A combined positioning system and method for ship segmented transport vehicles based on multi-source sensor IoT data acquisition.

Abstract

This invention discloses a combined positioning system and method for ship segmented transport vehicles based on multi-source sensor IoT acquisition. The system includes an IoT sensing system and an adaptive fusion positioning system. The IoT sensing system includes a wheel speedometer with a separate structure, an inertial measurement unit, and two sets of RTK antennas deployed diagonally on the vehicle platform. The positioning method includes: extracting wheel speed data, inertial data, and satellite positioning data from the IoT sensing system; performing dual-branch feature calculation to identify the chassis motion mode and obtain the relative motion state of the chassis and calculate the absolute coordinates of the vehicle's corner points; constructing a two-layer extended Kalman filter architecture, and dynamically adjusting the filter parameters according to the real-time sensor status through an online adaptive regulator of the covariance matrix, and fusing the output positioning results. This invention enables ship segmented transport vehicles to achieve high-precision positioning under conditions of no steering angle sensor, heavy load deformation, and weak GNSS.

Description

Technical Field

[0001] This invention relates to intelligent vehicle transportation and spatial fusion positioning, and in particular to a combined positioning system and method for ship segmented transport vehicles based on multi-source sensor IoT data acquisition. Background Technology

[0002] Ship section transport vehicles are core equipment in shipyards and ports for transferring large section components. Due to limitations imposed by the dense metal structures in the operating environment, tire radial deformation under heavy loads, and blind spots caused by the vehicle's large size, existing positioning solutions have the following shortcomings:

[0003] At the hardware level, traditional wheel speed gauges are installed on the rotating parts of the wheel, which poses risks of cable entanglement and slip ring wear. Furthermore, they cannot adapt to changes in wheel diameter caused by heavy loads, resulting in severe mileage drift.

[0004] At the positioning algorithm level, existing technologies face dual deficiencies in both underlying kinematics estimation and top-level fusion mechanisms. On the one hand, large flatbed trucks have numerous axle groups and high degrees of freedom, exhibiting various complex motion modes such as straight-line movement, lateral movement, and rotation. Due to limitations in sensor installation, the system can only acquire wheel speeds at the four corners, lacking crucial wheel steering angle observations. Existing estimation algorithms lack multi-motion-mode recognition mechanisms and rely excessively on a single, simplified kinematic model, making it impossible to achieve high-precision forward kinematics solutions for the chassis based solely on local wheel speeds when steering angles are missing. This results in significant cumulative errors in the front-end mileage estimation. On the other hand, in back-end fusion positioning, conventional GNSS navigation experiences a sharp drop in positioning accuracy, or even divergence failure, when faced with signal jumps or rejections caused by metal canopies and large segmented obstructions. Furthermore, single or co-located GNSS antennas are prone to signal interruptions when obstructed by irregularly shaped cargo segments, and there is a lack of anti-obstruction optimization design for weak GNSS environments. Furthermore, existing fusion algorithms lack effective covariance adaptive adjustment capabilities when faced with complex conditions such as front-end inference distortion and back-end observation degradation, which severely limits the overall robustness of multi-source fusion positioning systems. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a combined positioning system and method for ship segment transport vehicles based on multi-source sensor IoT data acquisition that can accurately and stably locate the vehicle under conditions of no steering angle observation, heavy load deformation, and weak GNSS.

[0006] Technical solution: The ship segment transport vehicle combined positioning system based on multi-source sensor IoT acquisition described in this invention includes an IoT sensing system and an adaptive fusion positioning system;

[0007] The IoT sensing system includes a wheel speedometer, an inertial measurement unit, and an RTK positioning component. The RTK positioning component includes two sets of antennas deployed diagonally on the vehicle platform. The wheel speedometer is configured to have a rotating part that moves with the wheel and a measuring part that maintains an attitude stable relative to the ground.

[0008] The adaptive fusion positioning system communicates with the IoT sensing system and is configured to perform positioning calculations, including a feature data parsing module and a two-layer adaptive fusion filtering module.

[0009] The feature data parsing module is used to correct the wheel diameter corresponding to the wheel speed data based on the vehicle load, identify the chassis motion mode by combining inertial measurement data and obtain the relative motion state of the chassis by forward solving, and evaluate the signal quality and optimize the benchmark of satellite positioning data to calculate the coordinates of the four corners of the vehicle body as a global position reference.

[0010] The dual-layer adaptive fusion filtering module includes a local extended Kalman filter node, a global extended Kalman filter node, and an online adaptive covariance matrix adjuster. The online adaptive covariance matrix adjuster is configured to dynamically adjust the covariance matrix parameters of the two extended Kalman filter nodes based on the evaluation results of satellite signal quality and chassis motion pattern recognition. The local extended Kalman filter node fuses the chassis relative motion state to output a local odometry, and the global extended Kalman filter node fuses the local odometry with the global position reference to output the final positioning result.

[0011] By setting up an IoT sensing system including wheel speed sensors, inertial measurement units (IMUs), and RTK positioning components, a multi-source heterogeneous data foundation was provided for subsequent fusion positioning. The wheel speed sensor employs a design where the rotating and measuring components are relatively separated, ensuring the measuring component maintains attitude stability during wheel rotation. This physically avoids cable entanglement and signal slip ring wear issues, guaranteeing stable wheel speed signal output. The feature data parsing module introduces a motion pattern recognition mechanism, combining inertial measurement data to complete the forward kinematics of the chassis even in the absence of a physical steering angle sensor, eliminating the calculation errors inherent in complex kinematic models. The dual-layer adaptive fusion filtering module uses a decoupled extended Kalman filter architecture. Local nodes process high-frequency chassis and inertial measurement data, while global nodes utilize absolute position to eliminate accumulated errors. Combined with an online adaptive covariance matrix adjuster, the fusion weights are dynamically adjusted based on the real-time sensor status, achieving high-precision and robust positioning under heavy load deformation, weak satellite signals, and complex motion modes.

[0012] Preferably, each of the two sets of antennas includes a main antenna and a secondary antenna; the antennas of the same set are arranged on the same side of the vehicle body to form a first baseline vector pointing to the rear of the vehicle, and the main antennas and secondary antennas of different sets are arranged at diagonal positions on the vehicle body platform; the RTK positioning components are arranged at the four corners of the vehicle body through an integrated co-frame structure, and the integrated co-frame structure is also equipped with ultrasonic sensor components; the integrated co-frame structure includes a fixed base and a bracket slot connected thereto, and the RTK positioning components and ultrasonic sensor components are connected to the bracket slot.

[0013] The main and secondary antennas of the same antenna group are arranged along the same side of the vehicle body, and the baseline vector formed by them is parallel to the longitudinal centerline of the vehicle body, establishing a defined spatial geometric constraint relationship, which facilitates the direct acquisition of heading information. Different groups of main and secondary antennas are arranged diagonally across the vehicle platform, making the two antenna groups physically independent. When the signal of the antenna on one side of the vehicle is attenuated due to cargo or environmental obstruction, the antenna at the opposite diagonal position on the other side can still maintain effective satellite reception, enhancing the system's signal robustness under complex obstruction conditions. The integrated co-frame structure integrates the RTK positioning components and ultrasonic sensor components at the four corners of the vehicle body, simplifying the installation layout and improving structural compactness.

