A dynamic zero offset calibration method for robot navigation based on multi-sensor fusion

Abstract

This invention relates to the field of robot navigation technology, specifically to a dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion. The technical solution includes: acquiring IMU measurements and velocity data; determining whether the current operating state is stationary based on the IMU measurements; if so, entering a static calibration mode, continuously acquiring IMU data in the stationary state and processing it to obtain an initial zero bias; if not, calculating the fusion residual value between the current IMU measurement and the multi-sensor fusion estimate; validating the acquired key constraint information by determining whether it exceeds the corresponding threshold range; if not, no correction is performed; if so, progressive zero-bias correction is applied to the IMU measurements to obtain corrected measured values, and subsequent IMU measurements are then used for zero-bias compensation correction. This invention uses a two-level calibration architecture of static calibration and online estimation, combined with a multi-sensor fusion algorithm and a progressive correction strategy with a minimal learning rate, to achieve accurate, real-time, and stable calibration of IMU zero bias.

Description

Technical Field

[0001] This invention relates to the field of robot navigation technology, and specifically to a dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion. Background Technology

[0002] With the rapid rise of intelligent manufacturing, unmanned operation, and smart services, mobile robots have become core equipment for promoting industrial upgrading, improving production efficiency, and reducing labor costs. After a decade of leapfrog development, they have evolved from track-dependent single industrial tools into intelligent agents applicable to all scenarios, achieving an industrial leap from technological follower to global leader.

[0003] Currently, the application scenarios of mobile robots have achieved full coverage, penetrating multiple core areas such as industrial production, commercial services, and special operations, showing a trend of large-scale, diversified, and intelligent development. In the industrial sector, robots are widely used in warehousing and logistics, automobile manufacturing, 3C electronics, and new energy lithium battery industries, undertaking tasks such as material handling, workpiece sorting, and equipment inspection. Among them, AMR robots in the warehousing and logistics field have achieved cluster applications of tens of thousands of units, significantly improving warehousing and sorting efficiency and reducing the intensity of manual labor. In the commercial service sector, robots for hotel delivery, supermarket shopping guides, cleaning and disinfection have gradually become widespread, effectively filling the labor shortage in the service industry and improving the level of service standardization. In the field of special operations, robots for power line inspection, mine exploration, and fire rescue can replace human labor in high-risk and complex environments, avoiding personnel safety risks and improving operational safety and efficiency. In addition, the application of robots in home services, medical assistance, and other fields is also gradually expanding, and market demand continues to be released.

[0004] In mobile robot navigation systems, the inertial measurement unit (IMU) is one of the core sensors. By collecting measurement data from accelerometers and gyroscopes, it calculates the robot's attitude, velocity, and position, providing crucial data support for autonomous navigation. However, due to non-ideal factors such as manufacturing defects, material stress, circuit noise, and thermodynamic inhomogeneities in the packaging structure of internal components (e.g., MEMS gyroscopes and accelerometers), the IMU exhibits significant bias error—that is, the deviation of the sensor's output signal from the ideal zero point when there is no input stimulus.

[0005] More importantly, the IMU bias is not constant but exhibits significant time-varying and environmental dependence. It drifts significantly with the accumulation of operating time and changes in ambient temperature. This drift error is continuously amplified during IMU integration, causing a continuous accumulation of errors in the robot's attitude angle and position calculations. Ultimately, this severely affects navigation accuracy and may even lead to robot navigation failure. For example, if the gyroscope bias is 0.01 rad / s, it only takes 100 seconds to accumulate an angle error of 1 radian (approximately 57.3°), which can have a disastrous impact on robot navigation applications that require high-precision attitude maintenance. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion. By designing a two-stage calibration mechanism, combining static initial calibration with dynamic online correction, and fully utilizing odometry constraints and multi-sensor fusion residual information, dynamic tracking calibration of IMU zero bias is achieved. The dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion includes: S1. Acquire IMU measurements and velocity data.

