Construction robot pose adjustment method and system

Abstract

The application relates to the technical field of robot control, and discloses a building robot posture adjusting method and system, which comprises the following steps: acquiring the absolute distance from a chassis to a threshold and a real-time linear velocity, establishing a crossing time window; reading inertial measurement unit data, combining the time window to execute drift truncation and secondary integration to obtain Z-axis dynamic displacement, and calculating a real-time vertical air gap height of a sensor; based on the air gap height, reversely compensating an original magnetic field signal into a normalized magnetic field signal and total voltage; in the time window, decoupling environmental magnetic anomalies according to the comparison result of the total voltage and a metal interference threshold, switching the weight of the magnetic field and visual navigation, calculating the magnetic field and visual offset respectively, fusing the weight to output a control target offset to adjust the wheel differential. The application solves the single navigation failure problem caused by the chassis bumping height mutation and metal interference when crossing the threshold, and guarantees the posture stability and trajectory precision of the robot when crossing the complex terrain.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, specifically to a method and system for adjusting the posture of a construction robot. Background Technology

[0002] With the development of construction automation, the application of construction robots on construction sites is gradually increasing. In actual working environments, robots often need to autonomously enter and exit elevators or cross various ledges, which requires them to have navigation and attitude control capabilities in uneven terrain. Currently, construction robots mostly use a combination of magnetic field navigation and visual navigation for path tracking.

[0003] When the robot traverses raised terrain such as elevator door sills, its chassis vibrates and tilts. This change in posture causes a transient change in the vertical air gap height between the magnetic sensors mounted on the bottom of the chassis and the ground. Since magnetic field strength decreases non-linearly with spatial distance, the increase in physical height directly causes magnetic field signal attenuation and distortion, leading to errors in the lateral offset calculated by the controller. Furthermore, elevator door sills are typically made of metal, which can shield and interfere with the ambient magnetic field, causing localized magnetic field anomalies.

[0004] Existing robot navigation and control systems struggle to effectively distinguish between physical attenuation caused by sudden changes in chassis height and magnetic shielding caused by interference from metallic materials when facing the aforementioned conditions. This coupling phenomenon prevents the robot from accurately compensating for underlying signals when crossing ridges. Furthermore, existing multi-sensor fusion strategies often rely on fixed threshold judgments and lack spatial feedforward prediction mechanisms for terrain-crossing processes. When faced with environmental magnetic anomalies, they are prone to weight switching lags or control command jumps, ultimately causing the robot to deviate from its intended trajectory or lose control when traversing ridge areas. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for adjusting the posture of construction robots. This method solves the problem that when construction robots cross sills, the physical attenuation of magnetic signals caused by sudden changes in chassis height and the magnetic interference caused by the metal material of the sills coupled together, leading to the failure of single navigation and deviation of the robot's trajectory.

[0006] To address the above problems, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides a method for adjusting the posture of a construction robot, comprising the following steps:

[0008] The absolute physical distance from the chassis to the sill is obtained by a depth camera, and combined with the real-time longitudinal linear velocity obtained by the encoder, the estimated time for the front wheel to contact the sill is calculated and a time window is established.

[0009] The multidimensional attitude angle and Z-axis original acceleration of the inertial measurement unit are read. Drift truncation and double integration are performed according to the system time and the boundary relationship of the time window to obtain the dynamic displacement of the Z-axis. The real-time vertical air gap height of the Hall sensor arrays on both sides is calculated in combination with the physical dimensions of the chassis.

[0010] Based on the preset attenuation function and the real-time vertical air gap height, the original magnetic field signals on both sides are reverse-compensated into a first normalized magnetic field signal, a second normalized magnetic field signal, and a normalized total voltage.

[0011] When the system time is within the time window, the environmental magnetic anomaly is decoupled based on the comparison result of the normalized total voltage and the metal interference characteristic threshold, and the state switching between magnetic field navigation weight and visual navigation weight is performed.

[0012] The magnetic field offset is calculated based on the first normalized magnetic field signal and the second normalized magnetic field signal. The visual offset is calculated based on the visual image. The magnetic field navigation weight and the visual navigation weight are fused to output the control target offset in order to adjust the wheel differential.

[0013] Furthermore, during the establishment of the time-crossing window, the absolute difference between the linear velocities of the left and right drive wheels is monitored. When the absolute difference is less than a preset slip tolerance threshold, the arithmetic mean is taken as the real-time longitudinal linear velocity; when the absolute difference is greater than or equal to the slip tolerance threshold, the average velocity within the historical slip time window is used as the real-time longitudinal linear velocity. Subsequently, the absolute physical distance is divided by the real-time longitudinal linear velocity after amplitude limiting verification to obtain the estimated time, and the estimated time is used as a reference to extend the preset time error tolerance in both directions before and after the time axis to generate a closed-interval time-crossing window.

[0014] Furthermore, when performing drift truncation, if the system time is before the starting boundary of the time window, it is determined to be a flat driving state, and the drift truncation command is forcibly executed to lock the dynamic displacement of the Z-axis to zero; and the moving average filtering algorithm is used to perform mean calculation on the original Z-axis acceleration before the starting boundary, and the background noise of the Z-axis acceleration is extracted and updated.

[0015] Furthermore, when the system time enters the spanning time window, the last Z-axis acceleration background noise before entering the spanning time window is latched, and the effective acceleration is obtained by subtracting the last Z-axis acceleration background noise from the real-time raw Z-axis acceleration. Continuous trapezoidal discrete numerical integration is performed on the effective acceleration to obtain the Z-axis dynamic displacement. The horizontal physical span and longitudinal physical distance in the chassis physical dimensions are called, and spatial geometric mapping is performed in combination with the Z-axis dynamic displacement and the roll angle and pitch angle in the multi-dimensional attitude angles to calculate the real-time vertical air gap height, and sensing distance boundary limiting is applied.

[0016] Furthermore, in the signal compensation stage, the static reference bias voltage is subtracted from the acquired original magnetic field signals on both sides. Using the real-time vertical air gap height as an index, a preset one-dimensional lookup table is retrieved and a linear interpolation algorithm is called to calculate the corresponding dynamic attenuation coefficient. The original magnetic field signal after subtracting the bias voltage is divided by the dynamic attenuation coefficient limited by the lower boundary for inverse compensation, outputting the first normalized magnetic field signal and the second normalized magnetic field signal respectively, and the two are arithmetically summed to output the normalized total voltage. The above compensation principle is based on the fact that the magnetic field strength decreases nonlinearly with spatial distance. By restoring the true physical air gap height through spatial geometric mapping and using inverse mathematical compensation, the magnetic field signal base offset caused by chassis vibration can be eliminated.

[0017] Further, it is determined whether the normalized total voltage is less than or equal to the product of the flat ground reference total voltage and the metal interference characteristic threshold. If the normalized total voltage is less than or equal to the product for several consecutive sampling periods of the set frame number, it is confirmed as an environmental magnetic anomaly, and the anomaly label is latched until the system time exceeds the upper limit of the spanning time window. When an environmental magnetic anomaly is confirmed, the magnetic field navigation weight is reset to 0, and the visual navigation weight is reset to 1. In the control cycle before the weight switching is executed, the last valid magnetic field offset is extracted and the initial visual offset is subtracted to generate the coordinate zero-point offset constant and latch it. This anomaly decoupling mechanism utilizes the total voltage drop characteristic caused by the shielding effect of the metal material on the magnetic field lines to decouple the physical height attenuation from the signal distortion caused by the metal magnetic shielding, thereby triggering the state switching of the multi-sensor weights.

