Wall-climbing robot four-wheel independent driving cooperative control method

Abstract

The application discloses a wall-climbing robot four-wheel independent driving cooperative control method and relates to the technical field of robots. The steps comprise the following steps: processing a physical state signal synchronously sampled to construct a same-frequency timestamp discrete data sequence; based on a body pose vector and a deformation boundary constraint model, the dynamic effective scale of a driving node is solved to reconstruct a dynamic Jacobian pseudo-inverse matrix and output a first control instruction; when the current discrete degree evaluation value is greater than a first threshold value, a zero space projection operator is constructed by using the dynamic Jacobian pseudo-inverse matrix to extract a zero space current component, and a second control instruction is generated to execute feedforward cancellation; when the slip acceleration extreme value is greater than a second threshold value, a first interrupt instruction is generated to pause the control flow and send a duty ratio pulse to a normal adhesion control end. The application eliminates wheel speed interference and motor heating and prevents slipping and falling.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to a method for coordinated control of four-wheel independent drive of a wall-climbing robot. Background Technology

[0002] In existing technologies, when wall-climbing robots traverse surfaces with varying curvature (such as storage tank surfaces or wind turbine blades), the uneven force distribution on each drive wheel leads to varying degrees of physical compression deformation in the flexible rubber-coated wheels. Current control methods typically calculate kinematic speeds based on a fixed wheel geometry. Because the actual physical compression of the wheels is ignored, the target speed issued by the system does not match the actual rolling state of the wheels. This mismatch between speed commands and the physical entity causes the four drive wheels to drag each other on the wall surface. This mechanical interference generates opposing currents within the motors, causing them to generate excessive heat and potentially triggering hardware crashes. Furthermore, when the wheels momentarily slip in the transition zone of varying curvature, existing systems rely on the drive motors to decelerate or stop. However, the mechanical closed-loop deceleration response cycle of the motors is relatively long, making it difficult to establish sufficient grip before the robot detaches from the wall, easily leading to the robot falling. Summary of the Invention

[0003] This invention provides a four-wheel independent drive cooperative control method for a wall-climbing robot to solve the technical problems in the prior art, such as target speed mismatch caused by ignoring the actual physical deformation of the wheels, resulting in mutual dragging of the drive wheels and motor overheating, and robot falling due to motor mechanical response lag during instantaneous slippage.

[0004] In view of the above problems, the present invention provides a four-wheel independent drive cooperative control method for a wall-climbing robot, applied to the wall-climbing robot comprising multiple drive nodes and a normal attachment control end, comprising the following steps: Step S1: Acquire the synchronously sampled physical state signal, which includes the body pose vector, the phase current discrete matrix of each driving node, and discrete displacement pulses; process the physical state signal to construct a discrete data sequence with the same frequency timestamp, which includes: the body pose vector, the phase current discrete matrix, and the slip acceleration extreme value sequence output after differentiating the discrete displacement pulses; Step S2: Receive the body pose vector, calculate the dynamic effective scale of each driving node based on the preset deformation boundary constraint model; reconstruct the dynamic Jacobian pseudo-inverse matrix in real time using the dynamic effective scale, and output the first control command; Step S3: Receive the phase current discrete matrix in real time and calculate the current discreteness evaluation value; when the current discreteness evaluation value is greater than a preset first threshold, extract the dynamic Jacobian pseudo-inverse matrix described in step S2 to construct a null space projection operator; use the null space projection operator to perform orthogonal decomposition on the phase current discrete matrix, extract the null space current components, and generate a second control command for the null space current components. Step S4: Monitor the slip acceleration extreme value sequence mentioned in step S1; when the extreme value in the slip acceleration extreme value sequence is detected to be greater than the preset second threshold, generate a first interrupt command, suspend the execution of steps S2 and S3, and send a duty cycle pulse to the normal attachment control terminal.

[0005] The technical solution provided in this application has at least the following technical effects: By acquiring the eccentricity of the body pose vector solution algorithm and inputting the normal eccentricity into the deformation boundary constraint model to output a dynamic effective scale, the dynamic Jacobian pseudo-inverse matrix is ​​reconstructed. This allows the system to dynamically correct the radius parameters involved in the solution based on the actual wheel compression. The built-in one-way truncation operator forces the eccentricity input to 0 when the wheel is suspended, avoiding the physical fallacy of inversely calculating wheel length using algebraic operations. This step outputs a rotational speed command consistent with the actual physical deformation, eliminating speed interference between the drive wheels.