[0014] Preferably, the wheel speed gauge includes a rotating base fixed to the wheel hub and equipped with a sensing gear ring, and a suspended measuring housing sleeved on the rotating base via a bearing; the suspended measuring housing is provided with a sensor that cooperates with the sensing gear ring to measure speed, and its bottom is provided with a gravity anchoring mechanism for maintaining a fixed attitude; the gravity anchoring mechanism is an eccentric counterweight, the center of mass of which is located vertically below the bearing axis, and the restoring torque generated by its gravity is greater than the maximum frictional torque of the bearing at the operating speed; the edge of the suspended measuring housing is provided with a flange, and the outer periphery of the rotating base is provided with a groove, the flange and the groove cooperating to form a non-contact labyrinth seal structure.

[0015] By positioning the center of mass of the eccentric counterweight vertically below the bearing axis, a stable balance is achieved using the principle of minimum gravitational potential energy. The design ensures that the restoring torque exceeds the maximum frictional torque, guaranteeing absolute stability of the suspended measuring housing under all operating conditions and ensuring accurate wheel speed measurement. The non-contact labyrinth seal structure, through staggered flanges and grooves forming a tortuous gap, effectively blocks the intrusion of external dust, mud, and other contaminants while avoiding the sliding friction resistance of traditional contact seals. This helps reduce the overall bearing frictional torque and assists the gravity anchoring mechanism in more reliably maintaining the housing's attitude.

[0016] As a preferred embodiment, the IoT sensing system also includes a control cabinet component, a sensing module, and a positioning module;

[0017] The control cabinet component, located on the vehicle platform as a data aggregation center, is used to perform timestamp synchronization and format parsing of multi-source heterogeneous data, and to calculate the real-time vehicle pose, task progress and equipment health status based on the parsed data. It also monitors data boundaries in real time and triggers alarms and cuts off the power of the actuators when there is abnormal intrusion, failure or loss of positioning.

[0018] The perception module includes an ultrasonic sensor assembly, a lidar assembly, and a vision sensor assembly. The ultrasonic sensor assembly is distributed in the lifting mechanism and is used to provide feedback on the lifting height and balance status. The lidar assembly includes a 2D lidar embedded in the opening of the vehicle platform and arranged at a preset angle to detect the position of segmented goods in the overhead transport space, and a 3D lidar installed at the corner edge of the vehicle platform and below the platform plane to obtain a 270-degree field of view for obstruction perception and avoidance. The vision sensor assembly is fixed at the center of the front and rear ends of the vehicle platform for visual tag reading.

[0019] The positioning module is used to transmit coordinates, pulse counts, point clouds, and raw inertial navigation data to the control cabinet components. It includes an RTK positioning component, wheel speedometer, 3D LiDAR, and an inertial measurement unit installed at the physical center of gravity of the vehicle platform.

[0020] By dividing the hardware system into control cabinet components, sensing modules, and positioning modules, a clear decoupling of data aggregation, environmental perception, and positioning functions is achieved. The control cabinet components integrate a complete logical chain of data cleaning, state calculation, and safety decision-making, providing core control assurance for vehicle operation. The sensing module, through a combination of ultrasonic, LiDAR, and visual sensors, achieves comprehensive monitoring of the lifting status, transport space, and surrounding obstacles. The 3D LiDAR adopts a recessed layout below the platform plane, effectively extending the field of view to approximately 270 degrees, eliminating near-field perception blind spots for wide-body transport vehicles in narrow passages.

[0021] The method for combined positioning of ship segmented transport vehicles based on multi-source sensor IoT data acquisition as described in this invention includes:

[0022] S1. Extract the raw physical pose information required for positioning from the IoT sensing system. The raw information includes at least the wheel speed data of the wheel speed meter, the inertial measurement data of the inertial measurement unit, and the satellite positioning data of the two antennas in the RTK positioning component.

[0023] S2. Perform dual-branch feature calculation on the original information. The first branch is based on inertial measurement data and wheel speed data to perform chassis motion pattern recognition and forward kinematics of the chassis, and outputs the relative motion state of the chassis. The second branch is based on satellite positioning data to perform spatial reference analysis and outputs the absolute coordinate reference of the vehicle body.

[0024] S3. Construct a two-layer filtering architecture that includes local extended Kalman filter nodes and global extended Kalman filter nodes. Input the chassis relative motion state and the vehicle absolute coordinates into the two-layer filtering architecture, and adaptively update the local and global nodes by dynamically adjusting the filtering covariance matrix online, and fuse the output positioning results.

[0025] Step S1 extracts heterogeneous physical layer data from multiple sources, providing a unified input for subsequent processing. Step S2 employs parallel feature solving with two branches. The first branch combines wheel speed and inertial measurement data to complete chassis motion pattern recognition and forward kinematics, acquiring the relative motion state of the chassis even in the absence of a physical steering angle sensor, thus eliminating errors in the calculation principle of complex kinematic models at the source. The second branch performs spatial benchmark analysis based on satellite positioning data, providing a global absolute position reference. Step S3 adopts a decoupled local and global two-layer extended Kalman filter architecture. Local nodes ensure smooth high-frequency tracks, while global nodes eliminate accumulated errors. Furthermore, online dynamic adjustment of the covariance matrix enables adaptive fusion weight allocation based on the reliability of satellite model observations and chassis calculations, ultimately outputting high-precision positioning results under different operating conditions and signal fluctuations.

[0026] Preferably, in step S2, before performing chassis motion pattern recognition, the following is also included:

[0027] When an off-center load is detected in the vehicle, the independent single-wheel loads of the left and right wheels are calculated according to the principle of vehicle force balance. Based on the preset mapping relationship between tire load and wheel diameter, the left wheel diameter compensation parameters and the right wheel diameter compensation parameters are generated to correct the wheel speed sensor data.

[0028] In step S2, spatial reference analysis is performed based on satellite positioning data, and the absolute coordinate reference of the vehicle body is output, including:

[0029] Obtain the baseline azimuth of the current reference source. If the baseline vector corresponding to the baseline azimuth points in the opposite direction to the vehicle's direction of travel, add 180 degrees compensation to the baseline azimuth to obtain the vehicle's global heading angle. Otherwise, use the baseline azimuth directly as the global heading angle.

[0030] Based on the absolute coordinates of the current reference source, the global heading angle, and the pre-calibrated longitudinal and lateral distance offsets of each corner point of the vehicle body relative to the current reference source, the coordinates of each corner point of the vehicle body are calculated by coordinate rotation projection as reference position coordinates.

[0031] By calculating the independent loads of the left and right wheels separately and generating corresponding independent wheel diameter compensation parameters, the differential effect caused by the different compression of the tires on both sides can be precisely corrected. This effectively suppresses the integral drift of the heading angle during straight-line driving and improves the accuracy of mileage estimation. To address the situation where the baseline vector of a diagonally arranged antenna may be opposite to the vehicle's direction of travel, a 180-degree phase compensation is applied to the baseline azimuth angle, achieving global uniformity of the heading angle. Combined with pre-calibrated offsets at each corner of the vehicle body, the coordinates of a single antenna are expanded into real-time coordinates of the four corners of the vehicle body through coordinate rotation projection, providing accurate geometric boundary references for collision avoidance and precise docking in confined spaces.