[0018] Further, the difference between the first normalized magnetic field signal and the second normalized magnetic field signal is calculated, and their sum is also calculated. The difference is divided by the sum with an added zero-point constant, and multiplied by a pre-calibrated linear mapping coefficient to output the magnetic field offset based on the difference ratio algorithm. The magnetic field offset is multiplied by the magnetic field navigation weight to obtain the first product, and the sum of the visual offset and the coordinate zero-point offset constant is multiplied by the visual navigation weight to obtain the second product. The first product and the second product are added to output the control target offset. Finally, the control target offset is input to a discrete position PID controller with an anti-integral saturation mechanism, and the target speed difference is calculated in closed loop. Combined with the basic feedforward driving linear velocity, the rolling radius of the chassis drive wheels is substituted to perform inverse kinematics, instructing the two wheels to generate differential speed to output the corrective yaw moment.

[0019] A second aspect of the present invention provides a posture adjustment system for a construction robot, comprising:

[0020] The feedforward prediction module is used to obtain the absolute physical distance from the chassis to the sill using a depth camera, and combined with the real-time longitudinal linear velocity obtained by the encoder, calculate the estimated time for the front wheel to contact the sill and establish a time window spanning the sill.

[0021] The pose calculation module is used to read the multidimensional attitude angles and Z-axis original acceleration of the inertial measurement unit, perform drift truncation and quadratic integration based on the system time and the boundary relationship of crossing the time window to obtain the dynamic displacement of the Z-axis, and calculate the real-time vertical air gap height of the Hall sensor arrays on both sides in combination with the physical dimensions of the chassis.

[0022] The magnetic field compensation module is used to reverse-compensate the original magnetic field signals on both sides into a first normalized magnetic field signal, a second normalized magnetic field signal, and a normalized total voltage based on a preset attenuation function and the real-time vertical air gap height.

[0023] The weight switching module is used to decouple the environmental magnetic anomaly based on the comparison result of the normalized total voltage and the metal interference characteristic threshold when the system time is within the time window, and to perform state switching between magnetic field navigation weight and visual navigation weight.

[0024] The differential control module is used to calculate the magnetic field offset based on the first normalized magnetic field signal and the second normalized magnetic field signal, calculate the visual offset based on the visual image, and output the control target offset by fusing the magnetic field navigation weight and the visual navigation weight to adjust the wheel differential.

[0025] This invention provides a method and system for adjusting the posture of a construction robot. It has the following beneficial effects:

[0026] 1. This invention utilizes an inertial measurement unit to perform drift truncation and quadratic integration across a time window to calculate the real-time Z-axis dynamic displacement and the sensor's vertical air gap height, and then uses this information to perform inverse compensation on the original magnetic field signal. This mechanism eliminates the magnetic field signal attenuation error caused by chassis bumps and posture tilting when the robot crosses ridges, maintaining the accuracy of underlying magnetic signal detection in uneven terrain.

[0027] 2. This invention decouples physical height attenuation from the magnetic field shielding phenomenon caused by the metal material by comparing the normalized total voltage with the metal interference characteristic threshold across a time window, and performs state switching between magnetic field navigation and visual navigation weights accordingly. This method avoids trajectory deviation caused by the failure of a single magnetic field navigation when the robot passes through a metal sill area, ensuring the stability of multi-source navigation data fusion and state transition.

[0028] 3. This invention utilizes absolute ranging from a depth camera and real-time linear velocity feedforward from the chassis to calculate the estimated time of front wheel contact with the curb, thus pre-constructing a time window as a trigger benchmark for subsequent pose calculation and weight switching. This prediction mechanism based on spatial distance and running speed provides accurate action timing for the underlying wheel differential control, improving the response efficiency and control coherence of robot posture correction. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention;

[0030] Figure 2 This is a system framework diagram of the present invention;

[0031] Figure 3 This is a flowchart illustrating the operational logic of the cross-time window prediction method of the present invention.

[0032] Figure 4 This is the operational logic diagram of pose calculation and height mapping based on state constraints of the present invention;

[0033] Figure 5 This is the operational logic diagram for the nonlinear normalization compensation of the magnetic field signal in this invention.

[0034] Figure 6 This is the operational logic diagram for environmental magnetic anomaly decoupling and navigation weight switching of the present invention;

[0035] Figure 7 This is the operational logic diagram for the anti-disturbance target offset calculation and closed-loop correction control of the present invention;

[0036] Figure 8 This is a comparative data chart showing the sill crossing performance of the present invention. Detailed Implementation

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see the appendix Figure 1 This invention provides a method for adjusting the posture of a construction robot, comprising the following steps:

[0039] S100 acquires elevator door image information and aligns the center position. It obtains the absolute physical distance between the construction robot chassis and the elevator door sill through the depth camera. Combined with the real-time longitudinal linear velocity obtained by the encoder, it calculates the estimated time for the front wheels to contact the sill and establishes the sill crossing time window.

[0040] S200 reads data from the chassis inertial measurement unit in real time to obtain multi-dimensional attitude angles and Z-axis original linear acceleration. Based on the boundary relationship between system time and time window, it performs drift truncation and quadratic integration to obtain the dynamic displacement of Z-axis. Combined with the physical dimensions of the chassis, it calculates the real-time vertical air gap height of the first Hall sensor array and the second Hall sensor array.

[0041] S300: Acquire the original magnetic field signals of the Hall sensor arrays on both sides, and compensate the original magnetic field signals in reverse to the nominal voltage value at the reference height according to the preset attenuation function, and output the first normalized magnetic field signal, the second normalized magnetic field signal and the normalized total voltage.

[0042] S400 determines whether the system time is within the time window. Based on the determination result and the comparison result between the normalized total voltage and the metal interference characteristic threshold, it decouples the environmental magnetic anomaly and performs the state switching between the magnetic field navigation weight and the visual navigation weight.

[0043] The S500 calculates the magnetic field offset based on the normalized magnetic field signal, calculates the visual offset based on the top camera image information, and integrates the navigation weights to output the final control target offset, which is then input to the chassis controller to calculate the target speed difference to adjust the wheel differential.

[0044] See attached document Figure 2 This invention provides a posture adjustment system for a construction robot, comprising:

[0045] The feedforward prediction module is used to acquire elevator door image information and align the center position. It obtains the absolute physical distance between the construction robot chassis and the elevator door sill through the depth camera, and calculates the estimated time for the front wheels to contact the sill by combining the real-time longitudinal linear velocity obtained by the encoder, and establishes the sill crossing time window.

[0046] The pose calculation module is used to read data from the chassis inertial measurement unit in real time to obtain multi-dimensional attitude angles and Z-axis original linear acceleration. Based on the boundary relationship between system time and time window, it performs drift truncation and quadratic integration to obtain Z-axis dynamic displacement, and calculates the real-time vertical air gap height of the first Hall sensor array and the second Hall sensor array in combination with the chassis physical dimensions.

[0047] The magnetic field compensation module is used to acquire the original magnetic field signals of the Hall sensor arrays on both sides, and to inversely compensate the original magnetic field signals to the nominal voltage value at the reference height according to the preset attenuation function, and output the first normalized magnetic field signal, the second normalized magnetic field signal and the normalized total voltage.