[0007] S2. Determine whether the current operating state is static based on the IMU measurement value. If yes, determine to enter the static standard mode and execute S3. If no, continue to collect data and execute S4.

[0008] S3. After entering the static calibration mode, continuously collect IMU data in a static state and process it to obtain the initial zero bias B0.

[0009] S4. Calculate the fusion residual value R between the current IMU measurement and the multi-sensor fusion estimate.

[0010] S5. Validate the key constraint information obtained to determine whether it exceeds the corresponding threshold range. If not, do not make any corrections; if so, proceed to S6.

[0011] S6. Perform progressive zero-bias correction on the IMU measurements to obtain corrected measured values.

[0012] Preferred method for determining whether a robot is in a stationary state includes: calculating the acceleration magnitude 'a' of the three-axis accelerometer data and determining whether the acceleration magnitude is within the range of stationary acceleration; simultaneously determining whether the absolute values ​​of the three-axis angular velocities of the gyroscope are all less than a preset angular velocity threshold; if both conditions are met, the robot is determined to be in a stationary state and enters the stationary calibration mode.

[0013] Preferred: acceleration modulus .

[0014] Preferably, the static acceleration range is [0.98g, 1.02g], and the preset angular velocity threshold is 0.01rad / s.

[0015] Preferably, the IMU measurements include: triaxial acceleration data from the accelerometer (a x a y a z ) and the three-axis angular velocity data of the gyroscope (ω x ω y ω z ).

[0016] Preferred method for calculating fusion residual value: M IMU E represents the IMU measurement. f This is the estimated value from multi-sensor fusion.

[0017] Preferred: Multi-sensor fusion estimate E f The methods for obtaining the state vector include: 1. Obtaining the system state vector based on the robot's pose and velocity, and constructing a fused state vector: X=[p, v, ... ] T Where p is the robot's three-dimensional position vector in the world coordinate system, v is the robot's three-dimensional velocity vector in the world coordinate system, and ϕ is the robot's attitude angle, including yaw, pitch, and roll angles; 2. The IMU measurement value after zero-bias compensation at the previous moment is used as input: IMU angular velocity measurement value ω; IMU acceleration measurement value a after zero-bias compensation; The data acquisition time is numbered as i; i=1, 2, ...; Then, inertial integration is performed recursively: Attitude angle at time i: ; The three-dimensional velocity vector at time i: ; The three-dimensional position vector at time i: ; Where T is the IMU sampling time interval; , , This represents the initial state when the filter is started. Based on the attitude angle at time i-1 Construct a 3×3 attitude rotation matrix; simultaneously update the covariance prediction: Where F is the system state transition matrix, F T 3. The system state transition matrix is ​​the transpose of the matrix, and Q is the process noise covariance matrix; 4. The odometer speed v' and displacement increment ΔS are used as observations: Construct the observation equation: Where H is the observation matrix and h is the observation noise; calculate the Kalman gain: H T P is the transpose of the observation matrix; i Let R be the covariance matrix updated at time i, representing the magnitude of the uncertainty in the current state estimate; o The odometer observation noise covariance matrix is ​​given; the prior state is corrected to obtain the posterior state estimate. Update and correct the covariance matrix: 4. The posterior state obtained after the above IMU prediction and odometry update. That is, E f .

[0018] Preferred methods for determining whether the robot's current motion state meets the constraints include: 1. Determining the validity of linear motion constraints, uniform motion constraints, fusion estimation constraints, sensor data validity, time continuity, and motion condition constraints.

[0019] Preferred: The specific zero-bias correction process includes: , where i is the data acquisition time number, i=1, 2, ...; α is the minimum learning rate, and R is the fusion residual value of the IMU measurement, and then it is continuously updated iteratively.

[0020] Preferably, the zero bias correction process further includes: Where i is the data acquisition time number, and I is the current data acquisition time number; i = 1, 2, ..., I, t i ti is the time elapsed since the preset calibration time point; t0 is the standard duration; R i The fusion residual value of the IMU measurement obtained for data numbered i is calculated.