[0032] Preferably, chassis motion pattern recognition includes:

[0033] An equivalent kinematic model of a multi-axle wheel set is established, which maps the multi-axle, multi-drive physical configuration of the vehicle to a test wheel set with four distributed node positions.

[0034] The actual wheel speed data output by the wheel speed meters installed on the four test wheel sets are acquired and defined as the first wheel speed, the second wheel speed, the third wheel speed, and the fourth wheel speed in sequence; at the same time, the Z-axis angular velocity, X-axis acceleration, and Y-axis acceleration in the inertial measurement data are acquired synchronously.

[0035] Based on the numerical matching relationship of the four wheel speeds, combined with the dynamic characteristics of Z-axis angular velocity and acceleration, the current motion mode of the chassis is identified and output. The judgment logic is configured as follows:

[0036] If the speed of the first round is equal to the speed of the fourth round, the speed of the second round is equal to the speed of the third round, and the speed of the first round is not equal to the speed of the second round, then it is identified as the figure-eight pattern.

[0037] If the speed of the first wheel is not equal to the speed of the second wheel, and the speed of the third wheel is equal to the speed of the fourth wheel, then it is identified as car mode;

[0038] If the speeds of the first to fourth wheels are all equal, and the vertical axis angular velocity is not zero, while the X-axis acceleration and Y-axis acceleration are both zero, then it is identified as a stationary turning mode.

[0039] If the speeds of the first to fourth wheels are all equal, and the vertical axis angular velocity is zero, and at least one of the X-axis acceleration and Y-axis acceleration is not zero, then it is identified as a translational motion mode.

[0040] By establishing an equivalent kinematic model of a multi-axis wheel set, the complex physical configuration is simplified into four test wheel sets. By utilizing the combined characteristics of wheel speed and inertial measurement data, a clear logic for numerical matching and dynamic characteristic judgment is established, which can accurately distinguish between figure-eight mode, car mode, stationary turning mode, and translational mode, providing an accurate mode premise for subsequent correct kinematic solution.

[0041] As a preferred approach, the steps for correct chassis kinematics solution include:

[0042] Obtain the vehicle's physical dimensions, including at least the lateral track width S of the test wheelset and the longitudinal dimension L related to the axis of rotation, where S and L are both in meters.

[0043] When identified as a figure-eight pattern or a car pattern:

[0044] Based on the actual wheel speed ratio between the inner and outer wheels, a nonlinear geometric equation is established that includes the lateral wheelbase S, longitudinal dimension L, and rotation center offset distance d.

[0045]

[0046] Among them, V 外 V represents the wheel speed of the outer wheel. 内 Indicates the wheel speed of the inner wheel;

[0047] Solve the nonlinear geometric equation to obtain the offset distance of the current rotation center. ;

[0048] Based on the trigonometric relationship between the offset distance d and the longitudinal dimension L, the virtual steering angle of the inner wheel is calculated. Virtual steering angle of the outer wheel And calculate the current centerline velocity v of the vehicle body by combining the offset distance d;

[0049] When identified as a stationary turning pattern:

[0050] The vehicle's current centerline velocity v is determined to be zero; the Z-axis angular velocity ω from the inertial measurement data is directly extracted as the heading update parameter.

[0051] When identified as translational mode:

[0052] The vehicle's current Z-axis angular velocity ω is determined to be zero; the average wheel speed of multiple wheels is extracted as the vehicle's current centerline velocity v, and the X-axis acceleration and Y-axis acceleration in the inertial measurement data are integrated to calculate the translational direction angle;

[0053] The centerline velocity obtained based on the above-mentioned calculation methods And the Z-axis angular velocity in inertial measurement data According to the following time elements Kinematic model updates chassis pose:

[0054]

[0055]

[0056]

[0057] in, Let t be the heading angle, X-axis coordinate, and Y-axis coordinate of the center of the vehicle chassis at time t; and Let X and Y be the coordinates of the center of the vehicle chassis at time t-1.

[0058] Under conditions where only wheel speeds are available but no physical steering angle sensors are available, a nonlinear equation is established using wheel speed ratios and geometric relationships to inversely deduce the rotation center offset distance. This allows for the calculation of the virtual steering angles of each equivalent wheel and the linear velocity of the vehicle's centerline, achieving an effective soft substitute for the missing hardware steering angle observations. Differentiated calculation strategies are employed based on different motion modes to ensure the accuracy of the forward kinematics solutions for various driving conditions.

[0059] Preferably, the operating logic of the two-layer filtering architecture in step S3 is as follows:

[0060] The locally extended Kalman filter node receives the linear velocity and angular velocity output from the forward kinematics solution of the chassis as wheel odometer observations, and extracts the Z-axis angular velocity from the inertial measurement data as an inertial observation; the locally extended Kalman filter node shields the linear acceleration observations in the inertial measurement data to suppress integral drift and fuses the output of the local odometer.

[0061] The global extended Kalman filter node uses local odometry as the state prior and introduces satellite positioning data after coordinate system transformation as the absolute position observation. The long-term drift of local odometry is eliminated by fusion calculation again, and the global corrected positioning result is output.

[0062] Both the local extended Kalman filter node and the global extended Kalman filter node are connected to an online adaptive regulator of the covariance matrix. This regulator is configured to dynamically adjust the diagonal components of the observation covariance matrix and the off-diagonal components that characterize the coupling noise between sensors online based on the real-time status of the sensors.

[0063] Through a two-layer decoupled extended Kalman filter architecture, local nodes shield linear acceleration observations susceptible to vibration interference to suppress cumulative integral drift, thereby prioritizing the processing of chassis and inertial measurement data and fusing outputs high-frequency local odometry with stable short-term motion states. Global nodes use the local odometry as a state prior and introduce satellite positioning data as absolute position observations, eliminating long-term drift of the local odometry through further fusion calculation. An online adaptive covariance matrix adjuster simultaneously adjusts the diagonal and off-diagonal components, achieving dynamic optimization matching of the fusion weights under different operating conditions.

[0064] Preferably, the dynamic adjustment logic executed by the online adaptive regulator of the covariance matrix includes:

[0065] When the satellite positioning solution is detected to be a fixed solution, an initial covariance value not exceeding a preset value is assigned to the absolute position observation; when the satellite positioning solution is a floating-point solution or a single-point solution, the diagonal covariance component of the absolute position observation of the global extended Kalman filter node is amplified by a preset multiple.

[0066] When abnormal jumps in chassis wheel speed data or deviations in inertial measurement data variance from the normal range are detected, the diagonal covariance component of the corresponding sensor is amplified synchronously.

[0067] When the satellite signal is below the set value or when multiple sensors are coupled, the off-diagonal components in the observation covariance matrix are adjusted online to dynamically adjust the error correlation weight between position observation and local odometry observation.