[0006] By calculating the current dispersion evaluation value of the phase current discrete matrix and extracting the null space current component when it exceeds a first threshold, the generated second control command is used as a feedforward compensation quantity to perform arithmetic subtraction in the underlying field-oriented control loop. This step strips away parasitic currents that do not generate macroscopic displacement of the robot but only counteract each other between motors at the control layer. Without changing the robot's normal driving trajectory, it directly cancels out the excess current generated by mutual dragging, reducing the heat generation of the motors.

[0007] By inputting discrete displacement pulses into a tracking differential observer to extract the extreme value sequence of slip acceleration, a first interrupt command is generated when the extreme value exceeds a second threshold, sending a duty cycle pulse to the normal adhesion control terminal. Upon detecting instantaneous slippage, the system skips the long-cycle mechanical deceleration closed loop of the motor and instantaneously increases the normal adhesion force by utilizing the extremely short response of the fluid or electromagnetic adhesion field, completing steady-state fallback before the slow mechanical drop occurs. Attached Figure Description

[0008] Figure 1 A flowchart of a four-wheel independent drive cooperative control method for a wall-climbing robot provided in an embodiment of the present invention; Figure 2 This is a hardware structure diagram of the present invention. Detailed Implementation

[0009] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0010] Example: Please refer to Figure 1 and Figure 2 The four-wheel independent drive cooperative control method for the wall-climbing robot operates on the underlying physical architecture. This architecture comprises four drive nodes arranged in a matrix and a normal attachment control terminal independent of the drive nodes. Each drive node includes a brushless DC motor, a reducer connected to the brushless DC motor, and an extended flexible rubber-coated wheel. The normal attachment control terminal receives electrical signal commands via an independent control bus. Its specific physical form manifests as a ducted fan hydrodynamic system, a pre-vacuum energy storage pneumatic circuit, or a high-voltage overexcited electromagnetic array. Along with the energization of the underlying physical architecture, synchronously sampled physical state signals are acquired by a heterogeneous sensing matrix. This matrix includes an inertial measurement unit for measuring gravity deflection, a current transformer for collecting underlying electrical characteristics, and a photoelectric encoder for monitoring physical rotational displacement. When the physical state signal is captured by the heterogeneous sensing matrix, a underlying hardware timer triggers data synchronization timestamp marking and alignment logic. The data synchronization timestamp marking and alignment logic binds the body pose vector output by the inertial measurement unit, the discrete phase current matrix output by the current transformer, and the discrete displacement pulse output by the photoelectric encoder to the same frequency timestamp, and outputs a discrete data sequence with the same frequency timestamp.

[0011] After the discrete data sequence with the same frequency and timestamp is generated, the multi-level logic arbitration mechanism executes the control flow based on the discrete data sequence with the same frequency and timestamp. In the basic tracking closed-loop state, the body pose vector is received by the deformation boundary constraint model. The deformation boundary constraint model calculates the dynamic effective scale based on the body pose vector, and the dynamic effective scale reconstructs the dynamic Jacobian pseudo-inverse matrix. The dynamic Jacobian pseudo-inverse matrix generates the first control command and sends it to the drive node. At the same time as the first control command is output, the null space projection operator is superimposed on the basic tracking closed-loop state. The null space projection operator calls the dynamic Jacobian pseudo-inverse matrix to perform orthogonal decomposition on the phase current discrete matrix and extracts the null space current component. When a physical abrupt change occurs in the slip acceleration extreme value sequence in the discrete data sequence and the extreme value in the slip acceleration extreme value sequence is greater than the preset second threshold, the multi-level logic arbitration mechanism generates the first interrupt command. The first interrupt command monopolizes the control bus, suspends the dynamic effective scale calculation logic and the null space current component extraction logic, and sends a duty cycle pulse to the normal attachment control terminal.