[0048] The weight switching module is used to determine whether the system time is within the time window. Based on the determination result and the comparison result of the normalized total voltage and the metal interference characteristic threshold, it decouples the environmental magnetic anomaly and performs the state switching between magnetic field navigation weight and visual navigation weight.

[0049] The differential control module is used to calculate the magnetic field offset based on the normalized magnetic field signal, calculate the visual offset based on the top camera image information, fuse the navigation weights to output the final control target offset, and input it to the chassis controller to calculate the target speed difference to adjust the wheel differential.

[0050] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of each step involved above and its internal processing flow.

[0051] See attached document Figure 3 , Figure 3 This is a flowchart illustrating the operational logic of cross-time window prediction according to an embodiment of the present invention. In the method provided by the present invention, step S100 specifically includes the following steps:

[0052] S101: The depth camera mounted at the front of the construction robot acquires image information and depth point cloud data of the elevator door directly in front. In engineering applications, to ensure the consistency of the visual ranging benchmark, the robot's motion axis needs to be spatially aligned with the physical environment target. Therefore, the control system performs edge feature extraction on the acquired image information, identifies the vertical edge line of the elevator door and the horizontal edge line of the bottom sill, and calculates the physical center coordinates of the elevator door in the image coordinate system. Based on this, the control system generates a lateral yaw control command to drive the chassis to turn, keeping the longitudinal centerline of the construction robot collinear with the center coordinates of the elevator door. In the chassis aligned state, the depth value of the corresponding sill area is extracted from the depth point cloud data. Considering that there is a fixed physical wheelbase difference between the depth camera installation position and the ground contact point of the construction robot's front wheels, the initial depth value read by the camera is subtracted from this physical wheelbase difference to calculate the absolute physical distance between the construction robot's front wheels and the elevator door sill. For image edge feature extraction and the transformation of depth data into a three-dimensional coordinate system of physical distance, those skilled in the art can use existing mature machine vision and coordinate projection algorithms for processing. The specific algorithm implementation is a well-known technology in this field and will not be described in detail here.

[0053] S102, the dual-channel hub motor encoders configured on the left and right drive wheels of the construction robot chassis collect pulse signals generated by wheel rotation in real time. The control system counts the number of received pulses according to the set high-frequency sampling period and calculates the actual angular velocity of the left and right drive wheels. Combined with the physical radius parameters calibrated at the factory of the drive wheels, the current longitudinal linear velocity of the chassis is calculated. Considering that the ground of construction sites often has gravel or water accumulation, which can easily cause local slippage of one side of the wheel, relying solely on the data of a single wheel or averaging without conditions will lead to speed measurement failure. In this embodiment, the absolute difference between the linear velocities of the left and right sides is monitored in real time. When the difference is less than the preset slip tolerance threshold, it indicates that the grip of the two wheels is balanced. At this time, the arithmetic mean of the linear velocities of the left and right sides is taken as the real-time longitudinal linear velocity. When the difference is greater than or equal to the slip tolerance threshold, it is determined that a slippage abnormality has occurred. The control system then adopts the single-side linear velocity that is closest to the historical speed average stored in the slippage time window (such as the first 0.5 seconds) before the slippage occurred, thereby enhancing the robustness of multi-source motion data fusion.

[0054] S103, Construct a kinematic mathematical model for the estimated time of the front wheel contacting the curb. Based on the fundamental physical principles of quasi-uniform linear motion in classical kinematics, the future spatial interference node can be deduced after obtaining the relative distance and instantaneous velocity. Based on the absolute physical distance and real-time longitudinal linear velocity obtained in the previous steps, the control system uses kinematic feedforward calculation logic to deduce the time node when the construction robot's front wheel arrives and contacts the curb. In particular, to prevent overflow anomalies where the denominator approaches zero in the underlying calculation, a limit check is performed on the velocity denominator before the division operation. The formula for calculating the estimated time is as follows:

[0055] ;

[0056] In the formula, The estimated time for the front wheels of the construction robot to contact the sill. This represents the absolute physical distance between the front wheels of the construction robot and the elevator door sill. For the real-time longitudinal linear velocity of the construction robot chassis, This is the set minimum safe effective linear velocity for the chassis. In this embodiment, The preferred value range is 0.02 m / s to 0.05 m / s, determined based on the minimum stable operating speed required for the robot's drive motor to overcome static friction. The physical purpose of this mathematical model is to proactively characterize the precise time point at which the robot makes mechanical contact with the physical ledge using the transient ratio of distance to velocity.

[0057] S104, Establish the sill crossing time window. Due to inherent noise in encoder measurements, variations in the ground friction coefficient, and physical response delays in issuing computational commands, the actual contact time will deviate slightly from the theoretical model. Therefore, a preset time error tolerance is introduced. The control system uses the predicted time as the reference center and extends this time error tolerance bidirectionally before and after the time axis, generating a closed-interval dynamic time window. The mathematical formula for the time window is as follows:

[0058] ;

[0059] In the formula, To establish a time window for crossing the threshold, This represents a preset time error tolerance. In this embodiment, The value range is set between 0.1 seconds and 0.3 seconds. The specific value of this parameter is calibrated based on the sum of the maximum communication delay time of the robot chassis controller's local area network bus and the motor's closed-loop response period. By establishing such a dynamic closed interval through the above formula, the most likely time domain for the robot to experience a sudden change in chassis attitude is accurately defined in a physical sense.

[0060] The system establishes the sill crossing time window as a global state trigger in the underlying control module. When the system's internal clock runs and enters the initial boundary of this time window, the control module's state machine switches from the normal smooth ground driving mode to the sill crossing preparatory state. This time window provides clear time-dimensional pre-constraints for the drift truncation of the inertial measurement unit's integrated data and the navigation weight state transitions of the multi-modal sensors in subsequent steps, avoiding control lag caused by the system relying on the passive response to physical vibrations after touching the sill.

[0061] See attached document Figure 4 , Figure 4 This is a logic diagram of state constraints, pose calculation, and height mapping according to an embodiment of the present invention.

[0062] Combined with appendix Figure 4 In addition to the overall architecture, the specific implementation of step S200, which involves multi-dimensional spatial pose calculation and inertial drift truncation under state constraints, may include the following sub-steps:

[0063] As a preferred approach, in S201, the pose calculation module reads the underlying data stream of the chassis inertial measurement unit in real time at a set high-frequency sampling rate via an internal high-speed data bus. In the multi-sensor fusion control architecture, considering that the sampling frequency of inertial sensors is usually much higher than that of vision and encoder modules, to avoid misalignment of state machine judgments caused by frequency differences in heterogeneous data sources, the pose calculation module timestamps each frame of inertial measurement data received in conjunction with a global clock.

[0064] For time nodes that are not strictly aligned, synchronous fusion of multi-source data in the time dimension is achieved by calling a zero-order hold or a linear interpolation algorithm. This allows for the extraction of the multi-dimensional attitude angles of the construction robot, specifically roll and pitch angles, and the simultaneous extraction of the original linear acceleration along the Z-axis. Regarding the bus communication interaction mechanism for the underlying data of the microelectromechanical system (MEMS) and the spatial attitude calculation algorithm based on quaternions, those skilled in the art can utilize existing SPI serial peripheral interface communication and complementary filtering algorithms. The specific communication and calculation processes are well-known technologies in this field and will not be elaborated upon here.