[0068] By actively reducing weights based on satellite signal status, the global extended Kalman filter node rapidly eliminates or reduces its dependence on low-quality satellite data when the signal degrades. Through auxiliary adjustment based on local sensor operating conditions, the diagonal covariance component of the corresponding sensor is synchronously amplified when data anomalies occur, preventing local odometer contamination by transient noise. Adaptive compensation of off-diagonal components absorbs systemic coupling noise under weak signal or multi-sensor coupling conditions. The synergistic effect of these multiple adjustment strategies ensures continuous convergence and positioning stability of the fusion filter under various harsh conditions.

[0069] Beneficial effects: By designing the wheel speed sensor with a structure where the rotating and measuring parts are relatively separated, cable entanglement and slip ring wear problems are avoided from the physical source; high-precision kinematic calculation of multi-axle chassis is achieved under the limited condition of missing physical steering angle through multi-motion mode recognition and nonlinear forward solving algorithms; the robustness of signal acquisition and positioning continuity in weak GNSS environment is significantly enhanced through the diagonal arrangement of dual antennas and a heading compensation mechanism; in the fusion positioning process, a two-layer extended Kalman filter architecture with local and global decoupling is adopted, locally digesting high-frequency chassis and inertial measurement data, and globally using absolute position to eliminate cumulative drift, and introducing an online covariance adjuster to dynamically adjust the fusion weights and filtering parameters according to signal attenuation and motion state. The above technical means work together to effectively overcome the problems of positioning error accumulation and divergence caused by tire deformation, model complexity, signal obstruction and solution degradation in ship segmented transport vehicles under heavy load, no steering angle observation and weak GNSS environment, and achieve reliable and accurate positioning under complex working conditions. Attached Figure Description

[0070] Figure 1 A schematic diagram of the sensor deployment and installation structure for a ship section carrier positioning system.

[0071] Figure 2 This is a schematic diagram of the installation and internal structure of a gravity-anchored independent wheel speed gauge assembly.

[0072] Figure 3 System hardware connection topology diagram;

[0073] Figure 4 The flowchart shows a two-layer adaptive localization algorithm based on multi-source fusion.

[0074] Figure 5 This is a logic block diagram for load-based wheel diameter dynamic compensation.

[0075] Figure 6 A schematic diagram of equivalent mapping of chassis wheel assembly and feature recognition of multiple motion modes;

[0076] Figure 7 A geometric model of the chassis kinematics forward solution based on motion patterns;

[0077] Figure 8 Logic block diagram for RTK benchmark selection and vehicle body four-corner coordinate calculation. Detailed Implementation

[0078] The ship segment carrier combined positioning system based on multi-source sensor IoT acquisition, as described in this invention, includes an IoT sensing system and an adaptive fusion positioning system. The IoT sensing system is used to collect vehicle's own status and environmental perception data. The adaptive fusion positioning system is communicatively connected to the IoT sensing system, receives the perception data, performs fusion positioning calculation, and outputs the vehicle's real-time pose.

[0079] like Figure 1 As shown, the hardware components of the IoT sensing system mainly include wheel speed sensors, RTK positioning components, inertial measurement units (IMUs), and lidar. The wheel speed sensors are installed on the outside of the vehicle's wheel hubs to provide wheel speed pulse signals. The RTK positioning component 1 includes two sets of antennas arranged diagonally on the vehicle platform. Each set of antennas includes a main antenna and a secondary antenna. The same set of antennas is positioned on the same side of the vehicle to form a first baseline vector pointing towards the rear of the vehicle. The main and secondary antennas of different sets are positioned diagonally on the vehicle platform. For example, the first antenna set is located at the rear of the vehicle, and the second antenna set is located at the front of the vehicle. If the main antenna of the first antenna set is located at the left rear corner of the vehicle, its secondary antenna is located at the right rear corner. Similarly, the main antenna of the second antenna set is located at the right front corner of the vehicle, and its secondary antenna is located at the left front corner. The RTK positioning component 1 is integrated into a common frame structure at the four corners of the vehicle. This common frame structure includes a fixed base 1-1 and a bracket slot 1-2 connected to it. The RTK positioning component 1 and the ultrasonic sensor assembly 2 are connected to the bracket slot 1-2.

[0080] The inertial measurement unit is installed at the physical center of gravity of the vehicle platform, and is covered by a protective housing. It is used to output raw data of angular velocity and acceleration.

[0081] The IoT sensing system also includes a control cabinet component, a sensing module, and a positioning module.

[0082] The control cabinet component 3 is located on the side of the vehicle platform. Its antenna 3-1 is externally mounted, and the wiring harness 3-2 is connected to the cabinet through the wiring hole. The core industrial control computer inside serves as a data aggregation center, performing data cleaning and alignment logic for timestamp synchronization and format parsing of multi-source heterogeneous data; state calculation and fusion logic for calculating the vehicle's real-time pose, task progress, and equipment health status based on the parsed data; and safety decision-making logic for real-time monitoring of data boundaries and triggering alarms and cutting off the power to the actuators in case of abnormal intrusion, fault, or loss of positioning.

[0083] The perception module is used for environmental and cargo status monitoring, including an ultrasonic sensor assembly 2, a lidar assembly, and a vision sensor assembly 4. The ultrasonic sensor assembly 2 is distributed within the lifting mechanism to provide feedback on lifting height and balance. The lidar assembly includes a 2D lidar 5 and a 3D lidar 6. Two 2D lidars are provided, embedded in openings at the center of the front and rear ends of the vehicle platform, respectively. They are maintained at a preset tilt angle relative to the horizontal plane (e.g., tilted 30 degrees inwards towards the vehicle body) by a bracket structure, for coordinated front-to-rear scanning and collective coverage of the transport space above the vehicle platform. The 3D lidar 6 is installed at the corner edges of the vehicle platform and below the platform plane to obtain an approximately 270-degree field of view for obstruction detection and avoidance. Its fixing structure includes a bracket base welded to the vehicle body and a detachable fixing bracket. The vision sensor assembly 4 is fixed at the center of the front and rear ends of the vehicle platform for visual tag reading.

[0084] The positioning module is used to provide full-scene location services. Based on the RTK positioning component, wheel speedometer, 3D LiDAR component, and IMU inertial navigation component 7 installed at the physical center of gravity of the vehicle platform and protected by an outer shell, it transmits coordinates, pulse counts, point clouds and raw inertial navigation data to the core industrial control computer.

[0085] like Figure 2As shown, the wheel speed sensor 8 adopts a freestanding structure based on gravity anchoring. The wheel speed sensor 8 includes a rotating base 8-2 fixed to the wheel hub and equipped with a sensing gear ring 8-1, and a suspended measuring housing 8-3 mounted on the rotating base 8-2 via bearings. The suspended measuring housing 8-3 is equipped with a sensor 8-4 that works with the sensing gear ring 8-1 to measure speed; specifically, it is a Hall effect sensor. Its probe faces the sensing gear ring 8-1 and is used to sense changes in the magnetic field generated by the rotation of the gear ring 8-1 and output pulse signals. The bottom of the suspended measuring housing 8-3 is equipped with a gravity anchoring mechanism 8-5 to maintain a fixed attitude. The gravity anchoring mechanism 8-5 is an eccentric counterweight whose center of mass is located vertically below the bearing axis, and the restoring torque generated by its gravity is greater than the maximum frictional torque of the bearing at its operating speed. Through the synergistic design of a large restoring torque and a low friction base, the suspended measuring housing 8-3 maintains a stable attitude relative to the ground under vehicle driving and vibration conditions, and does not rotate with the wheel.