[0012] As the physical input source triggering the first interrupt instruction, the slip acceleration extreme value sequence is generated by the data extraction logic. Under the physical conditions of gear meshing and flexible rubber-coated wheel bouncing, the discrete displacement pulses output by the photoelectric encoder contain high-frequency phase noise. The high-frequency phase noise manifests as disordered level jumps on the time axis. The acquired discrete displacement pulses are input into a preset tracking differential observer. The tracking differential observer has a built-in optimal control synthesis function and constructs a state-space equation. After the discrete displacement pulses are input into the optimal control synthesis function, the optimal control synthesis function constructs a second-order state derivation formula in the discrete-time domain. The state derivation formula is configured as follows: the displacement estimate of the system in the next clock cycle is equal to the algebraic sum of the product of the current displacement estimate and the current velocity estimate with the same integration step size; the velocity estimate of the system in the next clock cycle is equal to the algebraic sum of the current velocity estimate and the acceleration control quantity. The acceleration control quantity is calculated and output by the optimal control synthesis function using the physical difference between the current displacement estimate and the discrete displacement pulse input, the current velocity estimate, and the velocity tracking factor as independent variables, through a mathematical saturation function. The underlying state derivation formula is periodically invoked via clock interrupts. Under the physical condition of avoiding the phase lag of the low-pass filter, the second-order state derivative sequence is smoothly extracted and output. The second-order state derivative sequence is defined as the slip acceleration extremum sequence and flows to a multi-level logic arbitration mechanism to perform threshold comparison.

[0013] After the data synchronization timestamp marking and alignment logic is completed, the body pose vector is extracted by the multi-level logic arbitration mechanism and imported into the deformation boundary constraint model. The body pose vector contains the system's current three-dimensional spatial tilt angle and gravitational acceleration characteristics. The deformation boundary constraint model retrieves the physical geometric wheelbase preset in the storage medium. The physical geometric wheelbase is spanned by the spatial coordinates of four drive nodes. Using the three-dimensional spatial tilt angle and gravitational acceleration characteristics of the body pose vector as input, the deformation boundary constraint model constructs a set of rigid body static equilibrium equations, including Z-axis normal force balance, X-axis pitch moment balance, and Y-axis roll moment balance. The rigid body static equilibrium equations extract the scalar product of the system's total mass and gravitational acceleration, and project the scalar product along the three-dimensional spatial tilt angle using trigonometric functions to obtain the total Z-axis normal load and the eccentric moments located on the X and Y axes. The deformation boundary constraint model retrieves the physical distance vector from the center of each drive node to the geometric center of the body. The eccentric moment is divided by this physical distance vector, transforming it into a dynamic load variable added to the foundation load of each drive node. By solving the rigid body static equilibrium equations, the deviation of the normal load component independently borne by each drive node is decoupled and output. After dimensionless normalization, the normal load component deviation is output as the corresponding normal eccentricity for each drive node.

[0014] After the normal eccentricity is generated, it sequentially enters the algebraic operation loop of the dynamic effective scale. The extraction of the dynamic effective scale is based on the analytical model. in, The dynamic effective scale is the i-th driver node among the aforementioned driver nodes; This refers to the static reference scale; The maximum compressibility limit threshold characterizing the physical extreme value; The normal eccentricity of the i-th driving node is calculated based on the body pose vector; Let the hyperbolic tangent function characterize the first mapping operator; It is an exponential function with the natural constant as its base. Used to characterize the second mapping operator; This refers to the one-way truncation operator; This is the preset first weighting constant; This is the preset second weighting constant.

[0015] In the analytical model's computational sequence, the normal eccentricity triggers a one-way truncation operator. The one-way truncation operator embeds linear rectification logic. When the normal eccentricity is less than or equal to 0, the one-way truncation operator sets the normal eccentricity to 0 and truncates the output stream, making the subtraction term of the algebraic formula zero, and the dynamic effective scale equal to the preset static baseline scale. When the normal eccentricity is greater than 0, normal eccentricity data packets greater than 0 are synchronously input to the first and second mapping operators. The first mapping operator calls the hyperbolic tangent function kernel. The hyperbolic tangent function performs numerical compression on the input normal eccentricity, outputting a compressed deformation that approximates a preset physical extreme value. The preset physical extreme value corresponds to the maximum compressibility limit threshold, which characterizes the limiting geometric parameters of the physical interface of the flexible rubber-coated wheel in contact with the internal rigid metal hub. The second mapping operator calls a kernel derived from an exponential function. When the normal eccentricity falls within the range approaching 0, the second mapping operator outputs a very small suppression multiplier, reducing the gain steepness of the compressed deformation and eliminating high-frequency oscillations in the control loop caused by microscopic wall surface undulation noise. The value output by the first mapping operator is multiplied by the value output by the second mapping operator and then superimposed with the static reference scale to output the dynamic effective scale of each driving node.