[0021] The technical effects and advantages of this invention: Addressing the core shortcomings of existing IMU zero-bias calibration methods, such as zero-bias drift, complex calibration algorithms, and poor robustness, this invention employs a two-level calibration architecture of "static calibration-online estimation," combined with a multi-sensor fusion algorithm and a gradual correction strategy with an extremely small learning rate, to achieve accurate, real-time, and stable calibration of IMU zero bias. Compared to existing technologies, this invention possesses significant technical advantages and innovations, with specific beneficial effects as follows: 1. A two-level calibration closed-loop system is constructed to overcome the limitations of existing calibration methods. Existing static calibration can only obtain the initial zero bias and cannot cope with zero bias drift caused by temperature and time during operation, easily leading to overcorrection or lag problems. This invention quickly obtains the initial zero bias reference through static calibration, providing a reliable starting point for online estimation. Then, online estimation tracks the zero bias drift in real time, forming a closed-loop logic of "static reference → dynamic correction → iterative update". This not only avoids excessive initial zero bias error but also effectively suppresses zero bias drift caused by environment and time, significantly improving the long-term stability of calibration. Compared with static calibration, navigation accuracy is improved by more than 30%.

[0022] 2. This invention clarifies the definition of the fusion truth value and the fusion process, addressing the shortcomings of existing technologies such as residual ambiguity and non-closed-loop calibration logic. The fused output E... f As a reliable true value for zero-bias calibration, it standardizes the fusion process. By combining the advantages of IMU pre-integration prediction and odometry observation update, it obtains an estimate that is close to the robot's actual motion state. It clarifies the physical meaning of the residual as the error dominated by zero-bias drift, ensuring the pertinence of zero-bias correction and avoiding erroneous correction. Compared with existing calibration methods that do not clearly define the true value for fusion, its robustness is greatly improved and it can effectively resist sensor noise interference.

[0023] 3. Employing a gradual correction strategy with an extremely small learning rate, balancing calibration accuracy and navigation real-time performance. This invention uses a first-order complementary filter combined with an extremely small learning rate (0.0001~0.001), eliminating the need for complex matrix operations and improving computational efficiency by more than 50%. It can slowly correct zero-bias drift without interfering with the high-frequency dynamic signal acquisition of the IMU, meeting the real-time navigation requirements of robots and demonstrating strong engineering feasibility. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion proposed in this invention.

[0025] Figure 2 This is a flowchart illustrating the method for obtaining multi-sensor fusion estimates in a dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion proposed in this invention. Detailed Implementation

[0026] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the invention, and should not be construed as limiting the invention. Rather, embodiments of the invention include all variations, modifications, and equivalents falling within the spirit and scope of the appended claims.