[0065] S202, perform time-bounded benchmark self-calibration and static drift truncation. Due to the inherent thermal noise and zero-bias divergence physical characteristics of inertial sensors, directly performing continuous quadratic integration of acceleration can easily amplify small static errors quadratically over time, leading to displacement calculations deviating from the true physical state. To suppress this divergence trend, this embodiment introduces a time constraint mechanism based on physical feedforward.

[0066] The pose calculation module compares the current clock time with the previously generated sill crossing time window in real time. When the current time is before the starting boundary of the time window, it is physically determined that the construction robot is moving smoothly on the flat concrete ground outside the elevator. Under this benchmark self-checking state, the pose calculation module forcibly executes the drift truncation command, locking the Z-axis dynamic displacement state quantity to zero, cutting off the accumulation path of static error at the source of the algorithm. At the same time, the pose calculation module uses a moving average filtering algorithm to perform mean calculation on the original linear acceleration of the Z-axis collected during this flat time period, dynamically extracting the background noise of the Z-axis acceleration of the sensor under the current temperature and operating conditions. The update formula for the background noise is as follows:

[0067] ;

[0068] In the formula, For dynamically updated Z-axis acceleration background noise, This is the sampling window length for the moving average algorithm. For the first timeline The original linear acceleration along the Z-axis for each sampling period. In engineering practice, considering that the number of historical samples may be insufficient when the main control equipment has just started up or just finished the last state adjustment, the pose calculation module will calculate the original linear acceleration if the cumulative number of samples is less than a set length. At that time, the mean is calculated by dynamically using the total number of samples that have actually been sampled as the denominator, in order to prevent division by zero or out-of-bounds errors from occurring at the underlying level.

[0069] In this embodiment, based on the commonly used sampling frequency of 100Hz to 200Hz for inertial measurement units, the sampling window length is... The value is preferably set to an integer between 50 and 100. This step, utilizing the robot's determined steady-motion phase, provides the basis for measuring and separating the sensor's static zero-bias data.

[0070] S203: After obtaining a relatively stable baseline noise level, further processing of the dynamic data at the moment of crossing the sill is required. When the current clock time enters the sill crossing time window, the control state machine transitions from the baseline self-checking state to the dynamic integration state. Within this time domain, the pose calculation module freezes and latches the final baseline noise level acquired at the moment before entering the time window.

[0071] The latched background noise is subtracted from the current real-time sampled Z-axis linear acceleration to reduce the cumulative effect of static deviations and extract the effective acceleration reflecting the physical bumps of the chassis. Subsequently, a second continuous-time integral calculation is performed on this effective acceleration. Since the microcontroller can only process discretized digital signals, in this embodiment, the continuous-time integral is specifically implemented through a trapezoidal numerical integration algorithm to approximate the actual physical displacement. Its theoretical mathematical operation logic is as follows:

[0072] ;

[0073] In the formula, The calculated and outputted dynamic Z-axis displacement at the current moment. The current clock time within the time window. This is the lower limit for the start time of the integral calculation. and These are all intermediate time variables during the integration process. Because the integration operation is constrained within a dynamic time window spanning a sill that is typically less than 1 second wide, this extremely short integration period can maintain acceptable accuracy in transient vertical displacement calculation without relying on external absolute position sensors.

[0074] S204 completes the geometric mapping from multi-dimensional spatial pose to sensor air gap height. After calculating the vertical displacement of the chassis geometric center, based on the principle of rigid body spatial geometric transformation, the lateral tilt of the chassis when crossing a sill usually leads to an asymmetric change in the height of the chassis edge. The pose calculation module calls the chassis physical geometric dimension parameters pre-configured in the underlying memory, and combines them with the calculated Z-axis dynamic displacement and roll angle to calculate in real time the real-time vertical air gap height of the first Hall sensor array and the second Hall sensor array distributed on both sides of the chassis relative to the ground magnetic grating plane. The mapping formula is as follows:

[0075] ;

[0076] ;

[0077] In the formula, This represents the real-time vertical air gap height of the first Hall sensor array on the left. This represents the real-time vertical air gap height of the second Hall sensor array on the right. This refers to the factory-calibrated reference air gap height for construction robots on flat ground. This value is determined by the chassis mechanical assembly height. The horizontal physical span between the geometric center of the first or second Hall sensor array and the longitudinal symmetry centerline of the robot chassis. For real-time roll angle, This refers to the longitudinal physical distance from the geometric center of the Hall sensor array to the chassis's transverse drive shaft. The real-time pitch angle is obtained from the aforementioned steps. The sign of the pitch angle component in the formula depends on whether the sensor array is mounted in front of or behind the drive shaft.

[0078] To ensure the physical consistency of the underlying control logic and prevent negative values ​​or exceeding the effective sensing range of the sensor due to extreme impacts, the pose calculation module rigidly limits the height value to a certain value before outputting the final air gap height. Within the safe working area, This is the maximum effective sensing distance calibrated according to the Hall sensor hardware manual. This spatial mapping step provides the necessary physical state parameters to support subsequent compensation for magnetic field measurement errors.

[0079] See attached document Figure 5 , Figure 5 This is a logic diagram for nonlinear normalization compensation of magnetic field signals according to an embodiment of the present invention.

[0080] Combined with appendix Figure 5 In relation to the overall architecture, the specific implementation of step S300, which involves nonlinear highly normalized compensation based on the magnetic field signal, may include the following sub-steps:

[0081] S301, after acquiring the vertical air gap height of the sensor arrays on both sides, the magnetic field compensation module reads the original magnetic field signals output by the first Hall sensor array and the second Hall sensor array in real time through the underlying hardware interface. The specific data flow path is as follows: after the Hall elements distributed on the left and right sides of the chassis sense the magnetic field of the magnetic grating ruler laid on the ground, they generate a weak analog voltage signal based on the Hall effect.

[0082] The analog signal undergoes hardware-level signal conditioning and anti-aliasing filtering via a front-end operational amplifier circuit before being input to the analog-to-digital converter (ADC) built into the main control chip. The ADC discretizes the continuous analog voltage into digital voltage values ​​and stores them in specific data registers.

[0083] Considering that industrial-grade linear Hall sensors typically have a factory-set static reference bias voltage in zero-magnetic-field environments, directly using the original sampled value would introduce calculation errors. In this embodiment, the magnetic field compensation module extracts the digital voltage value from the register according to a set control cycle, subtracts the static reference bias voltage obtained in a pre-calibrated non-magnetic environment, and uses the net voltage difference after deducting the bias as the original valid magnetic field signal on the corresponding side. For the sampling filtering of the analog-to-digital converter and the data transfer based on direct memory access, those skilled in the art can implement it using existing hardware configurations; the underlying hardware interface communication is well-known in the field and will not be described further here.

[0084] After obtaining the original magnetic field signal in the clean state, S302 needs to further establish a mathematical mapping relationship between the spatial magnetic field strength and the physical air gap height. According to the basic physical laws of electromagnetism, the spatial magnetic field strength exhibits a non-linear decay characteristic with the increase of the induction distance. If high-order polynomials or logarithmic functions are directly used for floating-point analytical operations in the underlying firmware, it will consume a lot of computing resources and cause delays in the execution of control commands.