[0086] The suspended measuring housing 8-3 has a flange on its edge, and the rotating base 8-2 has a groove on its outer periphery. The flange and groove cooperate to form a non-contact labyrinth seal structure, which can prevent external dust from entering without introducing contact friction. The bearing is a deep groove ball bearing, equipped with non-contact metal dust covers on both sides, filled with low-viscosity grease to reduce rotational friction torque. The synergistic design of the non-contact labyrinth seal structure and the low-friction bearing effectively prevents the intrusion of external contaminants and minimizes the system's frictional resistance, assisting the gravity anchoring mechanism to maintain the housing's attitude more reliably.

[0087] like Figure 3 As shown, the core of the adaptive fusion positioning system is the on-board industrial control computer. Various sensors connect to the industrial control computer through multiple industrial communication interfaces: the inertial measurement unit outputs raw angular velocity and acceleration data via an RS485 serial interface; the wheel speed meter outputs pulse count and rotation direction signals via a high-speed pulse interface; the RTK positioning component outputs latitude and longitude coordinates, baseline azimuth, and calculation status indicators in NMEA-0183 format via an RS232 serial port; the 3D LiDAR outputs 3D point cloud data via an Ethernet interface using the UDP protocol; and the 2D LiDAR outputs laser distance and reflection intensity data via an Ethernet interface using the TCP / IP protocol. The industrial control computer, acting as the data aggregation center, performs timestamp synchronization and format parsing of the multi-source heterogeneous data, runs the fusion positioning algorithm, and outputs the vehicle's real-time pose.

[0088] The present invention describes a method for combined positioning of ship segmented transport vehicles based on multi-source sensor IoT acquisition. This method utilizes a two-layer adaptive error state extended Kalman filter framework, and the specific execution flow is combined with… Figures 4 to 8 Please provide an explanation.

[0089] Step S1: Multi-source data physical layer analysis

[0090] The raw physical pose information required for positioning is extracted from the IoT sensing system. This raw information includes at least wheel speed data from the wheel speedometer, inertial measurement data from the inertial measurement unit (IMU), and satellite positioning data from the two antennas in the RTK positioning component. The raw gyroscope and accelerometer values ​​output by the IMU are converted into physical quantities using a scaling factor, and coordinate system alignment is achieved through an installation error matrix. The angular velocity and acceleration are integrated into angle and velocity increments using a bisample algorithm. During state recursion, the raw data is pre-compensated using the zero bias and scaling factor estimated in the previous time step; the quaternion attitude is updated through a conic effect compensation term, and the velocity is updated through paddling and rotation effect compensation terms, resulting in a high-frequency predicted state containing position, velocity, and attitude.

[0091] Step S2: Dual-branch feature calculation

[0092] Step S2 performs dual-branch feature calculation on the original information, including the first branch and the second branch.

[0093] The first branch, based on inertial measurement data and wheel speed data, performs chassis motion pattern recognition and forward kinematics of the chassis, and outputs the relative motion state of the chassis.

[0094] Combination Figure 5 The diagram shown illustrates the load-based wheel diameter dynamic compensation logic. Before identifying the chassis motion pattern, the system first performs load-based wheel diameter correction on the wheel speed sensor data. The system has a pre-defined mapping relationship between tire load and wheel diameter, which characterizes the physical property that the actual rolling wheel diameter decreases as the single-wheel load increases. The specific operation for obtaining this mapping relationship is as follows: On a flat test road with high-quality RTK signals, the vehicle is driven under unloaded, half-loaded, and fully loaded conditions, and the stable wheel diameter scaling factor data after the filtering algorithm converges is recorded. Taking a standard carrier vehicle as an example, under unloaded conditions, the single wheel bears a load of approximately 2.5 tons, and the physical wheel diameter does not change significantly; the baseline scaling factor is calibrated to 0.00. Under half-loaded conditions, the single wheel load is approximately 6.0 tons, and the radial deformation of the tire causes a decrease in the rolling radius of approximately 2%, corresponding to a scaling factor convergence to -0.02. Under fully loaded conditions, the single wheel load is approximately 10.0 tons, and the rolling radius decreases by approximately 4.5%, corresponding to a scaling factor reaching -0.045. Based on the above scattered dataset, the system uses polynomial curve fitting (such as a first or second-order function) or constructs an interpolation lookup table to generate a mathematical mapping function between continuous single-wheel load and wheel diameter attenuation ratio factor (for example, fitting to find that for every 1 ton increase in effective load, the wheel diameter decreases by about 0.4%).

[0095] During the actual online operation of the system, this mathematical model is deeply integrated into the underlying algorithm logic, running in real time as a feedforward intervention mechanism. The system sets global variables to record the loading status of the previous data frame and frequently listens for Boolean value transitions from the chassis's "loading status sensor." When the system detects a sudden change in loading status from "empty" to "loaded," the feedforward compensation mechanism is immediately triggered. The system first reads the actual weight data of the currently loaded segment of cargo, adds it to the known vehicle weight (e.g., 30 tons), and divides it by the total number of tires fixed to the chassis. Based on the static uniform load assumption, it initially calculates the average load on a single wheel. Subsequently, this load is input into the mapping model to quickly calculate the expected wheel diameter ratio factor under the current working condition. In this stage, the system performs the most critical forced intervention: instead of allowing the Kalman filter (EKF) to converge slowly through observation data, the system directly and forcibly overlays the calculated expected scaling factor onto the corresponding dimension of the EKF state vector, and simultaneously resets the covariance matrix (P matrix) of that dimension to an initial variance with reasonable confidence (e.g., set to 0.01 squared, indicating that the system is confident that the initial prediction error is controlled within 1%). This "direct injection" method significantly shortens the filter convergence time during heavy-load start-up, allowing subsequent RTK or visual observation data to be fine-tuned on this precise benchmark. Conversely, when the system detects a change in state from "loaded" to "unloaded" (i.e., unloading completed), it immediately forces the filter's scaling factor state to reset to 0.0 and assigns a smaller covariance confidence level (e.g., 0.005 squared) to quickly match the elastic recovery of the tire's physical deformation after unloading.

