the arithmetic logic unit. For example, assuming a fixed motor torque constant... (dimensions are) ), and the currently smoothed received (dimensions are) Then, through the formula Calculated actual electromagnetic torque of the motor The value is (dimensions are) ).

[0057] S230: Import the generalized momentum observer to perform momentum residual integral calculation, remove environmental influences, and obtain an accurate estimate of the external interaction torque that represents the user's pulling or pressing action.

[0058] The orthogonal intent extraction module instantiates a generalized momentum observer model derived from the Newton-Euler nonlinear coupling equations. This model avoids interference with high-noise acceleration signals. Direct use.

[0059] S231, Momentum state memory and rate of change assessment based on ring buffer.

[0060] The orthogonal intent extraction module allocates a circular buffer queue of length N in contiguous memory space to store physical pointer structures. In each discrete control cycle... When triggered, the module extracts the angular velocity at the current moment. and multiply it by the preset total physical moment of inertia of the system. To obtain the current generalized momentum (dimensions are) Subsequently, the module reads the historical moment by shifting the pointer one position forward. Stored generalized momentum The processor performs the interpolation operation. And divide numerically by the system's discrete time step It accurately extracts the instantaneous rate of change of the physical momentum of the current system.

[0061] S232. Implement composite residual assessment and smooth integral compensation that includes multiple torque dimensions.

[0062] The orthogonal intent extraction module is based on a continuous time-domain observation model. The core solution is performed on the mapped backward Euler difference discretization equations. The iterative formula is set as follows: .in The observer gain factor is not set empirically, but determined through a systematic pole placement method to ensure that the observer has a well-defined dynamic response characteristic. The poles of the error dynamic equation of this first-order observer are located at... To ensure the observer can quickly and stably track external torques applied by the user, typically with bandwidths below 5 Hz, while effectively suppressing higher-frequency sensor noise, we will adjust the observer's closed-loop bandwidth poles. This bandwidth is set to a value much higher than the target tracking bandwidth, for example, 100 rad / s (approximately 16 Hz). This bandwidth directly determines the reciprocal of the time constant for the observer to converge to the truth. Therefore, the observer gain factor is directly set to this bandwidth value, i.e. (dimension is) This setting ensures that the estimation error can be kept within approximately five time constants (i.e., It converges to zero within milliseconds, achieving an effective balance between response speed and noise robustness, enabling those skilled in the art to reproduce the observer without ambiguity.

[0063] For example, we perform a process spanning three consecutive micro-time steps ( Micro-numerical simulations were performed to verify its convergence and stripping characteristics. Known parameters included: moment of inertia. (dimensions are) ), Integral step size (dimensions are) ), observer gain (dimensions are) Under this high-gain configuration, the coefficient .

[0064] At the initial moment The system maintains uniform motion, has no external intention, and its state value is... , Base torque Total torque of the motor .

[0065] In the cycle The user suddenly applied a constant external force, causing the physical momentum to rise to [a certain value]. The input obtained at this moment from the previous moment: , , .

[0066] Perform the calculation of the rate of change of momentum in S231: (dimensions are) ).

[0067] Perform residual integral calculation using S232: (dimensions are) ).

[0068] In the cycle Due to the user's continued application of force, the motor current controller did not fully respond, and the momentum continued to rise to The previous input was updated to: , , .

[0069] Perform the calculation of the rate of change of momentum in S231: (dimensions are) ).

[0070] Perform residual integral calculation using S232: (dimensions are) ).

[0071] Through this integral recursive approximation based on extremely short time slices, the module effectively transforms signals containing various noise sources into a clean and progressively converging intention curve. The estimated external interaction torque value... Logically equivalent to the output bus variable middle.

[0072] To address the environmental robustness requirements of the system during long-term operation, this application provides a temperature drift closed-loop compensation strategy as an independent functional mechanism.

[0073] During the continuous acquisition of phase current by the multi-source sensing fusion module, the heating of the inverter power devices and motor windings causes a nonlinear increase in the hardware resistance parameters. If not corrected, under a fixed motor torque constant, the increase in current will be misinterpreted by the orthogonal intent extraction module as a constant "false reverse intent".