[0085] To balance computational accuracy and system real-time performance, this embodiment employs a firmware-based calculation principle that combines a one-dimensional lookup table with a linear interpolation algorithm to fit the nonlinear attenuation function in the magnetic field compensation module. Developers pre-calibrated the magnetic field attenuation ratios corresponding to different discrete height points in a flat laboratory environment and stored them in non-volatile memory to form a one-dimensional lookup table. During actual operation, the real-time vertical air gap height calculated in the previous step is used as the index input. To prevent memory access errors caused by the air gap height exceeding the calibration range, boundary limiting is applied to the input height before the lookup table. When the input height is less than the minimum node height in the lookup table, the attenuation coefficient corresponding to the minimum node is directly output; when the input height is greater than the maximum node height, the attenuation coefficient corresponding to the maximum node is output. Within the safe range, if the real-time air gap height accurately matches a discrete node in the lookup table, the corresponding attenuation coefficient is directly read; if the real-time air gap height falls between two adjacent nodes, the linear interpolation algorithm is called to calculate the current dynamic attenuation coefficient. The mathematical formula for the interpolation operation is as follows:

[0086] ;

[0087] In the formula, To obtain the dynamic attenuation coefficient corresponding to the current real-time air gap height, The input is the real-time vertical air gap height, which is the real-time vertical air gap height of the first or second Hall sensor array calculated above. and These are the adjacent lower and upper discrete heights of the current height HH envelope in the one-dimensional lookup table, respectively. and These represent the calibrated attenuation coefficients corresponding to the discrete heights at the lower and upper bounds in the lookup table. Typically, the attenuation coefficient at the reference installation height is calibrated to 1, and as the height increases, the attenuation coefficient exhibits a non-linear decrease within the interval (0,1). Using this calculation principle, basic arithmetic and lookup table instructions replace complex non-linear function analysis, enabling the restoration of attenuation characteristics in resource-constrained microcontrollers.

[0088] S303, after solving for the dynamic attenuation coefficient, performs inverse mapping of the magnetic field signal. The multidimensional spatial attitude changes generated when the chassis crosses a sill manifest as asymmetrical fluctuations in the height of the air gaps on both sides, leading to spurious attenuation in the measured magnetic field voltage that is unrelated to horizontal position deviation. To effectively reduce this measurement error caused by physical height changes, the magnetic field compensation module uses the acquired dynamic attenuation coefficient to inversely compensate and map the original magnetic field signal to the nominal voltage range at the calibrated reference height. Simultaneously, to prevent division overflow caused by the attenuation coefficient approaching zero due to extremely high transient height, the attenuation coefficient is limited to a set minimum effective threshold before the division operation. Above, The value of can be determined based on the minimum voltage signal-to-noise ratio that the underlying hardware can distinguish, with a preferred range of 0.05 to 0.1. The output logic of the normalized magnetic field signal is as follows:

[0089] ;

[0090] ;

[0091] In the formula, The first normalized magnetic field signal is the output. The original effective magnetic field signal acquired after subtracting zero bias from the first Hall sensor array. Based on the left air gap height The left-side dynamic attenuation coefficient is calculated by interpolation from a table. This is the second normalized magnetic field signal output. The original effective magnetic field signal acquired after subtracting zero bias from the second Hall sensor array. Based on the right air gap height The dynamic attenuation coefficient on the right side is calculated by interpolation from a table. The minimum effective attenuation coefficient threshold is set.

[0092] Based on this, the magnetic field compensation module performs an arithmetic summation of the normalized magnetic field signals on both sides, outputting a normalized total voltage. The calculation formula is as follows:

[0093] ;

[0094] In the formula, This represents the normalized total voltage. In this embodiment, after the aforementioned inverse mapping process, the first and second normalized magnetic field signals significantly mitigate the spatial three-dimensional attitude coupling interference caused by chassis roll, pitch, and vertical swaying. The final output value primarily characterizes the degree of horizontal one-dimensional physical deviation of the construction robot relative to the ground magnetic ruler. Simultaneously, this normalized total voltage eliminates the influence of changes in physical distance, providing a parameter basis for subsequent steps to determine whether pure environmental magnetic field absorption caused by the elevator metal sill exists.

[0095] See attached document Figure 6 , Figure 6 This is a logic diagram of environmental magnetic anomaly decoupling and navigation weight switching according to an embodiment of the present invention.

[0096] Combined with appendix Figure 6 In relation to the overall control architecture, the specific implementation of step S400, namely the environmental magnetic anomaly decoupling and multimodal navigation weight hard switching, may include the following sub-steps:

[0097] S401, the navigation decision module runs a dual-time verification mechanism in parallel within the main control program to determine whether the construction robot is indeed in the sill-crossing phase. In actual construction environments, occasional scattered ferromagnetic construction waste on the ground can easily cause local magnetic field distortion. If the sill-crossing is determined solely by fluctuations in the magnetic field signal, it will typically lead to frequent false triggering of the navigation state machine. To improve the anti-interference capability of anomaly detection, this embodiment introduces the dynamic time window generated based on vision-kinematic feedforward in the aforementioned steps as a prerequisite constraint in terms of timing. The control state machine reads the high-frequency hardware clock in real time and compares it with the start and end boundaries of the dynamic time window. The time verification logic is as follows:

[0098] ;

[0099] In the formula, This is a Boolean flag for time verification; a value of 1 indicates that the time verification passed, and a value of 0 indicates that the time verification failed. This is the current real-time hardware clock time. The aforementioned estimated time for the front wheels of the construction robot to contact the curb, derived from kinematic deduction, is... This is the preset time error tolerance. and These are the lower and upper bounds of the dynamic time window, respectively.

[0100] when When the value is set to 1, it means that the current moment conforms to the theoretical expectation of crossing the threshold at the physical kinematic level, and the state machine grants the permission to detect subsequent magnetic field anomalies. Conversely, if the time does not fall within this window, the underlying logic will ignore any sudden magnetic field drop signals. This double check in the time dimension helps to shield against random electromagnetic interference in unexpected areas to a large extent.

[0101] S402, when the aforementioned time verification flag is valid and physical height interference has been isolated, the system needs to further quantitatively determine the environmental magnetic anomaly attributes. After the aforementioned nonlinear height normalization compensation, the output normalized total voltage is theoretically unaffected by vertical air gap changes caused by chassis bumps. On a flat concrete surface, this normalized total voltage should remain within a relatively stable nominal range. However, elevator sills are typically made of thick steel or stainless steel, which have strong magnetic shielding and absorption effects. When the sensor array moves directly above the sill, the magnetic field lines emitted by the magnetic grating ruler are significantly short-circuited or absorbed by the metal material, resulting in a significant abrupt attenuation of the normalized total voltage, independent of physical height. The navigation decision module utilizes this attenuation characteristic to construct a magnetic anomaly determination formula:

[0102] ;

[0103] In the formula, The real-time normalized total voltage output from the previous step is calculated. The reference total voltage constant is pre-calibrated on a flat, undisturbed ground. The threshold value for determining magnetic field attenuation is set.

[0104] In this embodiment, based on the magnetic permeability characteristics of common elevator sill materials, The preferred value range is 0.3 to 0.5. To avoid logical misjudgments caused by sampling glitches during a single analog-to-digital conversion, the state machine is equipped with a continuous frame counter. Only when the real-time normalized total voltage is continuous... Only when all sampling periods satisfy the above inequality conditions will the state machine formally confirm the current operating condition as an environmental magnetic anomaly. (Number of consecutive confirmation frames) Typically, the sampling rate is set to 3 to 5 frames based on the control bus sampling rate.