[0016] After the dynamic effective scale is output from each driving node, it is injected into the basic motion space matrix. The basic motion space matrix reconstructs the forward kinematic transformation relationship based on the dynamic effective scale, generating a dynamic Jacobian matrix. A multi-level logical arbitration mechanism extracts the dynamic Jacobian matrix and performs pseudo-inverse calculation. To prevent motion control from triggering matrix singularities under the topological configuration, thus causing computational overflow and crashes, the pseudo-inverse calculation introduces damped least squares for regularization, outputting a dynamic Jacobian pseudo-inverse matrix. The dynamic Jacobian pseudo-inverse matrix is ​​then multiplied by the underlying motion trajectory planning vector to generate the first control command. This first control command immediately flows to each driving node to trigger physical rotation execution.

[0017] Within the timeline of the continuous generation and flow of the first control command, the multi-level logic arbitration mechanism synchronously receives the phase current discrete matrix returned by each drive node. The phase current discrete matrix is ​​input into the discreteness evaluation mathematical model. The discreteness evaluation mathematical model extracts the algebraic variance of each spatial element in the phase current discrete matrix. The algebraic variance characterizes the degree of uneven force distribution among the four drive nodes due to differences in physical terrain. The discreteness evaluation mathematical model has a built-in monotonically increasing empirical mapping function. The algebraic variance is input as an independent variable to the empirical mapping function, which outputs the corresponding stiffness deterioration coefficient. The stiffness deterioration coefficient is numerically equivalent to the deterioration trend of the ratio of the maximum to the minimum eigenvalue of the drive stiffness matrix, and is thus directly assigned the physical definition of the condition number of the drive stiffness matrix by the system. The condition number of the drive stiffness matrix is ​​given a numerical definition of the current discreteness evaluation value. The current discreteness evaluation value and a preset first threshold are entered into a Boolean comparator. The first threshold characterizes the critical point at which the internal physical heat generation power of the motor crosses the safety envelope. When the current discreteness evaluation value is greater than the first threshold, the multi-level logic arbitration mechanism triggers the generation link of the second control command.

[0018] With the activation of the second control command generation link, the multi-level logic arbitration mechanism extracts the existing dynamic Jacobian pseudo-inverse matrix and the corresponding dynamic Jacobian matrix from memory. Using the dynamic Jacobian pseudo-inverse matrix and the dynamic Jacobian matrix as input basis, the multi-level logic arbitration mechanism performs algebraic matrix multiplication and performs a difference operation with the identity matrix, outputting a transformation matrix. The transformation matrix establishes a mapping plane orthogonal to the basic motion space in the 3D vector space, forming a null space projection operator. The phase current discrete matrix is ​​imported into the null space projection operator to perform spatial linear mapping. The spatial linear mapping decomposes the phase current discrete matrix, filters out the current components that induce macroscopic physical displacement of the robot, and extracts the internal topological current characteristics independent of the basic motion space. The internal topological current characteristics are defined as null space current components. After scaling, the null space current components generate a second control command for the null space current components. The second control command is sent to the field-oriented control bus inside the drive node. The magnetic field orientation control bus inside the drive node performs arithmetic subtraction by using the second control command as a feedforward compensation in the quadrature axis current feedback loop to cancel out the mutual internal current loss between each drive node.

[0019] As the first and second control commands are continuously issued, a multi-level logic arbitration mechanism synchronously extracts and monitors the slip acceleration extreme value sequence within a discrete data sequence with the same frequency timestamp. When an extreme value in the slip acceleration extreme value sequence is detected to be greater than a preset second threshold, the multi-level logic arbitration mechanism generates a first interrupt command. The generation logic of the first interrupt command is based on the physical boundary of time constant mismatch. In the chassis physical kinematics test mechanism of the wall-climbing robot, the mechanical response period of each drive node executing deceleration or stopping through a mechanical closed loop is much larger than the physical response time constant of the normal attachment control end establishing a fluid attachment field or electromagnetic attachment field. The first interrupt command responds to the time constant scissors difference between the mechanical response period and the physical response time constant, and takes over the control of the system bus.