[0074] The temperature drift closed-loop compensation strategy is defined as a closed-loop correction algorithm that includes low-frequency state monitoring and dynamic parameter calibration. The strategy explicitly receives the raw analog-to-digital converted voltage value from the chip temperature sensor built into the microcontroller, as well as the baseline current observation vector of the motor in a stationary state, as multi-dimensional inputs. The microcontroller maps the raw voltage value from the temperature sensor to a Celsius temperature value. When the system state machine determines, via a time counter, that the support pole has been in a stationary, locked state with zero speed and not held by the user for a preset number of frames, the compensation logic is activated. The microcontroller uses the linear demagnetization attenuation formula... Adjust the motor torque constant. Among these adjustments... This refers to the irreversible demagnetization decay coefficient specific to permanent magnet materials. For example, the current temperature value is monitored as input. The reference calibration temperature value is stored in the read-only memory. for Preset attenuation coefficient (Dimensionless). The algorithm performs subtraction and multiplication to obtain the attenuation compensation ratio, representing the percentage of temperature deviation. The system extracts the initial reference motor torque constant. The attenuation compensation ratio is multiplied. The temperature drift closed-loop compensation strategy ultimately outputs the model parameter values ​​updated with precise derating. (dimensions are) The output is directly overwritten to the corresponding field of the system parameter dictionary structure in memory via pointer indirect addressing. This closed-loop correction operation ensures the absolute objectivity of the algebraic calculation of actual electromagnetic torque under extreme high-temperature conditions.

[0075] Similarly, this application constructs an intent-triggered threshold verification mechanism with clear statistical judgment logic, the purpose of which is to completely eliminate the false activation of the force field caused by electronic white noise.

[0076] To prevent minute force field oscillations caused by the inherent noise of the current sensor when the system is stationary, the system determines the zero-point boundary based on a probabilistic statistical model rather than manually set dynamic settings. Within an initial noise sampling period of 2 seconds (preset duration), the system records... A sample set is formed by estimating the intentional moment values ​​under conditions of no external human interference. The system's scheduling mathematical processing library functions perform standard deviation and expectation calculations on this sample set to determine if it conforms to a Gaussian distribution and has a mean. and statistical standard deviation The system calculates the intent trigger threshold that meets the three sigma confidence criterion. In each main control loop, the logical judgment instruction reads the absolute value of the extracted external interaction torque estimate. Then, a floating-point comparison operation is performed between it and the trigger threshold variable.

[0077] For example, suppose that the statistical mean of the static noise floor is calculated. (dimensions are) ), standard deviation (dimensions are) The system performs scalar addition calculations and sets the intent trigger threshold. (dimensions are) When the latest extracted external interaction torque module output is When the floating-point comparison instruction returns false, the system determines it as a white noise environmental disturbance, and the locking force field response output is zero; when the external interaction torque rises sharply to When the comparison instruction returns True, the trigger output enable flag is set to a hexadecimal high level of 0xFF. This enable flag is written to the register as a Boolean output instruction to activate the state machine transition of the subsequent predictive virtual impedance control engine.

[0078] S300. Input the estimated value of the external interactive torque into the preset rotational virtual admittance model to generate the desired angle trajectory, and through the explicit model prediction controller, locate and output the compensation torque command in the pre-divided polyhedral state space based on the difference between the desired angle trajectory and the current pitch angle.

[0079] The predictive virtual impedance control engine, as a highly integrated advanced control kernel, transforms the extracted pure user intent into a dynamic response sequence that characterizes specific tactile physical properties, and incorporates the optimal control law based on the multi-physics boundary constraints of the motor hardware.

[0080] S310. Input the estimated value of the external interaction torque into the preset rotational virtual admittance model, and generate the desired angle trajectory and its derivative by solving the discretized Euler integral.

[0081] The admittance trajectory generation subunit incorporates an ordinary differential dynamic mapping mathematical model based on a second-order mechanically damped spring oscillator system. The model receives the scalar external interaction torque in an active state. It serves as the sole external excitation source input for the system equations.

[0082] S311. Configure and instantiate the multi-degree-of-freedom virtual impedance parameter space.