[0096] To further address the extremely harsh operating conditions in actual storage yards, this embodiment introduces two advanced optimization mechanisms—eccentric load independent compensation and Zero-Up-Temperature (ZUPT)—into the basic compensation logic described above. Given that the center of gravity of large ship sections often deviates significantly from the vehicle's geometric center during loading, using only the average load assumption would result in greater tire compression on the heavily loaded side than on the lightly loaded side, leading to a physical yaw torque similar to "track differential speed" when the vehicle is traveling in a straight line. To address this, the system extracts cargo center of gravity offset information by reading differential pressure sensor data from the lifting hydraulic cylinder group. Based on a torque balance model, it decouples the total load and calculates the actual force loads on the left and right wheel arrays separately. Based on this, the system substitutes these values ​​into a mapping model to generate two independent sets of compensated wheel diameter parameters for the left and right sides of the vehicle, thereby eliminating the physical yaw phenomenon caused by eccentric load at its source and significantly suppressing the integral drift of the heading angle in weak GNSS environments. In addition, since the segmented cargo lifting and unloading operations both require the vehicle to remain absolutely stationary, when the system detects a change in loading status and the pulse of the bottom wheel speed meter remains at zero, the system immediately triggers the zero-speed correction window. By forcibly locking the relative speed observation value to an absolute zero value, the system reverses the high-frequency calibration of the random walk and zero bias error inside the IMU, thus preparing the most ideal initial physical state for the vehicle to start and enter the obstructed area.

[0097] After completing high-precision compensation of wheel speed data, combined with Figure 6 The diagram shown illustrates the equivalent mapping of the chassis wheelset and the feature recognition of multiple motion modes. The system performs chassis motion mode recognition. First, an equivalent kinematic model of the multi-axle wheelset is established, mapping the vehicle's multi-axle, multi-drive physical configuration to four test wheelsets at distributed node positions. The actual wheel speed data output from the wheel speed meters mounted on the four test wheelsets are acquired and sequentially defined as the first wheel speed, second wheel speed, third wheel speed, and fourth wheel speed. Simultaneously, the Z-axis angular velocity, X-axis acceleration, and Y-axis acceleration from the inertial measurement data are acquired.

[0098] Based on the numerical matching relationship of the four wheel speeds, combined with the dynamic characteristics of the Z-axis angular velocity and acceleration, the current motion mode of the chassis is identified and output. The judgment logic is as follows: if the first wheel speed is equal to the fourth wheel speed, the second wheel speed is equal to the third wheel speed, and the first wheel speed is not equal to the second wheel speed, then it is identified as a figure-eight mode; if the first wheel speed is not equal to the second wheel speed, and the third wheel speed is equal to the fourth wheel speed, then it is identified as a car mode; if the first to fourth wheel speeds are all equal, and the vertical axis angular velocity is not zero, and the X-axis acceleration and Y-axis acceleration are both zero, then it is identified as a stationary turning mode; if the first to fourth wheel speeds are all equal, and the vertical axis angular velocity is zero, and at least one of the X-axis acceleration and Y-axis acceleration is not zero, then it is identified as a translational mode.

[0099] Combination Figure 7The diagram shown illustrates the forward kinematics geometric model of the chassis based on motion patterns. After motion pattern recognition, the system executes the forward kinematics of the chassis. Since the equivalent test wheelset is not equipped with a physical steering angle sensor, the system combines the preset vehicle lateral track and longitudinal dimensions, uses wheel speed data to inversely deduce the rotation center offset distance, establishes a nonlinear geometric equation to solve for the virtual steering angle of each equivalent wheel and the linear velocity of the vehicle centerline, and uses IMU data to complete position updates, outputting a high-precision chassis relative motion state (including equivalent linear velocity and angular velocity), which is then used as the direct observation input for subsequent local filtering.

[0100] The specific process of the forward kinematics solution of the chassis is as follows: obtain the physical dimension parameters of the vehicle, including at least the lateral track width S of the test wheelset and the longitudinal dimension L related to the axis of rotation.

[0101] When the vehicle is identified as a figure-eight pattern or a car pattern, due to the steering motion of the vehicle body, a nonlinear geometric equation is established based on the actual wheel speed ratio between the inner and outer wheels, which includes the lateral track width S, the longitudinal dimension L, and the rotation center offset distance d:

[0102]

[0103] Among them, V 外 V represents the wheel speed of the outer wheel. 内 Let represent the wheel speed of the inner wheel; d represents the offset distance of the rotation center. Solve the above nonlinear geometric equation to obtain the current offset distance d of the rotation center. This is based on the trigonometric relationship between the offset distance d and the longitudinal dimension L. The virtual steering angle α of the inner wheel and the virtual steering angle β of the outer wheel are calculated respectively, and the current centerline velocity v of the vehicle body is calculated in combination with the offset distance d.

[0104] When the vehicle is identified as in a stationary turning mode, the current centerline velocity v of the vehicle is determined to be zero, and the Z-axis angular velocity ω in the inertial measurement data is directly extracted as the heading update parameter.

[0105] When the vehicle is identified as in translation mode, the current Z-axis angular velocity ω of the vehicle body is determined to be zero. The average wheel speed of multiple wheels is extracted as the current centerline velocity v of the vehicle body. The X-axis acceleration and Y-axis acceleration in the inertial measurement data are integrated to calculate the translation direction angle.

[0106] Based on the centerline velocity v obtained from the above-mentioned modes and the Z-axis angular velocity ω from the inertial measurement data, the chassis pose is updated according to the following kinematic model of time element dt:

[0107]

[0108]

[0109]

[0110] in, Let t be the heading angle, X-axis coordinate, and Y-axis coordinate of the center of the vehicle chassis at time t; and Let dt be the X-axis and Y-axis coordinates of the chassis center at time t-1, and dt be the time element. This outputs the relative motion state of the chassis, including the equivalent linear velocity and angular velocity.

[0111] The second branch performs spatial reference analysis based on satellite positioning data, outputting the absolute coordinates of the vehicle body. Combined with... Figure 8 The diagram shown illustrates the logic block diagram for RTK reference selection and vehicle four-corner coordinate calculation. It acquires satellite positioning data from two sets of antennas in the RTK positioning component. Each set of data includes absolute coordinates, baseline azimuth, and a solution status identifier (fixed solution, floating-point solution, or single-point solution). Following a preset rule that fixed solutions have higher priority than floating-point solutions, and floating-point solutions have higher priority than single-point solutions, the set with the highest solution status priority among the two antenna sets is selected as the current reference source.

[0112] Determine the reference position coordinates based on the current reference source: Obtain the baseline azimuth angle of the current reference source. If the baseline vector points in the opposite direction to the vehicle's forward direction, add 180 degrees compensation to the azimuth angle to obtain the vehicle's global heading angle. Otherwise, use that azimuth angle directly as the vehicle's global heading angle. Based on the absolute coordinates of the main antenna of the current reference source. Global heading angle of the vehicle And the pre-calibrated longitudinal distance offset of each corner of the vehicle body relative to the main antenna. and lateral distance offset Geometric calculations are performed based on a rotational projection model in a two-dimensional coordinate system. The virtual boundary coordinates of the target corner points are calculated. , as reference position coordinates.

[0113] The coordinate rotation projection model is as follows:

[0114]

[0115]

[0116] Based on the calculated coordinates of the four corners of the vehicle body, the system generates a global position reference for subsequent global filtering nodes to perform absolute position observation and updates.