[0020] After the first interrupt instruction is generated, it activates the potential energy release logic through the highest priority external interrupt pin within the microcontroller unit. At the hardware register level of the microcontroller unit, the potential energy release logic bypasses the pulse width modulation waveform generator, suspending the closed-loop electrical flow between the first and second control instructions. During the physical clock cycle that suspends the normal flow sequence, the potential energy release logic generates a first trigger signal containing the maximum duty cycle parameter and sends this first trigger signal as a duty cycle pulse to the gate drive circuit of the normal attachment control terminal. The duty cycle pulse manifests as a step shock wave level with the maximum duty cycle characteristic. The duty cycle pulse forces the normal attachment control terminal to complete a step change in the physical attachment field state within a time constant threshold less than a preset threshold. By activating the potential energy release logic, the multi-level logic arbitration mechanism restores the normal adhesion force between the wall-climbing robot chassis and the physical wall before the mechanical deceleration hysteresis period arrives. With the restoration of the normal adhesion force, the multi-level logic arbitration mechanism continuously monitors the slip acceleration extreme value sequence through a tracking differential observer. When the extreme values ​​in the slip acceleration extreme value sequence fall back from above the preset second threshold to below or equal to the preset second threshold, the multi-level logic arbitration mechanism triggers the state recovery logic. The state recovery logic sends a steady-state hold command to the normal attachment control terminal to maintain the current attachment field, cancels the bypass control of the potential energy release logic on the pulse width modulation waveform generator, removes the closed-loop electrical flow restriction between the first control command and the second control command, and restores the normal execution of the basic tracking closed loop and the zero-space orthogonal decomposition.

[0021] In the closed-loop algebraic computation of the deformation boundary constraint model, the characteristic parameters of the analytical model conform to a 3D spatial convergence base, preventing hardware division by zero crashes or data overflows caused by the algebraic model output. As a specific parameter characterizing physical extrema, the factory-calibrated value of the maximum compressibility limit threshold is constrained within a specific physical geometric range. The upper limit of this specific physical geometric range is defined by subtracting the difference between the preset static benchmark scale and the physical radius of the rigid metal hub inside each drive node. The numerical constraint of the maximum compressibility limit threshold ensures that the underlying algebraic derivation of the analytical model conforms to the 3D Euclidean geometric boundary, cutting off the logical possibility of mathematical formula output penetrating the rigid boundary of the physical entity.

[0022] The analytical model's built-in first and second weighting constants are based on the physical mapping principles of materials engineering. The first weighting constant determines the slope gain of the hyperbolic tangent function. During the physical factory calibration stage, the first weighting constant is mapped to the Shore hardness value of the flexible rubber-coated wheel manufacturing formula. The higher the Shore hardness value of the flexible rubber-coated wheel, the smaller the calibration value of the first weighting constant, which algebraically reduces the sensitivity of the first mapping operator in deformation feedback. The second weighting constant determines the convergence time constant of the exponential decay function. During the physical factory calibration stage, the second weighting constant is mapped to the spatial filtering characteristics of the external tread depth of the flexible rubber-coated wheel. The deeper the tire tread depth of the flexible rubber-coated wheel, the smaller the calibration value of the second weighting constant, which algebraically increases the passivation absorption depth of the second mapping operator for microscopic undulation physical noise.

[0023] With the solidification of the physical calibration domain and dimensional analysis parameters, the algebraic convergence of the deformation boundary constraint model under extreme physical environments exhibits rigid boundary constraints. When a wall-climbing robot operates on a complex 3D physical surface and encounters a unilateral physical suspension condition, the driving node on the suspended side loses its normal physical contact force. The multibody dynamics algebraic equations, based on the physical suspension state, output a normal eccentricity approaching negative infinity. The normal eccentricity data packet approaching negative infinity triggers the linear rectification gating logic within the unidirectional truncation operator. The unidirectional truncation operator performs a Boolean negative value judgment on the normal eccentricity approaching negative infinity, outputting 0. 0 is synchronously injected into the multiplication operation link within the analytical model, erasing the algebraic weights of the first and second mapping operators, forcing the compressed deformation variable to be truncated to 0. The final algebraic summation result of the dynamic effective scale is locked as the static baseline scale. The zero-output mechanism of the one-way truncation operator blocks the computational fallacy of negative eccentricity input stretching the physical radius from the mathematical level, eliminating the engineering risk of spatial mapping singularity explosion caused by injecting erroneous radius data into the dynamic Jacobian pseudo-inverse matrix.