[0083] The admittance trajectory generation subunit defines three core state configuration parameters in memory for reshaping the dynamic characteristics of the strut. One of these is the virtual moment of inertia, which determines the initial resistance and acceleration hysteresis characteristics. (dimensions are) Secondly, it is used to provide energy dissipation, limit the maximum steady-state velocity, and simulate the smooth feel of advanced hydraulic hinges. (dimensions are) Thirdly, it is used to provide virtual rotational stiffness for simulating position recovery trends and elastic potential energy. (dimensions are) Different combinations of these parameters instantiate different abstract scenario application patterns.

[0084] S312. Solving the virtual physical response based on the explicit integral difference model.

[0085] The admittance trajectory generation subunit is based on the constructed continuous-time domain rotational admittance equation. Discretization rewriting and numerical iterative difference solving are performed in the arithmetic logic unit of the microcontroller. During computation, the algorithm explicitly solves for the angular acceleration variable by rearranging the terms.

[0086] For example, a complete parameter and numerical calculation step-by-step operation is performed. To construct a "low-resistance, high-speed responsive hovering" virtual feel mode suitable for elderly users, the system logic incorporates virtual stiffness... Reset to zero (dimensions are) To eliminate the illusion of heavy physical inertia caused by the large size of the metal struts, the system uses virtual inertia. Manually reduce the value to the minimum. (dimensions are) Simultaneously, the system will implement virtual damping. Set appropriately (dimensions are) This ensures that the smooth, sticky feel is not produced when the movement stops.

[0087] Set the current microcontroller clock to enter a certain discrete time step. The estimated torque value of the current input intent is read from memory. (dimensions are) The expected angle from the previous cycle is read from memory. (dimensions are) and desired angular velocity (dimensions are) ).

[0088] First operational stage: The microcontroller subtractor calculates the effective driving torque term applied to the virtual inertia after removing damping and elastic force. Effective driving torque. (dimensions are) ).

[0089] Second operation phase: The microcontroller divider uses the effective driving torque to calculate the desired angular acceleration generated in the current cycle. (dimensions are) ).

[0090] Third operation stage: The microcontroller multiplier and adder utilize the integration constant (assuming...) Performing a first-order numerical Euler integral on the acceleration (in seconds) yields the updated desired angular velocity. (dimensions are) ).

[0091] Fourth operation stage: Perform a double integral on the angular velocity to obtain the desired angular trajectory position point. (dimensions are) This coherent multi-step pipelined computation ensures that the reference trajectory output by the virtual model does not undergo any abrupt changes in the mathematical space, possessing absolute smoothness, continuity, and physical plausibility.

[0092] S320: Through the explicit model predictive controller, based on the preset motor output extreme value constraints, the optimal feedforward control law is found in the pre-divided polyhedral state space, and a sequence of compensation torque commands in the future prediction time domain is generated.

[0093] The explicit model predictive control mechanism employed in this application is based on solving a computationally complex constrained optimization problem offline in a single step, and storing the solution (i.e., the control law) as a partitioned affine function form that can be queried online at high speed. This offline solution process is based on a precisely mathematically defined multi-parameter quadratic programming problem (mp-QP), which consists of three core elements: the system prediction model, the cost function to be optimized, and the physical constraints.

[0094] First, the system prediction model is constructed as a discrete-time state-space equation describing the dynamic evolution of the tracking error. Among them, the state vector These refer to the angle tracking error and angular velocity tracking error defined in S321. Control input. The compensation torque command output for the current cycle State matrix and input matrix It is obtained by discretization based on Newton's second law, for example, for a system with a total moment of inertia. The dominant second-order system, in the sampling period Down, , .

[0095] Second, the cost function to be optimized is defined as a function in the finite prediction time domain. Quadratic objective functional within .in, and It is a positive definite weight matrix. Used to penalize state errors, the larger the diagonal element, the more forcefully the system will eliminate the corresponding error. The size of the penalty control input is used; a larger value means the system tends to use less torque and complete the tracking task more energy-efficiently. For example, it can be set... Prioritize ensuring angle tracking accuracy and set To allow for larger instantaneous control output.

[0096] Third, physical constraints are defined as a series of linear inequalities used to define the safety boundaries of the system operation. These constraints mainly include limitations on control inputs, such as the peak output torque constraint of the motor. (For example, (N·m), and restrictions on system states, such as the maximum permissible tracking error.