[0105] Furthermore, to prevent frequent jittering of the state machine caused by signal fluctuations at threshold edges, this embodiment introduces a state latching mechanism in the control logic. Once the operating condition is confirmed as an environmental magnetic anomaly, the state machine locks the anomaly tag until the system hardware clock cycles. The lock can only be released after the vehicle has driven out of the upper boundary of the aforementioned dynamic time window. This decision logic decouples and separates the false magnetoweakness caused by chassis attitude from the real magnetoweakness caused by environmental materials.

[0106] S403: Once the environmental magnetic anomaly is successfully confirmed, the main control program immediately initiates the state transition control strategy based on multimodal navigation weights. After determining that the system has entered a strong magnetic shielding area created by the sill, the reliability of the original ground magnetic grating ruler navigation loop significantly decreases. Continuing to rely on data from this side often leads to large deviations in the chassis lateral control. Therefore, the navigation decision module triggers a hard switch command for the underlying control route.

[0107] In this state transition logic, the control algorithm forcibly assigns a value of 0 to the magnetic field navigation weight parameter, thereby blocking the magnetic field offset from entering the control loop at the data flow level. Simultaneously, the preset visual navigation weight is synchronously awakened, and its parameter is abruptly set from 0 in the standby state to 1. The mathematical expression for the hard weight switching is as follows:

[0108] ;

[0109] In the formula, To assign control weights to the magnetic field navigation loop, To assign control weights to the visual navigation loop, This represents the current operating state of the control state machine. Represents a normal, stable driving state. This indicates a confirmed environmental magnetic anomaly.

[0110] To avoid severe mechanical step vibration of the chassis during the switching moment due to the zero-point deviation between the underlying coordinate systems of magnetic field navigation and visual navigation, the navigation decision module extracts the last effective lateral positioning deviation of the current magnetic field loop as the initial bias constant in the control cycle before executing the above-mentioned weight change. In subsequent calculations, it is superimposed on the visual positioning reference to achieve a smooth handover of control flow.

[0111] After the weight switching, the main control chip immediately retrieves high-frame-rate real-time images captured by industrial cameras positioned on top of the construction robot's car. This top camera array is primarily used to observe the elevator door guide rails or visual guidance markers on the top, providing a lateral positioning reference in blind spots lacking ground magnetic field characteristics. For the image acquisition protocol, visual feature recognition and matching, and pixel offset calculation based on the pinhole camera model, those skilled in the art can utilize existing visual positioning algorithm libraries for deployment. The specific image processing and positioning mechanisms are well-known technologies in the field and will not be elaborated upon here. Through the aforementioned hardware-level weight blocking and wake-up strategy, it is helpful to ensure that the underlying guidance closed loop does not experience control blind spots when the robot traverses complex magnetic interference media.

[0112] See attached document Figure 7 , Figure 7 This is a logic diagram for calculating the anti-disturbance target offset and implementing closed-loop correction control according to an embodiment of the present invention.

[0113] Combined with appendix Figure 7 In relation to the overall control architecture, the specific implementation of step S500 may include the following sub-steps:

[0114] In S501, within the main control chip, the offset calculation module performs parallel processing of heterogeneous sensor data to acquire magnetic field and visual offsets in real time. Regarding the magnetic field loop, considering the potential for gaps in the splicing of the magnetic grating ruler laid on the ground or uneven magnetization, the absolute magnetic field amplitude often exhibits global fluctuations. To effectively suppress this common-mode interference, the offset calculation module employs a difference ratio algorithm to perform dimensionless processing on the normalized magnetic field signals from both sides output in step S300, thereby calculating the actual physical offset. The formula for calculating the magnetic field offset is as follows:

[0115] ;

[0116] In the formula, The calculated magnetic field offset is... and These are the first normalized magnetic field signal and the second normalized magnetic field signal, respectively. The smallest positive constant fixed in the program is preferred to be 10. -5This is designed to prevent a bottom-level division-to-zero crash caused by an abnormal power outage of the sensor. In this embodiment, the magnetic field difference ratio and the linear mapping coefficient from numerical value to physical lateral distance are used. The value of is determined by the physical span of the Hall sensor array and calibration experiments. Technicians can deduce this constant coefficient by inducing a known lateral displacement of the chassis on a horizontal test bench. When the center of the robot chassis is precisely aligned with the central axis of the magnetic scale, and The expectation is close, at this time It is zero.

[0117] Within the same clock cycle of magnetic field data computation, the vision loop is synchronously in a data-parallel ready state. The top camera array captures images of the elevator car door guide rails or top guide strips, extracting the centerline of the guide features. For image grayscale conversion, edge feature extraction, and pixel-to-physical coordinate mapping transformation based on the camera intrinsic parameter matrix, those skilled in the art can use existing machine vision algorithm libraries. The image processing and coordinate system mapping mechanisms are well-known technologies in the field and will not be elaborated upon here. After coordinate transformation, the vision loop outputs a visual offset that matches the current environment. .

[0118] In step S502, after parallel calculation of the two independent positioning errors, the system dynamically schedules the heterogeneous information sources at the algorithm level through a multimodal fusion strategy. The navigation decision module calls the currently active multimodal navigation weights output in step S400 to perform weighted fusion of the heterogeneous sensor offsets, generating the final control target offset used to execute the kinematic closed loop at the underlying level. The fusion formula is as follows:

[0119] ;

[0120] In the formula, This is the control target offset that the system ultimately outputs. and These are the control weights assigned to the magnetic field navigation loop and the visual navigation loop, respectively. The calculated magnetic field offset, The calculated visual offset, This is the coordinate zero-point offset constant latched at the moment of multimodal weight switching, used to compensate for the physical installation deviation between the origin of the camera coordinate system and the origin of the magnetic sensor coordinate system.

[0121] because and Controlled by a state machine and mutually exclusive between 0 and 1, this fusion formula will only adopt the measurement result of one source as the dominant control quantity at any single moment, thus avoiding the mutual conflict of heterogeneous sensor input vectors from a logical mechanism and maintaining the uniqueness of control commands.

[0122] S503, Control target offset After generation, it is passed as an error input to the chassis kinematics closed-loop controller. In this embodiment, the chassis closed-loop controller uses a discrete position PID (proportional-integral-derivative) algorithm to calculate the target additional angular velocity that can correct the deviation based on the current lateral deviation and its rate of change over time. The operational equation of the closed-loop controller is as follows:

[0123] ;

[0124] In the formula, For the first The target speed difference calculated in each control cycle This is the index of the current discrete control cycle. This is the index variable for the discrete time step starting from the start of the control system or when the integral is cleared. and The first The and the first The deviation input amount of each (i.e., the previous cycle) control cycle, of which The value is equivalent to the current period's. , For the first The deviation input amount per control cycle , and These are the proportional, integral, and derivative control gain parameters, which can be tuned on a real vehicle using the critical proportional gain method in conjunction with the chassis's mass inertia.

[0125] Meanwhile, considering that the prolonged existence of lateral deviation can easily lead to problems in the integral term during hard switching of multimodal weights or when the chassis encounters physical obstacles, the situation is also considered. Infinite accumulation can lead to integral divergence (Windup phenomenon), so the closed-loop controller introduces an anti-integral saturation mechanism at the algorithm's underlying level.