[0024] When the physical load on the wall-climbing robot exceeds the limit operation threshold, or when it encounters extreme gravity overturning moment distortion conditions, the load-bearing side drive node bears saturated physical normal pressure. The multibody dynamics algebraic equations, based on this saturated physical normal pressure, output a normal eccentricity approaching positive infinity. This approaching normal eccentricity sequentially passes through the unidirectional truncation operator and is fully injected into the first mapping operator. The hyperbolic tangent function inside the first mapping operator performs a limit operation on the approaching normal eccentricity. Based on the inherent mathematical asymptotic characteristics of the hyperbolic tangent function, the limit operation result converges to a mathematical upper limit of 1. The upper limit of 1 is multiplied by the maximum compressibility limit threshold using a scalar. The algebraic operation sequence of the dynamic effective scale outputs a constant lower bound, equal to the static baseline scale minus the maximum compressibility limit threshold. The limit convergence mechanism of the hyperbolic tangent function encapsulates the dynamic effective scale within a finite upper and lower bound scalar space, establishing the global boundedness of matrix operations under extreme physical distortion conditions.

[0025] In microcontroller physical operating environments where floating-point units are lacking or computing resources are limited, the deformation boundary constraint model performs closed-loop operations based on a degraded operation branch. The degraded operation branch strips away hyperbolic tangent floating-point operation instructions and reconstructs the physical boundary constraints using a lookup table approach. The saturation convergence mathematical properties of the first mapping operator and the origin decay mathematical properties of the second mapping operator within the deformation boundary constraint model are pre-discreteized to generate discrete numerical matrices corresponding to specific normal eccentricity nodes. These discrete numerical matrices are physically burned into read-only memory (ROM), constructing a 1D or 2D constant data table. Upon receiving the real-time calculated normal eccentricity, the microcontroller unit uses the normal eccentricity as the address key to retrieve adjacent discrete constant nodes from ROM. The microcontroller unit performs linear interpolation algebra operations on the retrieved adjacent discrete constant nodes, outputting approximate compressed deformation variables. The extraction process of the approximate compressed deformation variables skips the complex calculus floating-point operation cycle while simultaneously reproducing the mathematical response properties defined by the first and second mapping operators.

[0026] As a parallel alternative to the lookup table logic, the Taylor expansion polynomial substitution logic is pre-placed in the computation queue for small eccentricity operation conditions. When the algebraic magnitude of the normal eccentricity is within a preset small fluctuation range, the deformation boundary constraint model truncates the hyperbolic tangent function kernel, activating a 3rd or 5th order truncated polynomial algebraic operation sequence. The truncated polynomial algebraic operation sequence consists of a static baseline scale minus the product of the first constant term and the cube of the one-way truncation operator output value, and superimposed with the product of the second constant term and the 5th power of the one-way truncation operator output value. Based on the geometric curvature characteristics of odd-power functions, the truncated polynomial outputs a smooth mathematical hysteresis signal near the origin and exhibits amplified convergence characteristics as the input independent variable increases. The output result of the truncated polynomial algebraic operation sequence maps saturation convergence properties and origin decay properties. The Taylor expansion polynomial substitution logic, through a basic algebraic multiplication-addition instruction sequence, completes the extraction of the dynamic effective scale, constraining the algorithm result within the physical defense envelope defined by the deformation boundary constraint model.

[0027] In the matrix decomposition implementation path of the deformation boundary constraint model, the multi-level logic arbitration mechanism extracts the null-space current components by invoking the singular value decomposition algorithm. The multi-level logic arbitration mechanism performs singular value decomposition on the dynamically reconstructed dynamic Jacobian matrix, decoupling the dynamic Jacobian matrix into the product of a left singular matrix, a singular value matrix, and a right singular matrix. The multi-level logic arbitration mechanism traverses each element in the singular value matrix, filters out the 0 singular values, and extracts the right singular vector group corresponding to the 0 singular values ​​from the right singular matrix. The extracted right singular vector group undergoes matrix transpose and combination operations to construct the transformation matrix. The calculation process of constructing the transformation matrix through the singular value decomposition algorithm completes the separation of current characteristics independent of the basic motion space in algebraic space, achieving a physical mapping orthogonal to the basic motion space.