[0097] By inputting the above model, cost function, and constraints into a dedicated multi-parameter programming solver, a "solution graph" covering the entire state space can be generated offline, namely the pre-compiled polyhedral constraint addressing tree described in S322, thereby reducing the online computational load to a predictable and extremely low level.

[0098] Traditional implicit model predictive control strategies heavily rely on the microprocessor to repeatedly call interior-point or effective-set methods to solve the convex quadratic programming objective functional within extremely short control cycles down to the millisecond level. This inevitably leads to severe timeouts and crashes in control timing due to chip computational bottlenecks. This application employs an explicit model predictive control mechanism based on multi-parameter programming theory. It transfers the massive, high-intensity constraint optimization iterative search process to the offline compilation stage outside the chip. Thus, during the online embedded execution stage, the extremely complex matrix inversion and optimization search are reduced to a single high-speed spatial geometric partitioning condition addressing decision based on data structures.

[0099] S321. Construct the tracking error state vector and perform multi-dimensional state evaluation.

[0100] In the online execution step of the explicit model predictive control engine, the polyhedral addressing query subunit receives multiple real-time physical variables from the upstream module. The subunit first extracts the desired angle trajectory output from the virtual admittance model. And the pitch angle calculated from the actual attitude. The arithmetic logic unit performs difference calculations to obtain the system tracking deviation error. Similarly, calculate the tracking deviation rate of change error of the angular velocity. The polyhedral addressing query subunit packages and incorporates these two error components into the high-speed register group to construct a one-dimensional real-time state evaluation vector representing the current control situation. .

[0101] S322. Perform condition determination and topology shuttle search in the precompiled polyhedral constraint addressing tree.

[0102] In the main flash memory area of ​​the system, an explicit control law addressing tree data structure, generated offline by an offline multi-parameter quadratic programming solver, is pre-programmed. This structure indicates that the state space has been divided into thousands of states that satisfy different control boundaries (such as peak phase current not exceeding a certain threshold). The discrete convex polyhedral region set (e.g., bus input voltage constraints, etc.).

[0103] The polyhedral addressing query subunit initializes its memory pointer to the starting physical address of the root node of the binary search tree. At the current access depth, the subunit extracts the row of the normal vector matrix of the boundary hyperplane inequality system corresponding to the current node through the direct memory access mechanism. and the corresponding boundary intercept scalar The floating-point multiply-accumulate unit of the subunit executes the one-dimensional real-time state evaluation vector. With the normal vector High-speed matrix-vector dot product judgment operation By performing a floating-point comparison instruction between the dot product result and the intercept scalar: if the result is less than or equal to... This indicates that the current state is located in the left half of the open space defined by the hyperplane, and the pointer moves and jumps to the left subtree branch; if it is greater than, the pointer moves to the right subtree branch. The above topology shuttle logic continuously loops and progresses within a finite and extremely short clock cycle until the memory traversal pointer falls into the terminal leaf data area without child nodes, indicating that the unique target polyhedral region index number that satisfies all multidimensional physical envelope constraints has been accurately captured and located in all polyhedral region sets.

[0104] S323. Extract the target region parameters and execute affine feedback feedforward calculation to generate the absolute optimal compensation instruction.

[0105] When the addressing process safely terminates in a terminal leaf polyhedral region, the polyhedral addressing query subunit extracts pre-fixed affine feedback coefficients bound to that region from a contiguous compact memory block of that leaf node. These coefficients include a control gain matrix. and a constant bias vector Subsequently, the prediction engine's operator executes a single matrix algebraic mapping equation. Calculate the final physical control action directly.

[0106] A small data validation example is provided to illustrate the dimensionality reduction process of this instruction. Assume that after 12 conditional jump comparisons, the pointer is positioned on the 88th convex polyhedron. The program then extracts the feedback coefficient matrix for region 88 from flash memory. Contains two weight elements Bias constant Values The one-dimensional real-time state evaluation vector buffered in the current register is obtained, which is the tensor array composed of the position error and velocity error. The microcontroller feeds the above values ​​into a hardware multiplier-accumulator to perform a dot product merging operation: Add the result of the multiplication and addition to the bias scalar: The deterministic floating-point output result This is defined as the current output scalar term in the compensated torque command sequence generated under the boundary conditions of motor maximum safe current saturation and dynamic voltage drop constraints, which has a globally optimal time response cost. (dimensions are) ).