[0126] When the target speed difference reaches the set physical boundary, the controller will stop the error accumulation calculation in the current direction. The closed-loop controller internally processes the final output... Upper and lower limit protection was implemented, restricting it to... Within the safe zone, The speed was determined based on the maximum rated speed of the hub motor and the chassis's mechanical anti-rollover safety specifications.

[0127] After calculating the target speed difference, the controller combines it with the basic feedforward linear velocity of the construction robot chassis and distributes it to the drive wheels on both sides via inverse kinematics. The physical response equation for wheel differential adjustment is as follows:

[0128] ;

[0129] ;

[0130] In the formula, and These are the real-time control angular velocities sent to the left and right wheel hub motors, respectively. The basic straight-line angular velocity issued by the global path planner. This represents the effective physical rolling radius of the chassis drive wheels. When the chassis tends to drift to the left, the calculated target speed difference will increase the speed of the left wheel and decrease the speed of the right wheel, thus generating a yaw moment to the right, forcing the robot to correct its course. This torque-based physical response mechanism is expected to enable the construction robot to maintain relatively stable lateral tracking accuracy when crossing elevator sills with complex magnetic interference, helping to mitigate the physical risk of colliding with elevator door frames.

[0131] To further aid in understanding the present invention, a specific application embodiment is provided below, along with experimental verification and effect comparison.

[0132] Specific application examples:

[0133] Scene and parameter settings:

[0134] In this specific application embodiment, the object is an indoor construction and material handling robot using a dual-wheel differential drive. Its main task is to autonomously transport heavy building materials such as tiles into a construction elevator along a magnetic grid ruler pre-set on the ground.

[0135] The relevant basic physical parameters are as follows:

[0136] The reference air gap height of the magnetic sensor as specified by the chassis manufacturer. =30mm;

[0137] Horizontal physical span of Hall sensor array =300mm, longitudinal physical distance from the drive shaft =200mm;

[0138] Reference total voltage constant calibrated on a flat, undisturbed ground =3.0V; Magnetic field attenuation threshold =0.4;

[0139] Effective physical rolling radius of hub motor =0.1m.

[0140] Time window prediction (step S100): When the robot approaches the elevator, the visual depth camera measures the absolute physical distance between the chassis and the stainless steel sill in front to be 1.2m. At this time, the longitudinal linear velocity calculated by the encoder is 0.4m / s. The feedforward module calculates the estimated time of contact with the sill. =1.2 / 0.4=3.0s. Set the error tolerance. =0.2s, thus establishing the dynamic time window for crossing the sill as [2.8s, 3.2s].

[0141] Pose calculation and height mapping (step S200): When the system clock reaches 2.8s, the state machine triggers a dynamic integral state, latching the background noise of the flat ground. At the instant the front wheel contacts the curb, the IMU calculates that the chassis has generated a pitch angle of +0.08 rad (the front of the vehicle rises) and a dynamic Z-axis displacement of +15 mm. The pose calculation module uses the corrected kinematic formula that includes roll and pitch angles to accurately calculate that the instantaneous true air gap height of the left and right Hall sensors has changed abruptly from 30 mm to approximately 48 mm.

[0142] Magnetic field nonlinearity compensation (step S300): Due to the increase in physical height, the original total magnetic field voltage drops sharply to 1.5V. Without compensation, the system is very likely to misjudge that it has entered the metal sill area. At this time, the magnetic field compensation module uses the 48mm air gap height as an index to interpolate and calculate the attenuation coefficient from a one-dimensional lookup table, and inversely compensates the original signal to about 2.9V (close to the reference value of 3.0V), effectively eliminating the false magnetic weakness phenomenon caused by vehicle body bumps.

[0143] Abnormal decoupling and weight switching (step S400): When the sensor array is truly hovering directly above the stainless steel sill (at which point the clock is within the time window), due to strong metal absorption, the compensated normalized total voltage drops sharply to 0.9V, satisfying... The anomaly detection condition is ≤0.4×3.0V=1.2V. The state machine immediately adjusts the magnetic field weight. Set to 0, visual weight Set to 1. Simultaneously record the difference between the last effective magnetic field offset (e.g., +5mm) and the initial visual offset (e.g., +2mm), and generate the coordinate zero-point offset constant. =3mm, enabling seamless switching.

[0144] Closed-loop correction (step S500): During the vision-driven phase, the system detects a tendency for the robot to drift 10mm to the left due to ground bumps. The chassis closed-loop controller calculates the target rotational speed difference. Subsequently, based on the correct inverse kinematics solution, the left wheel speed was increased and the right wheel speed was decreased, generating a yaw torque to the right. The robot quickly made a slight adjustment to the right, successfully aligning itself with the elevator door's central axis and entering the car without colliding with the door frame.

[0145] Experimental verification and effect comparison:

[0146] To verify the reliability and technical advantages of the present invention, the applicant built a simulation verification platform containing a standard elevator sill (45mm high, 120mm wide, made of 304 stainless steel) in a standard test site and conducted comparative experiments.

[0147] Comparison method settings

[0148] Traditional control group: adopts conventional magnetic field navigation, without IMU Z-axis height compensation, uses a fixed all-time low-pressure threshold to trigger visual switching, and has no predictive time window constraints.

[0149] Experimental group of this invention: equipped with the zero-jitter hard handover logic of the prediction time window, IMU pose compensation (including pitch and roll angle correction) and coordinate offset compensation of this invention.

[0150] Each group of robots underwent 50 fully automated elevator entry and crossing tests, with a set base travel speed of 0.4 m / s.

[0151] The experimental data are shown in Table 1:

[0152]

[0153] in conclusion:

[0154] According to Table 1 and Figure 8 Based on the data, the attitude adjustment method proposed in this invention achieves substantial improvements in all core control indicators when crossing elevator sills with complex magnetic interference and physical bumps. Thanks to the introduction of the feedforward prediction time window and the multi-dimensional pose air gap compensation mechanism, this scheme reduces the navigation mode switching delay from 185ms to 15ms and eliminates the 12 false anomaly triggers that occurred in the traditional control group, effectively verifying the accuracy of the environmental magnetic anomaly decoupling logic and the efficiency of the system's timing response.

[0155] Furthermore, at the underlying closed-loop execution level, the weight transfer strategy based on coordinate zero-point offset compensation effectively suppressed the transient jitter during multimodal switching to 1.8mm, significantly reducing the maximum lateral deviation of the chassis during crossing from 48.5mm to 12.2mm. The synergistic effect of these technologies ultimately increased the robot's collision-free elevator entry success rate from 76% to 100%, effectively ensuring the smoothness of control and motion safety of the construction robot during switching in complex heterogeneous environments.