[0028] To solidify the operational platform of the collaborative control method and address the object suitability of the software control algorithm, the control flow logic is physically mapped to a specific collaborative controller hardware base. The collaborative controller hardware base is equipped with a heterogeneous bus interface layer, which includes a Controller Area Network (CAN) Flexible Data Rate Bus (FDRB) interface or an Industrial Ethernet Fieldbus (IFET) interface. The collaborative controller connects to four distributed drive nodes at the bottom layer via the CANRB or IFET interfaces, using either a serial or ring topology. Each distributed drive node integrates an analog-to-digital converter (ADC), which directly connects to the drive circuit to perform physical sampling of phase currents and transmits the data to the collaborative controller via the bus interface. The heterogeneous bus interface layer includes a high-speed serial peripheral interface bus. The collaborative controller connects to the chassis inertial measurement unit (IMU) via this bus and synchronously reads the body pose data from the IMU's internal physical registers using a clock signal. The computing power of the collaborative controller consists of a microprocessor core with a digital signal processing (DSP) instruction set or a field-programmable gate array (FPGA) coprocessor, used to perform discrete matrix multiplication and differential observation operations.

[0029] The virtual logic module inside the collaborative controller establishes the exclusivity of the multi-level logic arbitration mechanism through a hardware timer and interrupt vector table mapping mechanism. The basic tracking closed-loop control logic that outputs the first control instruction is attached to the basic timer interrupt service function, and the execution cycle of the basic timer interrupt service function is calibrated according to the control update rate of the chassis motion planning. The zero-space orthogonal decomposition logic that extracts the zero-space current component and outputs the second control instruction is attached to the current loop sampling interrupt service function, and the execution cycle of the current loop sampling interrupt service function is calibrated according to the motor electrical time constant and the response bandwidth of the field-oriented control. The interrupt priority of the current loop sampling interrupt service function is higher than that of the basic timer interrupt. The transient extreme value fallback logic that generates the first interrupt instruction and issues the duty cycle pulse is directly connected to the non-maskable interrupt pin of the collaborative controller through a hardware pin. The non-maskable interrupt pin has high-level hardware interrupt vector table privileges at the hardware circuit level. When the non-maskable interrupt pin is activated by an extreme value in the sliding acceleration extreme value sequence that is greater than the preset second threshold trigger level, the collaborative controller suspends the register clocks corresponding to the basic timer interrupt service function and the current loop sampling interrupt service function, and exclusively executes the duty cycle pulse sending action. The co-controller is internally configured with a hardware-level interrupt nesting monitoring register. When the trigger level of the non-maskable interrupt pin is deactivated and the extreme value in the slip acceleration extreme value sequence meets the fallback condition, the co-controller writes an interrupt clear flag to the hardware-level interrupt nesting monitoring register. The interrupt clear flag forcibly resets the execution suspension state of the non-maskable interrupt, and the hardware clock tree reallocates register clock cycles to the basic timer interrupt service function and the current loop sampling interrupt service function, realizing a seamless degradation and control handover between different physical interrupt priorities through a multi-level logical arbitration mechanism.

[0030] The binary machine code, which includes the construction of discrete data sequences, reconstruction of dynamic Jacobian pseudo-inverse matrices, null space orthogonal decomposition, and hardware interrupt state transition logic, is physically embedded in the non-volatile flash memory medium or embedded multimedia memory card medium within the co-controller.

[0031] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for coordinated control of four-wheel independent drive of a wall-climbing robot, characterized in that, The method applied to the wall-climbing robot, which includes multiple drive nodes and a normal attachment control end, includes the following steps: Step S1: Acquire the synchronously sampled physical state signal, which includes the body pose vector, the phase current discrete matrix of each driving node, and discrete displacement pulses; process the physical state signal to construct a discrete data sequence with the same frequency timestamp, which includes: the body pose vector, the phase current discrete matrix, and the slip acceleration extreme value sequence output after differentiating the discrete displacement pulses; Step S2: Receive the body pose vector, calculate the dynamic effective scale of each driving node based on the preset deformation boundary constraint model; reconstruct the dynamic Jacobian pseudo-inverse matrix in real time using the dynamic effective scale, and output the first control command; Step S3: Receive the phase current discrete matrix in real time and calculate the current discreteness evaluation value; when the current discreteness evaluation value is greater than a preset first threshold, extract the dynamic Jacobian pseudo-inverse matrix described in step S2 to construct a null space projection operator; use the null space projection operator to perform orthogonal decomposition on the phase current discrete matrix, extract the null space current components, and generate a second control command for the null space current components. Step S4: Monitor the slip acceleration extreme value sequence mentioned in step S1; when the extreme value in the slip acceleration extreme value sequence is detected to be greater than the preset second threshold, generate a first interrupt command, suspend the execution of steps S2 and S3, and send a duty cycle pulse to the normal attachment control terminal.