[0117] Step S3: Two-layer adaptive fusion filtering

[0118] Combination Figure 4The flowchart of the dual-layer adaptive localization algorithm based on multi-source fusion is shown. Step S3 constructs a dual-layer filtering architecture containing local extended Kalman filter nodes and global extended Kalman filter nodes. The relative motion state of the chassis and the absolute coordinates of the vehicle body are used as reference inputs to the dual-layer filtering architecture. The local and global nodes are adaptively updated by dynamically adjusting the filtering covariance matrix online, and the local and global nodes are fused and output as local positioning results.

[0119] The specific operating logic of the two-layer filtering architecture is as follows:

[0120] A local extended Kalman filter node is constructed: it receives the linear and angular velocities output from the forward kinematics solution of the chassis as wheel odometer observations, and extracts the Z-axis angular velocity from the inertial measurement data as an inertial observation. A local filter node (local EKF node) shields the linear acceleration observations in the inertial measurement data, which are susceptible to vibration interference, to suppress integral drift, prioritizes the processing of chassis and inertial measurement data, and fuses and outputs a high-frequency local odometer with stable short-term motion.

[0121] Construct a global extended Kalman filter node: with local odometry as the state prior, introduce satellite positioning data after coordinate system transformation as the absolute position observation, and eliminate the long-term drift of local odometry by fusing and solving again, and output the globally corrected positioning result.

[0122] Both the local filter node and the global filter node are connected to an online adaptive regulator of the covariance matrix. This regulator dynamically adjusts the diagonal components of the observation covariance matrix and the off-diagonal components that characterize the coupling noise between sensors online based on the real-time status of the sensors.

[0123] The dynamic adjustment logic executed by the online adaptive covariance matrix regulator is as follows: When the satellite positioning solution is detected to be a fixed solution, a small initial covariance value (not exceeding a set value) is assigned to the absolute position observation; when the satellite positioning solution is a floating-point solution or a single-point solution, the diagonal covariance component of the absolute position observation of the global filter node is amplified by a preset multiple or exponentially to quickly remove or reduce the global filter node's dependence on degraded satellite data. When abnormal jumps in chassis wheel speed data or a brief deviation of inertial measurement data variance from the normal range are detected, the diagonal covariance component of the corresponding sensor is amplified synchronously to prevent local odometers from being contaminated by instantaneous noise. When the satellite signal is below a set value or in a multi-sensor coupling condition, the regulator enables cross-covariance compensation through the interface, allowing the system to modify the off-diagonal components in the observation covariance matrix online to dynamically adjust the error correlation weight between position observations and local odometer observations, absorb systematic coupling noise, and ensure the continuous convergence of the fusion filter.

[0124] The globally extended Kalman filter node performs measurement updates based on the adaptively adjusted parameters, estimates the current 21-dimensional error state vector, feeds back the error state to correct the predicted state, and outputs the corrected high-precision global positioning result for the vehicle. After completing the measurement update, the estimated inertial measurement unit zero-bias error and scaling factor error are accumulated into the corresponding sensor parameters to achieve online calibration, and the error state vector is reset to zero to prepare for the next filtering cycle.

Claims

1. A combined positioning system for ship segmented transport vehicles based on multi-source sensor IoT data acquisition, characterized in that, This includes IoT sensing systems and adaptive fusion positioning systems; The IoT sensing system includes a wheel speedometer, an inertial measurement unit, and an RTK positioning component. The RTK positioning component includes two sets of antennas deployed diagonally on the vehicle platform. The wheel speedometer is configured to have a rotating part that moves with the wheel and a measuring part that maintains an attitude stable relative to the ground. The adaptive fusion positioning system communicates with the IoT sensing system and is configured to perform positioning calculations, including a feature data parsing module and a two-layer adaptive fusion filtering module. The feature data parsing module is used to correct the wheel diameter corresponding to the wheel speed data based on the vehicle load, identify the chassis motion mode by combining inertial measurement data and obtain the relative motion state of the chassis by forward solving, and evaluate the signal quality and optimize the benchmark of satellite positioning data to calculate the coordinates of the four corners of the vehicle body as a global position reference. The dual-layer adaptive fusion filtering module includes a local extended Kalman filter node, a global extended Kalman filter node, and an online adaptive covariance matrix adjuster. The online adaptive covariance matrix adjuster is configured to dynamically adjust the covariance matrix parameters of the two extended Kalman filter nodes based on the evaluation results of satellite signal quality and chassis motion pattern recognition. The local extended Kalman filter node fuses the chassis relative motion state to output a local odometry, and the global extended Kalman filter node fuses the local odometry with the global position reference to output the final positioning result.

2. The system according to claim 1, characterized in that: The two sets of antennas each include a main antenna and a secondary antenna; the antennas of the same set are arranged on the same side of the vehicle body to form a first baseline vector pointing to the rear of the vehicle, and the main antennas and secondary antennas of different sets are arranged at diagonal positions on the vehicle body platform; the RTK positioning components are arranged at the four corners of the vehicle body through an integrated co-frame structure, and the integrated co-frame structure is also equipped with ultrasonic sensor components; the integrated co-frame structure includes a fixed base and a bracket slot connected thereto, and the RTK positioning components and ultrasonic sensor components are connected to the bracket slot.

3. The system according to claim 1, characterized in that: The wheel speed meter includes a rotating base fixed to the wheel hub and equipped with a sensing gear ring, and a suspended measuring housing mounted on the rotating base via a bearing. The suspended measuring housing is equipped with a sensor that cooperates with the sensing gear ring to measure speed, and its bottom is equipped with a gravity anchoring mechanism for maintaining a fixed attitude. The gravity anchoring mechanism is an eccentric counterweight, the center of mass of which is located vertically below the bearing axis, and the restoring torque generated by its gravity is greater than the maximum frictional torque of the bearing at the operating speed. The edge of the suspended measuring housing is provided with a flange, and the outer periphery of the rotating base is provided with a groove. The flange and the groove cooperate to form a non-contact labyrinth seal structure.

4. The system according to claim 1, characterized in that, The IoT sensing system also includes a control cabinet component, a sensing module, and a positioning module; The control cabinet component, located on the vehicle platform as a data aggregation center, is used to perform timestamp synchronization and format parsing of multi-source heterogeneous data, and to calculate the real-time vehicle pose, task progress and equipment health status based on the parsed data. It also monitors data boundaries in real time and triggers alarms and cuts off the power of the actuators when there is abnormal intrusion, failure or loss of positioning. The perception module includes an ultrasonic sensor assembly, a lidar assembly, and a vision sensor assembly. The ultrasonic sensor assembly is distributed in the lifting mechanism and is used to provide feedback on the lifting height and balance status. The lidar assembly includes a 2D lidar embedded in the opening of the vehicle platform and arranged at a preset angle to detect the position of segmented goods in the overhead transport space, and a 3D lidar installed at the corner edge of the vehicle platform and below the platform plane to obtain a 270-degree field of view for obstruction perception and avoidance. The vision sensor assembly is fixed at the center of the front and rear ends of the vehicle platform for visual tag reading. The positioning module is used to transmit coordinates, pulse counts, point clouds, and raw inertial navigation data to the control cabinet components. It includes an RTK positioning component, wheel speedometer, 3D LiDAR, and an inertial measurement unit installed at the physical center of gravity of the vehicle platform.