[0107] S400: The base torque and the compensation torque command are added together and fused, and converted into a target stator quadrature axis current command, which is then sent to the bottom current loop for closed-loop vector control drive.

[0108] The underlying signal synthesis logic module in the microcontroller performs physical energy aggregation and actuator signal distribution modulation at the controller instruction level.

[0109] S410. The base torque and the compensation torque command are added together to convert the result into the target stator quadrature axis current command, and the three-phase inverter bridge arm is controlled to achieve the designed force field space dynamic response.

[0110] The microcontroller directly loads two key torque maintenance variables, independently calculated and preserved in previous steps, from the shared address space of specific global variables in memory. One of them is the base torque, which characterizes the potential energy consumed to maintain the current mechanical structure against gravity during a fall. Secondly, it carries the advanced planning intent of the model and the compensation torque command that reflects the subjective adaptation to the feel of the action. The high-speed arithmetic logic unit inside the microcontroller executes non-blocking double-precision floating-point scalar addition instructions to generate the total thrust synthesis instructions required by the system drive terminals in a highly deterministic manner. In terms of mathematical expression, this total thrust synthesis command not only statically encompasses the static equilibrium solution required to maintain hovering in space, but also perfectly incorporates the controlled virtual inertia, elasticity, and viscous damping combined nonlinear dynamic response characteristics linearly excited by the user's subtle pushing and pulling intentions.

[0111] In order to project this abstract mechanical torque programmable onto the underlying electromagnetic actuator without loss, the microcontroller retrieves a preset motor torque constant calibration value uniquely bound to the matched motor hardware from a fixed read-only memory sector. The divider unit of the microcontroller performs a hardware-level floating-point division operation between the numerator of the generated total thrust synthesis command and the denominator of the motor torque constant, thereby accurately converting it into the target stator quadrature-axis current command setpoint in the stator magnetic field rotating coordinate system. .

[0112] For example, the gravity hovering base torque temporarily stored in the previous cycle was... (dimensions are) The optimal compensation torque, provided by the prediction engine feedback, reflects the user's intention to accelerate the door opening. (dimensions are) The combined instruction formed after the two are converged is: (dimensions are) The torque constant corresponding to the constant air gap flux constraint of the motor. Under the influence of the division operation, the target stator quadrature-axis current command is finally determined as follows: (dimensions are) ).

[0113] The target stator quadrature axis current command The reference level is immediately injected and updated to the underlying field-programmable gate array or independent servo control core, operating at a frequency of 10,000 Hz or even higher, in a hard real-time high-speed proportional-integral (PI) closed-loop current controller regulator network.

[0114] The proportional-integral controller calculates the error between the set current and the actual feedback quadrature-axis current after hardware acquisition and filtering, and generates a reference voltage vector amplitude and rotation phase angle including control compensation through integral accumulation. The built-in space vector pulse width modulation (SVPWM) generator algorithm, based on the spatial sector position of the voltage vector and according to a determined geometric projection timing allocation algorithm (such as specific on / off time calculation rules including zero vector and effective voltage working vector), maps the continuous reference voltage to a value derived from the input voltage. , , A high- and low-level digital pulse switching sequence signal with strictly defined time width. The modulated digital sequence is applied and amplified in parallel to drive the high- and low-side gate driver array of the power circuit network of a three-phase six-transistor full-bridge insulated-gate bipolar transistor or metal-oxide-semiconductor field-effect transistor inverter.

[0115] This nanosecond-level precise periodic switching action forces the DC bus voltage to reconstruct the required AC three-phase current within the motor winding stator, thereby accurately establishing the required alternating rotating air gap magnetic field. The rigid closed-loop action of this ultra-high frequency actuator enables the external rotor shaft end to precisely output mechanical thrust after layers of dimensionality reduction, decoupling, and optimization through complex multi-core calculations at the top level.

[0116] like Figure 3 As shown in the simulation diagram of the intentional torque observation in a), when the system is under the time-varying environmental base torque (the low-frequency torque caused by the slope disturbance shown by the dashed line in the figure), and is subjected to the real external interactive torque (i.e., the instantaneous push-pull operation of the user) during a specific time period, the generalized momentum observer constructed in this application can suppress high-frequency sensing noise and output a smooth external interactive torque observation estimate (as shown by the solid line in the figure) with extremely small first-order delay and extremely high steady-state accuracy.