[0027] Example 1 refer to Figure 1 This embodiment proposes a dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion. By designing a two-level calibration mechanism, combining static initial calibration and dynamic online correction, and fully utilizing odometry constraints and multi-sensor fusion residual information, dynamic tracking calibration of IMU zero bias is achieved. The dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion includes: S1. After the robot is powered on, the control IMU enters the initialization state and continuously collects IMU measurement values ​​and velocity data. The robot navigation system of this application may include: an IMU, an odometry unit, a main controller, and a fusion calculation module, etc.; wherein, the IMU is used to collect three-axis acceleration and three-axis angular velocity data of the robot's motion state, and the sampling frequency can be set to 100-200Hz (adapting to the characteristics of conventional MEMS sensors); the odometry unit (which can be a wheeled odometry unit, visual odometry unit, etc.) is used to collect the robot's motion speed and displacement information, providing motion constraints for zero-bias calibration; the main controller is used to execute calibration algorithms and process sensor data; the fusion calculation module is used to fuse IMU and odometry data, output robot navigation information (attitude, position), and generate fusion residuals, providing a basis for online zero-bias estimation. IMU measurement values ​​may include: three-axis acceleration data from the accelerometer (a... x a y a z ) and the three-axis angular velocity data of the gyroscope (ω x ω y ω zFor example, using a wheeled inspection robot as the application platform, the robot navigation system includes: an IMU (MEMS type, with a three-axis accelerometer measurement range of ±16g, a three-axis gyroscope measurement range of ±250rad / s, a sampling frequency of 200Hz, and a built-in temperature sensor), a wheeled odometer (sampling frequency of 10Hz, speed measurement accuracy of ±0.01m / s), a main controller (using STM32H743, main frequency of 480MHz), and a fusion calculation module (achieving multi-sensor fusion based on error state EKF). The robot operates in a factory workshop with an ambient temperature range of 10℃-35℃ and slight ground vibration (noise interference). The navigation task is to inspect along a preset path, requiring continuous operation for 2 hours with an attitude angle error ≤0.5° and a position error ≤0.1m. After the robot is powered on, the main controller initializes the IMU, odometry, and fusion calculation module, clears the cache, and sets relevant parameters: accelerometer modulus judgment range [0.98g, 1.02g], gyroscope angular velocity static threshold 0.01rad / s, static calibration sampling duration 2 seconds, online estimation learning rate α=0.001, sliding window size 50 groups, temperature trigger threshold 5℃, and reasonable range of fusion residual ±0.05g (accelerometer) and ±0.005rad / s (gyroscope). A new temperature-assisted calibration, anomaly handling, and sliding window filtering mechanism are added. When there is a sudden temperature change, the correction speed is adaptively adjusted or static calibration is restarted; when there is abnormal data, the correction is paused, effectively avoiding calibration failure caused by temperature drift and abnormal data.

[0028] S2. Determine whether the current operating state is stationary based on the IMU measurement value. If yes, enter the stationary standard mode and execute S3. If no, continue collecting data and execute S4. The method for determining whether it is stationary may include: calculating the acceleration modulus 'a' of the accelerometer's three-axis data, i.e., the acceleration modulus. To determine whether the magnitude of acceleration is within the range of rest acceleration, the range of rest acceleration can be [0.98g, 1.02g] (where g is the acceleration due to gravity, with a value of 9.8 m / s²). 2 Simultaneously, it checks whether the absolute values ​​of the three-axis angular velocities of the gyroscope are all less than a preset angular velocity threshold (e.g., 0.01 rad / s). If both conditions are met, the robot is determined to be stationary and enters a stationary calibration mode. If not, data is continuously collected until a stationary state is detected (the maximum detection time does not exceed 5 seconds to avoid excessive initialization time). For example, the main controller controls the IMU to collect three-axis acceleration and three-axis angular velocity data for 1 second, calculating the accelerometer modulus value to be 9.78 m / s². 2(i.e., 0.998g), the absolute values ​​of the angular velocities of the three axes of the gyroscope are all 0.003rad / s, which meets the conditions for judging the stationary state, and enters the stationary calibration mode. The stationary state here can include the linear uniform motion state, but linear uniform motion when the IMU measurement value judges it to be stationary is very rare, and will not be elaborated here.

[0029] S3. After entering static calibration mode, continuously acquire IMU data in a static state and process it to obtain the initial zero bias B0. Processing may include removing outliers, calculating the average values ​​of the accelerometer's three-axis zero bias and the gyroscope's three-axis zero bias as the initial zero bias B0, and storing the initial zero bias. Here, B can be a 6-dimensional vector, i.e. B ax B ay B az These are the three-axis zero bias of the accelerometer, B ωx B ωy B ωz These represent the three-axis zero bias of the gyroscope. For example, after continuously acquiring 400 sets of IMU data for 2 seconds and discarding invalid data, the average three-axis zero bias of the accelerometer is calculated as: B ax =0.02m / s 2 B ay =0.018m / s 2 B az =0.022m / s 2 ; Gyroscope three-axis zero-bias average value: B ωx =0.002rad / s, B ωy =0.0018rad / s, B ωz =0.0022 rad / s; that is, the initial zero bias B0 = [0.02, 0.018, 0.022, 0.002, 0.0018, 0.0022]. Of course, it can also be the median or mode of the three-axis zero bias of the accelerometer and the three-axis zero bias of the gyroscope, etc., which will not be elaborated here.