[0156] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for adjusting the posture of a construction robot, characterized in that, Includes the following steps: The absolute physical distance from the chassis to the sill is obtained by a depth camera, and combined with the real-time longitudinal linear velocity obtained by the encoder, the estimated time for the front wheel to contact the sill is calculated and a time window is established. The multidimensional attitude angle and Z-axis original acceleration of the inertial measurement unit are read. Drift truncation and double integration are performed according to the system time and the boundary relationship of the time window to obtain the dynamic displacement of the Z-axis. The real-time vertical air gap height of the Hall sensor arrays on both sides is calculated in combination with the physical dimensions of the chassis. Based on the preset attenuation function and the real-time vertical air gap height, the original magnetic field signals on both sides are reverse-compensated into a first normalized magnetic field signal, a second normalized magnetic field signal, and a normalized total voltage. When the system time is within the time window, the environmental magnetic anomaly is decoupled based on the comparison result of the normalized total voltage and the metal interference characteristic threshold, and the state switching between magnetic field navigation weight and visual navigation weight is performed. The magnetic field offset is calculated based on the first normalized magnetic field signal and the second normalized magnetic field signal. The visual offset is calculated based on the visual image. The magnetic field navigation weight and the visual navigation weight are fused to output the control target offset in order to adjust the wheel differential.

2. The method for adjusting the posture of a construction robot according to claim 1, characterized in that, The step of combining the real-time longitudinal linear velocity obtained from the encoder to calculate the estimated time for the front wheel to contact the curb and establishing a crossing time window specifically includes: The absolute difference between the linear velocities of the left and right drive wheels is monitored. When the absolute difference is less than a preset slip tolerance threshold, the arithmetic mean is taken as the real-time longitudinal linear velocity. When the absolute difference is greater than or equal to the slip tolerance threshold, the average velocity within the historical slip time window is adopted as the real-time longitudinal linear velocity. The estimated time is obtained by dividing the absolute physical distance by the real-time longitudinal linear velocity after amplitude limiting verification, and a preset time error tolerance is extended in both directions before and after the estimated time as a reference to generate the spanning time window of the closed interval.

3. The method for adjusting the posture of a construction robot according to claim 1, characterized in that, The drift truncation based on the relationship between the system time and the boundary of the time window specifically includes: When the system time is before the starting boundary of the time window, it is determined to be a flat driving state, and the drift cutoff command is forcibly executed to lock the dynamic displacement of the Z-axis to zero; The mean value of the original Z-axis acceleration before the starting boundary is calculated using a moving average filtering algorithm to extract and update the background noise of the Z-axis acceleration.

4. The method for adjusting the posture of a construction robot according to claim 3, characterized in that, The second integral yields the dynamic displacement along the Z-axis, which, combined with the chassis physical dimensions, is used to calculate the real-time vertical air gap height of the Hall sensor arrays on both sides. Specifically, this includes: When the system time enters the time-crossing window, the last Z-axis acceleration background noise before entering the time-crossing window is latched, and the real-time Z-axis original acceleration is subtracted from the last Z-axis acceleration background noise to obtain the effective acceleration; The effective acceleration is subjected to continuous trapezoidal discrete numerical integration to obtain the dynamic displacement along the Z-axis. The horizontal physical span and longitudinal physical distance in the chassis physical dimensions are called in, and spatial geometric mapping is performed in combination with the Z-axis dynamic displacement and the roll angle and pitch angle in the multi-dimensional attitude angles to calculate the real-time vertical air gap height, and the sensing distance boundary is limited.

5. The method for adjusting the posture of a construction robot according to claim 1, characterized in that, The method, based on a preset attenuation function and the real-time vertical air gap height, inversely compensates the original magnetic field signals on both sides into a first normalized magnetic field signal, a second normalized magnetic field signal, and a normalized total voltage, specifically including: The static reference bias voltage is subtracted from the original magnetic field signals on both sides obtained respectively; Using the real-time vertical air gap height as an index, a preset one-dimensional lookup table is retrieved and a linear interpolation algorithm is called to calculate the corresponding dynamic attenuation coefficient; The original magnetic field signal after deducting the bias voltage is divided by the dynamic attenuation coefficient limited by the lower boundary for inverse compensation, and the first normalized magnetic field signal and the second normalized magnetic field signal are output respectively. The two are then arithmetically summed to output the normalized total voltage.

6. The method for adjusting the posture of a construction robot according to claim 1, characterized in that, The decoupling of environmental magnetic anomalies based on the comparison result of the normalized total voltage and the metal interference characteristic threshold specifically includes: Determine whether the normalized total voltage is less than or equal to the product of the flat ground reference total voltage and the metal interference characteristic threshold; If the normalized total voltage is less than or equal to the product for a consecutive set number of frame sampling periods, then an environmental magnetic anomaly is confirmed, and the anomaly tag is latched until the system time crosses the upper boundary of the time window.

7. The method for adjusting the posture of a construction robot according to claim 1, characterized in that, The state switching between the execution of magnetic field navigation weights and visual navigation weights specifically includes: After confirming the environmental magnetic anomaly, the magnetic field navigation weight is abruptly set to 0, and the visual navigation weight is abruptly set to 1; In the control cycle preceding the weight mutation, the last valid magnetic field offset is extracted and the initial visual offset is subtracted to generate the coordinate zero-point offset constant and latch it.

8. The method for adjusting the posture of a construction robot according to claim 1, characterized in that, The calculation of the magnetic field offset based on the first normalized magnetic field signal and the second normalized magnetic field signal specifically includes: Calculate the difference between the first normalized magnetic field signal and the second normalized magnetic field signal, and calculate the sum of the two; Divide the difference by the sum with an added zero minimum constant, and multiply by a pre-calibrated linear mapping coefficient to output the magnetic field offset based on the difference ratio algorithm.

9. The method for adjusting the posture of a construction robot according to claim 7, characterized in that, The process of integrating the magnetic field navigation weights and the visual navigation weights to output a control target offset in order to adjust the wheel differential speed specifically includes: The magnetic field offset is multiplied by the magnetic field navigation weight to obtain a first product. The sum of the visual offset and the coordinate zero point offset constant is multiplied by the visual navigation weight to obtain a second product. The first product and the second product are added together to output the control target offset. The target offset is input to a discrete position PID controller that includes an anti-integral saturation mechanism, and the target speed difference is calculated in closed loop. The target speed difference is combined with the basic feedforward driving linear velocity, and the rolling radius of the chassis drive wheel is substituted to perform inverse kinematics solution, instructing the wheels on both sides to generate differential speed to output yaw correction torque.

10. A construction robot posture adjustment system, applied to the construction robot posture adjustment method as described in any one of claims 1-9, characterized in that, include: The feedforward prediction module is used to obtain the absolute physical distance from the chassis to the sill through the depth camera, and combine it with the real-time longitudinal linear velocity obtained by the encoder to calculate the estimated time for the front wheel to contact the sill and establish a time window spanning the sill. The pose calculation module is used to read the multidimensional attitude angle and Z-axis original acceleration of the inertial measurement unit, perform drift truncation and double integration according to the system time and the boundary relationship of the time window to obtain the dynamic displacement of the Z-axis, and calculate the real-time vertical air gap height of the Hall sensor arrays on both sides in combination with the physical dimensions of the chassis. The magnetic field compensation module is used to inversely compensate the original magnetic field signals on both sides into a first normalized magnetic field signal, a second normalized magnetic field signal, and a normalized total voltage based on a preset attenuation function and the real-time vertical air gap height. The weight switching module is used to decouple the environmental magnetic anomaly based on the comparison result of the normalized total voltage and the metal interference characteristic threshold when the system time is within the time window, and to perform state switching between magnetic field navigation weight and visual navigation weight. The differential control module is used to calculate the magnetic field offset based on the first and second normalized magnetic field signals, calculate the visual offset based on the visual image, and output the control target offset by fusing the magnetic field navigation weight and the visual navigation weight to adjust the wheel differential.

Images