[0117] like Figure 3 As shown in the simulation diagram of desired trajectory tracking in b), under the drive of the pure intention torque extracted above, the rotating virtual admittance model generates a continuous and abrupt desired angle trajectory (as shown by the dashed line in the figure). Moreover, under the premise of strictly adhering to the motor extreme output constraint, the explicit model predictive controller controls the actual angle trajectory (as shown by the solid line in the figure) to achieve near-deviation-free and overshoot-free dynamic tracking of the desired trajectory.

[0118] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described virtual force field balancing strut method based on intent recognition.

[0119] The computer-readable storage medium may be a non-volatile memory such as a read-only memory, random access memory, flash memory, hard disk storage, solid-state storage device, or optical disk.

[0120] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.

[0121] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for balancing a virtual force field strut based on intent recognition, characterized in that, The method includes: Acquire multi-source physical state signals of the system, which include the pitch angle, pitch angular velocity and cross-axis component of the motor stator current relative to the absolute reference plane of the strut; Obtain the system's solidified physical parameters, and based on the pitch angle, the pitch angular velocity, and the system's solidified physical parameters, construct and output the base moment characterizing gravity and inherent drag under the current spatial pose; The actual electromagnetic torque of the motor is obtained based on the quadrature axis component of the motor stator current, and the estimated value of the external interactive torque after orthogonally stripping environmental interference is extracted by using a preset generalized momentum observer based on the residual integral of the base torque and the actual electromagnetic torque of the motor. The estimated external interaction torque is input into a preset rotational virtual admittance model to generate a desired angle trajectory. The explicit model prediction controller locates and outputs a compensation torque command in a pre-divided polyhedral state space based on the difference between the desired angle trajectory and the current pitch angle.

2. The method according to claim 1, characterized in that, The process of acquiring the system's fixed physical parameters, and constructing and outputting the base moment characterizing gravity and inherent drag under the current spatial pose based on the pitch angle, the pitch angular velocity, and the system's fixed physical parameters, includes: Extract the pitch angle and pitch angular velocity from multiple consecutive frames, and perform an arithmetic average operation on the pitch angle and pitch angular velocity from the multiple consecutive frames to generate smooth state data aligned to the clock domain. Read the total physical mass, physical distance between the center of mass and the Coulomb friction torque constant from the system parameter dictionary; The gravity compensation component is obtained by multiplying the total physical mass of the system, the physical gravitational acceleration constant, the physical distance of the center of mass, and the sine value of the pitch angle. Extract the polarity sign of the pitch angular velocity, and multiply the polarity sign with the Coulomb friction torque constant to calculate the dynamic dry friction resistance component; The base torque is obtained by summing the gravity compensation component and the dynamic dry friction resistance component.

3. The method according to claim 1, characterized in that, The step of extracting the estimated external interactive torque after orthogonally stripping environmental interference, based on the residual integral of the base torque and the actual electromagnetic torque of the motor using a preset generalized momentum observer, includes: The generalized momentum of the current system is determined based on the product of the system's moment of inertia and the pitch angular velocity; Calculate the rate of change of the generalized momentum of the current system relative to the physical momentum of the previous discrete time step; The actual electromagnetic torque and the estimated external interaction torque of the previous cycle are summed, and the base torque of the previous cycle is subtracted to obtain the intermediate torque variable. The difference between the rate of change of physical momentum and the intermediate torque variable is calculated to obtain a scalar residual representing the degree of absolute imbalance. Multiply the scalar residual by the observer gain factor and the integration time step to obtain the intended update amount; The intention update value is accumulated and integrated with the external interaction torque estimate of the previous cycle to obtain the external interaction torque estimate of the current control cycle.

4. The method according to claim 3, characterized in that, The method further includes a temperature drift closed-loop compensation strategy, which includes: Receive the raw voltage value of the chip temperature sensor and the baseline current observation vector when the motor is stationary; When the system is determined to be in a stationary, locked state with zero speed and no external intervention, the original voltage value of the chip temperature sensor is converted into the current temperature value. Calculate the temperature difference between the current temperature value and the reference temperature value, and multiply the temperature difference by a preset attenuation coefficient to obtain the attenuation compensation ratio; The initial motor torque constant is multiplied by the attenuation compensation ratio to obtain the updated model parameter value, and the updated model parameter value is used to overwrite the currently used motor torque constant.