[0030] S4. Calculate the fusion residual value R between the current IMU measurement and the multi-sensor fusion estimate. The fusion residual value can be calculated using the following methods: M IMU E represents the IMU measurement. f This is the multi-sensor fusion estimate; the essence of this fusion residual value is the difference between the IMU measurement and the optimal fusion estimate, mainly caused by IMU zero-bias drift, and can serve as the core basis for zero-bias correction. Multi-sensor fusion estimate E f An odometer or IMU can be directly connected. However, using an odometer may result in inaccurate calculations, slippage, low frequency, and lag. Using an IMU may also result in inaccurate data and inherent zero-bias drift. (Reference) Figure 2 Multi-sensor fusion estimate E f The methods for obtaining the state vector may include: 1. Obtaining the system state vector based on the robot's pose and velocity, and constructing a fused state vector: X=[p, v, ϕ] T Where p is the robot's 3D position vector in the world coordinate system, v is the robot's 3D velocity vector in the world coordinate system, and ϕ is the robot's attitude angle, including yaw, pitch, and roll angles. The processed fused optimal state As E f 2. Using the IMU measurement value after zero-bias compensation at the previous moment as input: IMU angular velocity measurement ω (after zero-bias compensation); IMU acceleration measurement a after zero-bias compensation; number the data acquisition time as i; i=1,2,…; then perform inertial integration recursion: Attitude angle at time i: ; The three-dimensional velocity vector at time i: ; The three-dimensional position vector at time i: ; Where T is the IMU sampling time interval (integration step size), in seconds, which is determined by the IMU sampling frequency, and typically ranges from 0.005s to 0.01s; , , The initial state during filter startup is given directly at the power-on / startup time and will not be elaborated here; Based on the attitude angle at time i-1 A 3×3 attitude rotation matrix is ​​constructed to transform the acceleration from the IMU body coordinate system to the world coordinate system. Simultaneously, the covariance prediction is updated. Where F is the system state transition matrix, F T Let Q be the transpose of the system state transition matrix, and let Q be the process noise covariance matrix. 3. Use the odometer speed v' and displacement increment ΔS as observations: Construct the observation equation: Where H is the observation matrix and h is the observation noise; calculate the Kalman gain: H T P is the transpose of the observation matrix. i Let the updated covariance matrix (posterior covariance) at time i represent the magnitude of the uncertainty in the current state estimate. R oThis is the odometer observation noise covariance matrix, with dimensions consistent with the odometer observations (typically 4×4 or 6×6), and units representing the variance of the corresponding observations. It indicates how "unreliable / noisy" the odometer measurement data itself is. It can be preset offline based on odometer accuracy indicators and experimental calibration, directly given at robot startup and remaining unchanged during operation; its value is typically 10. -4 ~10 -2 Correcting the prior state yields the posterior state estimate: Update and correct the covariance matrix: , where N is the identity matrix and covariance matrix of the same dimension (e.g., 9×9 or 15×15), with 1s on the diagonal and 0s elsewhere. 4. The posterior state obtained after the above IMU prediction and odometry update. That is, E f , including fusion location Fusion speed and fusion posture The specific numerical values ​​will not be elaborated here.