2. The four-wheel independent drive cooperative control method for a wall-climbing robot according to claim 1, characterized in that, In step S2, the calculation of the dynamic effective scale of each driving node specifically includes: Using the body pose vector, the normal eccentricity corresponding to each of the driving nodes is calculated; The normal eccentricity is input into the deformation boundary constraint model; wherein, the deformation boundary constraint model is configured as follows: When the normal eccentricity is greater than zero, a first mapping operator with saturation convergence property is used to output a compressed deformation that approximates a preset physical extreme value based on a preset first weight constant; and a second mapping operator with origin decay property is used to filter the characteristic response of the normal eccentricity within a preset interval based on a preset second weight constant. When the normal eccentricity is less than or equal to zero, the compressed deformation is truncated to zero using a one-way truncation operator, and the dynamic effective scale is made equal to the preset static reference scale.

3. The four-wheel independent drive cooperative control method for a wall-climbing robot according to claim 2, characterized in that, The dynamic effective scaling is calculated using the following analytical model: in, The dynamic effective scale is the i-th driver node among the aforementioned driver nodes; This refers to the static reference scale; The maximum compressibility limit threshold characterizing the physical extreme value; The normal eccentricity of the i-th driving node is calculated based on the body pose vector; Let the hyperbolic tangent function characterize the first mapping operator; It is an exponential function with the natural constant as its base. Used to characterize the second mapping operator; This refers to the one-way truncation operator; The first weighting constant is preset; This is the preset second weighting constant.

4. The four-wheel independent drive cooperative control method for a wall-climbing robot according to claim 1, characterized in that, In step S3, the orthogonal decomposition of the phase current discrete matrix using the null projection operator specifically includes: Using the dynamic Jacobian pseudo-inverse matrix and the dynamic Jacobian matrix corresponding to the dynamic Jacobian pseudo-inverse matrix as a basis, a transformation matrix orthogonal to the basic motion space is constructed as the null space projection operator. The phase current discrete matrix is ​​input into the transformation matrix for linear mapping to extract the current characteristics independent of the basic motion space, which are then used as the zero-space current components.

5. The four-wheel independent drive cooperative control method for a wall-climbing robot according to claim 1, characterized in that, In step S1, the sequence of extreme values ​​of slip acceleration output after differentiating the discrete displacement pulse specifically includes: The obtained discrete displacement pulses are input into a nonlinear tracking differentiator; The discrete displacement pulses are processed using the nonlinear tracking differentiator, and the second-order state derivative sequence is output as the extreme value sequence of the slip acceleration.

6. The four-wheel independent drive cooperative control method for a wall-climbing robot according to claim 1, characterized in that, In step S4, sending a duty cycle pulse to the normal attachment control terminal is specifically configured as follows: In response to the first interrupt instruction, a preset time constant threshold is obtained, wherein the time constant threshold is less than the mechanical response period of each of the drive nodes; A first trigger signal is generated as the duty cycle pulse and sent to the normal attachment control terminal; wherein the time interval from the generation of the first trigger signal to the response output of the normal attachment control terminal is less than the time constant threshold.

7. The four-wheel independent drive cooperative control method for a wall-climbing robot according to any one of claims 1 to 6, characterized in that, Steps S2, S3, and S4 are controlled by a multi-level logical arbitration mechanism, which includes: When the extreme value in the slip acceleration extreme value sequence is greater than the second threshold, step S4 is executed exclusively. When the extreme value in the slip acceleration extreme value sequence falls back from being greater than the second threshold to being less than or equal to the second threshold, the normal execution of steps S2 and S3 is resumed. When the extreme value in the slip acceleration extreme value sequence is less than or equal to the second threshold, and the current dispersion evaluation value is greater than the first threshold, steps S2 and S3 are executed simultaneously. When the extreme value in the slip acceleration extreme value sequence is less than or equal to the second threshold, and the current dispersion evaluation value is less than or equal to the first threshold, step S2 is executed separately.

Images