5. The method according to claim 1, characterized in that, The method further includes an intent-triggered threshold verification mechanism, which includes: Within a preset noise floor sampling period, multiple noise floor sample sets are recorded under conditions of no external interference. Statistical calculations were performed on the aforementioned low-noise sample set to determine the mean and standard deviation of the Gaussian distribution; The intention trigger threshold is calculated by adding the mean of the Gaussian distribution to the standard deviation by a preset multiple. Compare the absolute value of the extracted external interaction torque estimate with the intent trigger threshold; When the absolute value is greater than the intent trigger threshold, an enable flag is output to activate the explicit model prediction controller.

6. The method according to claim 1, characterized in that, The step of inputting the estimated external interaction torque value into a preset rotational virtual admittance model to generate the desired angle trajectory includes: Configure the virtual moment of inertia, virtual rotational damping, and virtual rotational stiffness of the rotating virtual admittance model; Subtract the damping torque generated by the virtual rotational damping and the expected angular velocity of the previous cycle from the estimated value of the external interaction torque, and subtract the elastic torque generated by the virtual rotational stiffness and the expected angle of the previous cycle to obtain the effective driving torque; The desired angular acceleration for the current cycle is calculated by dividing the effective driving torque by the virtual moment of inertia. The desired angular acceleration is integrated first-order to obtain the desired angular velocity of the current period, and the desired angular velocity is integrated second-order to obtain the desired angular trajectory.

7. The method according to claim 1, characterized in that, The explicit model predictive controller, based on the difference between the desired angle trajectory and the current pitch angle, locates and outputs a compensation torque command in a pre-divided polyhedral state space, including: Perform difference calculation to obtain the tracking error between the expected angle trajectory and the actual pitch angle, and calculate the tracking error change rate to construct a one-dimensional real-time state evaluation vector; Starting from the root node of the explicit control law addressing tree in memory, extract the boundary hyperplane inequality normal vector of the current polyhedral region; Perform a dot product operation between the one-dimensional real-time state evaluation vector and the boundary hyperplane inequality normal vector, and terminate the addressing process when all boundary constraints of the polyhedral region are satisfied. Extract the solidification feedback coefficient matrix and bias constant corresponding to the target polyhedral region; The solidified feedback coefficient matrix and the one-dimensional real-time state evaluation vector are multiplied and added together, and the bias constant is superimposed to output the compensation torque command that satisfies the electrical and physical boundary constraints of the motor.

8. The method according to claim 1, characterized in that, After locating and outputting the compensation torque command, the method further includes: The base torque and the compensation torque command are added together to generate the total thrust synthesis command required by the system. Divide the total thrust synthesis command by the motor torque constant to convert it into the target stator quadrature-axis current command; The target stator quadrature axis current command is input into the closed-loop current controller, which outputs a space vector pulse width modulation digital sequence to control the air gap of the three-phase inverter bridge arm drive motor to establish a rotating magnetic field.

9. A virtual force field balancing strut system based on intent recognition, characterized in that, The system includes: The multi-source sensing fusion module is configured to acquire multi-source physical state signals of the system, which include the pitch angle, pitch angular velocity and cross-axis component of the motor stator current relative to the absolute reference plane. The base environment calculation module is configured to acquire the system's solidified physical parameters, and based on the pitch angle, the pitch angular velocity, and the system's solidified physical parameters, construct and output the base torque characterizing gravity and inherent drag under the current spatial pose. The orthogonal intention extraction module is configured to obtain the actual electromagnetic torque of the motor based on the quadrature axis component of the motor stator current, and extract the estimated value of the external interactive torque after orthogonally stripping environmental interference based on the residual integral of the base torque and the actual electromagnetic torque of the motor through a preset generalized momentum observer. The predictive virtual impedance control engine is configured to input the estimated external interaction torque into a preset rotational virtual admittance model to generate a desired angle trajectory, and through an explicit model predictive controller, locate and output a compensation torque command in a pre-divided polyhedral state space based on the difference between the desired angle trajectory and the current pitch angle.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

Images