[0031] S5. To avoid excessive zero-bias correction due to invalid constraints, the validity of the acquired key constraint information is judged to determine whether it exceeds the corresponding threshold range. If not, no correction is performed; if so, proceed to S6. Key constraint information consists of combinations of IMU measurements and velocity data, which will not be elaborated here. For odometry motion constraints (key constraint information validity judgment), it is determined whether the robot's current motion state meets the constraint conditions. For example, 1. Validity judgment conditions for linear motion constraints: 1) The rate of change of the heading angle output by the odometry is less than the angle threshold (e.g., ≤1° / s); 2) The speed difference between the left and right wheels is less than the speed threshold; 3) The fused estimated angular velocity is stably close to 0 for a preset number of consecutive points; If satisfied: the constraint is deemed valid, and the zero bias can be corrected using the true value of the angular velocity = 0; If not satisfied: turning, slipping, or bumping, the constraint is invalid and no correction is performed. Conditions for judging the effectiveness of uniform motion constraints: 1) Odometer velocity change ≤ 0.02 m / s; 2) IMU measured motion acceleration close to 0; If satisfied: the robot can be considered to have no acceleration or deceleration, used for accelerometer zero bias correction; If not satisfied: the constraint is invalid during start-up, braking, and acceleration. Conditions for judging the effectiveness of static constraints: 1) Acceleration magnitude ∈ [0.98g, 1.02g]; 2) All three-axis angular velocities are less than the static threshold (e.g., ≤ 0.01 rad / s); 3) Odometer velocity ≈ 0; If satisfied: enter strong correction mode; If not satisfied: static calibration is not triggered. 2. Judgment of the effectiveness of fusion estimation constraints: fusion covariance P is less than the set threshold, filtering does not diverge, and no jump occurs; Judgment of the rationality of fusion residuals: judge whether the fusion residuals are in the normal range: angular velocity residual ≤ 0.01 rad / s, acceleration residual ≤ 0.05g, fusion residual too large: indicates that it is affected by noise / impact / slippage, and the constraint is invalid; stable residuals: mainly caused by zero bias drift, the constraint is effective. Sensor data validity assessment: IMU data validity: acceleration and angular velocity are within their range, with no sudden pulses or obvious spike noise. Temperature is within the normal operating range (e.g., -20℃ to 60℃). Odometer data validity: wheel speedometer has no missed steps or overflow, the speed sign is consistent with the direction of movement, and there is no obvious slippage (speed sudden change but attitude remains unchanged). Time continuity assessment: constraints are continuously effective for ≥ a certain number of frames (e.g., ≥ 20 frames) to avoid single noise-induced false triggering of corrections. Motion condition constraint assessment (avoiding dynamic over-correction): Motion amplitude assessment: constraints are invalid for rapid acceleration, sharp turns, and bumpy jumps. Constraints are effective for smooth straight driving and low-speed cruising. Correction frequency limitation: even if constraints are effective, the zero-bias correction frequency is limited to ensure only slow drift and no rapid changes with dynamic motion. For fusion residuals, the absolute value of the residuals is assessed to see if it is within a reasonable range (excluding abnormal residuals caused by sensor noise and sudden motion changes); other assessment methods are not excluded, but will not be elaborated here.

[0032] S6. Perform progressive bias correction on the IMU measurements to obtain corrected measured values. Subsequent IMU measurements are then corrected using this method. The specific bias correction process is as follows: Using a minimal learning rate α (α ranges from 0.0001 to 0.001), the IMU bias is slowly updated using a first-order progressive correction formula: , where i is the data acquisition time number, i=1,2,…; α is the minimum learning rate, and R is the fusion residual value of the IMU measurement. Then, the process is continuously iterated and updated: after each frame of IMU data acquisition, steps 1-5 are repeated to continuously iterate and correct the zero bias, thereby achieving real-time suppression of zero bias drift without interfering with the high-frequency dynamic response of the IMU.