5. A method for combined positioning of ship segmented transport vehicles based on multi-source sensor IoT data acquisition, characterized in that, include: S1. Extract the raw physical pose information required for positioning from the IoT sensing system. The raw information includes at least the wheel speed data of the wheel speed meter, the inertial measurement data of the inertial measurement unit, and the satellite positioning data of the two antennas in the RTK positioning component. S2. Perform dual-branch feature calculation on the original information. The first branch is based on inertial measurement data and wheel speed data to perform chassis motion pattern recognition and forward kinematics of the chassis, and outputs the relative motion state of the chassis. The second branch is based on satellite positioning data to perform spatial reference analysis and outputs the absolute coordinate reference of the vehicle body. S3. Construct a two-layer filtering architecture that includes local extended Kalman filter nodes and global extended Kalman filter nodes. Input the chassis relative motion state and the vehicle absolute coordinates into the two-layer filtering architecture, and adaptively update the local and global nodes by dynamically adjusting the filtering covariance matrix online, and fuse the output positioning results.

6. The method according to claim 5, characterized in that, In step S2, before performing chassis motion pattern recognition, the following steps are also included: When an off-center load is detected in the vehicle, the independent single-wheel loads of the left and right wheels are calculated according to the principle of vehicle force balance. Based on the preset mapping relationship between tire load and wheel diameter, the left wheel diameter compensation parameters and the right wheel diameter compensation parameters are generated to correct the wheel speed sensor data. In step S2, spatial reference analysis is performed based on satellite positioning data, and the absolute coordinate reference of the vehicle body is output, including: Obtain the baseline azimuth of the current reference source. If the baseline vector corresponding to the baseline azimuth points in the opposite direction to the vehicle's direction of travel, add 180 degrees compensation to the baseline azimuth to obtain the vehicle's global heading angle. Otherwise, use the baseline azimuth directly as the global heading angle. Based on the absolute coordinates of the current reference source, the global heading angle, and the pre-calibrated longitudinal and lateral distance offsets of each corner point of the vehicle body relative to the current reference source, the coordinates of each corner point of the vehicle body are calculated by coordinate rotation projection as reference position coordinates.

7. The method according to claim 5, characterized in that, Chassis motion pattern recognition includes: An equivalent kinematic model of a multi-axle wheel set is established, which maps the multi-axle, multi-drive physical configuration of the vehicle to a test wheel set with four distributed node positions. The actual wheel speed data output by the wheel speed meters installed on the four test wheel sets are acquired and defined as the first wheel speed, the second wheel speed, the third wheel speed, and the fourth wheel speed in sequence; at the same time, the Z-axis angular velocity, X-axis acceleration, and Y-axis acceleration in the inertial measurement data are acquired synchronously. Based on the numerical matching relationship of the four wheel speeds, combined with the dynamic characteristics of Z-axis angular velocity and acceleration, the current motion mode of the chassis is identified and output. The judgment logic is configured as follows: If the speed of the first round is equal to the speed of the fourth round, the speed of the second round is equal to the speed of the third round, and the speed of the first round is not equal to the speed of the second round, then it is identified as the figure-eight pattern. If the speed of the first wheel is not equal to the speed of the second wheel, and the speed of the third wheel is equal to the speed of the fourth wheel, then it is identified as car mode; If the speeds of the first to fourth wheels are all equal, and the vertical axis angular velocity is not zero, while the X-axis acceleration and Y-axis acceleration are both zero, then it is identified as a stationary turning mode. If the speeds of the first to fourth wheels are all equal, and the vertical axis angular velocity is zero, and at least one of the X-axis acceleration and Y-axis acceleration is not zero, then it is identified as a translational motion mode.

8. The method according to claim 7, characterized in that, The steps for correct chassis kinematics analysis include: Obtain the vehicle's physical dimensions, including at least the lateral track width S of the test wheelset and the longitudinal dimension L related to the axis of rotation, where S and L are both in meters. When identified as a figure-eight pattern or a car pattern: Based on the actual wheel speed ratio between the inner and outer wheels, a nonlinear geometric equation is established that includes the lateral wheelbase S, longitudinal dimension L, and rotation center offset distance d. ， Among them, V 外 V represents the wheel speed of the outer wheel. 内 Indicates the wheel speed of the inner wheel; Solve the nonlinear geometric equation to obtain the offset distance of the current rotation center. ; Based on the trigonometric relationship between the offset distance d and the longitudinal dimension L, the virtual steering angle of the inner wheel is calculated. Virtual steering angle of the outer wheel And calculate the current centerline velocity v of the vehicle body by combining the offset distance d; When identified as a stationary turning pattern: The vehicle's current centerline velocity v is determined to be zero; the Z-axis angular velocity ω from the inertial measurement data is directly extracted as the heading update parameter. When identified as translational mode: The vehicle's current Z-axis angular velocity ω is determined to be zero; the average wheel speed of multiple wheels is extracted as the vehicle's current centerline velocity v, and the X-axis acceleration and Y-axis acceleration in the inertial measurement data are integrated to calculate the translational direction angle; The centerline velocity obtained based on the above-mentioned calculation methods And the Z-axis angular velocity in inertial measurement data According to the following time elements Kinematic model updates chassis pose: ， ， ， in, Let t be the heading angle, X-axis coordinate, and Y-axis coordinate of the center of the vehicle chassis at time t; and Let X and Y be the coordinates of the center of the vehicle chassis at time t-1.

9. The method according to claim 5, characterized in that, The operating logic of the two-layer filter architecture in step S3 is as follows: The locally extended Kalman filter node receives the linear velocity and angular velocity output from the forward kinematics solution of the chassis as wheel odometer observations, and extracts the Z-axis angular velocity from the inertial measurement data as an inertial observation; the locally extended Kalman filter node shields the linear acceleration observations in the inertial measurement data to suppress integral drift and fuses the output of the local odometer. The global extended Kalman filter node uses local odometry as the state prior and introduces satellite positioning data after coordinate system transformation as the absolute position observation. The long-term drift of local odometry is eliminated by fusion calculation again, and the global corrected positioning result is output. Both the local extended Kalman filter node and the global extended Kalman filter node are connected to an online adaptive regulator of the covariance matrix. This regulator is configured to dynamically adjust the diagonal components of the observation covariance matrix and the off-diagonal components that characterize the coupling noise between sensors online based on the real-time status of the sensors.

10. The method according to claim 5, characterized in that, The dynamic adjustment logic executed by the covariance matrix online adaptive regulator includes: When the satellite positioning solution is detected to be a fixed solution, an initial covariance value not exceeding a preset value is assigned to the absolute position observation; when the satellite positioning solution is a floating-point solution or a single-point solution, the diagonal covariance component of the absolute position observation of the global extended Kalman filter node is amplified by a preset multiple. When abnormal jumps in chassis wheel speed data or deviations in inertial measurement data variance from the normal range are detected, the diagonal covariance component of the corresponding sensor is amplified synchronously. When the satellite signal is below the set value or when multiple sensors are coupled, the off-diagonal components in the observation covariance matrix are adjusted online to dynamically adjust the error correlation weight between position observation and local odometry observation.

Images