[0033] The zero-bias correction process can also be as follows: Where i is the data acquisition time number, and I is the current data acquisition time number, i.e., the total number of data acquisitions. i = 1, 2, ..., I, t i Let t0 be the time interval from time i to a preset calibration time point. This allows for the calibration of a preset time point, such as calibrating and evaluating data within 10 minutes of the current time. t0 is the standard duration, the value of which can be determined based on actual conditions, for example, 1 minute. However, this is just a simple example and may not be universally applicable. i The fusion residual value of the IMU measurement obtained for data numbered i is calculated. This method takes into account the timeliness of zero-bias correction, allows evaluation of data over a preset time period, and weakens long-term data without completely discarding it. This avoids the problem of a one-size-fits-all approach to data, prevents calculation errors caused by data fluctuations, and improves the accuracy of zero-bias correction.

[0034] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0035] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion, characterized in that, The dynamic zero-bias calibration method for robot navigation based on multi-sensor fusion includes: S1. Acquire IMU measurements and velocity data; S2. Determine whether the current operating state is a static state based on the IMU measurement value. If yes, determine to enter the static standard mode and execute S3. If no, continue to collect data and execute S4. S3. Enter the static calibration mode, continuously collect IMU data in a static state, and process it to obtain the initial zero bias B0; S4. Calculate the fusion residual value R between the current IMU measurement and the multi-sensor fusion estimate; S5. The validity of the acquired key constraint information is judged to determine whether it exceeds the corresponding threshold range. If not, no correction is made; if so, S6 is executed. S6. Perform progressive zero-bias correction on the IMU measurements to obtain corrected measured values.

2. The method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 1, characterized in that, The method for determining a stationary state includes: calculating the acceleration modulus 'a' of the three-axis accelerometer data, and determining whether the acceleration modulus is within the range of stationary acceleration; simultaneously determining whether the absolute values ​​of the three-axis angular velocities of the gyroscope are all less than a preset angular velocity threshold; if both conditions are met, the robot is determined to be in a stationary state and enters the stationary calibration mode.

3. The method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 2, characterized in that, The acceleration modulus .

4. The method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 2, characterized in that, The static acceleration range is [0.98g, 1.02g], and the preset angular velocity threshold is 0.01rad / s.

5. The method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 1, characterized in that, IMU measurements include: triaxial acceleration data from the accelerometer (a x a y a z ) and the three-axis angular velocity data of the gyroscope (ω x ω y ω z B0 is a 6-dimensional vector, that is... B ax B ay B az These are the three-axis zero bias of the accelerometer, B ωx B ωy B ωz These are the three-axis zero-bias gyroscopes.

6. The method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 1, characterized in that, The method for calculating the fusion residual is as follows: M IMU E represents the IMU measurement. f This is the estimated value from multi-sensor fusion.

7. The method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 6, characterized in that, Multi-sensor fusion estimate E f The methods for obtaining the state vector include:

1. Obtaining the system state vector based on the robot's pose and velocity; 2. Using the IMU measurement value after zero-bias compensation at the previous moment as input, performing inertial integration recursion; 3. Using the odometry velocity v' and displacement increment ΔS as observations; 4. Obtaining the posterior state after the above IMU prediction and odometry update. That is, E f .

8. The method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 7, characterized in that, The fusion state vector is defined as: X = [p, v, ϕ] T , where p is the robot's three-dimensional position vector in the world coordinate system, v is the robot's three-dimensional velocity vector in the world coordinate system, and ϕ is the robot's attitude angle, including yaw, pitch, and roll angles.

9. The method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 1, characterized in that, Determining whether the robot's current motion state meets the constraints includes: judging the validity of linear motion constraints, judging the validity of uniform motion constraints, judging the validity of fusion estimation constraints, judging the validity of sensor data, judging time continuity, and judging motion condition constraints.

10. A method for dynamic zero-bias calibration of robot navigation based on multi-sensor fusion according to claim 6, characterized in that, The zero-bias correction process includes: , where i is the data acquisition time number, i=1, 2, ...; α is the minimum learning rate, and R is the fusion residual value of the IMU measurement, and then it is continuously updated iteratively.

Images