A robot high-precision positioning method and system based on environment perception

Abstract

The application discloses a kind of robot high-precision positioning method and system based on environmental perception, it is related to the field based on environmental perception of robot high-precision positioning, for improving the positioning accuracy and navigation stability of robot, comprising: the difference information between the first pose information and the second pose information is calculated;Obtain the first uncertainty information of inertial recursion module state, and the second uncertainty information of visual observation module, to generate the uncertainty information of difference information;Based on difference information, the uncertainty information of difference information, and the duration that system is in pure inertial recursion mode, it is judged whether the source type of difference information is inertial drift accumulation type;When source type is inertial drift accumulation type, determine the accommodation capacity parameter of the kinematic parameter output by robot chassis;Based on difference information and accommodation capacity parameter, determine the adjustment parameter of the motion state of robot;According to adjustment parameter, form gradual state correction process.

Description

Technical Field

[0001] This invention relates to the field of high-precision robot positioning based on environmental perception, and more particularly to a method and system for high-precision robot positioning based on environmental perception. Background Technology

[0002] In industrial inspection, mobile robots typically employ environmental perception methods such as visual sensors and inertial measurement units (IMUs) for real-time pose estimation to achieve autonomous navigation within complex factory environments. Under normal circumstances, a visual-inertial tightly coupled filter maintains stable positioning results through joint pose optimization. However, when the robot travels to areas with high-contrast lighting, such as below a smoke exhaust skylight on a factory roof, the alternation of direct sunlight and shadows can cause large areas of overexposure and underexposure in the image sequence. At this point, the number of inliers matched by the visual front-end drops sharply, and the system typically determines that the visual observation quality is below a threshold, automatically switching to a degradation-robust mode based on pure inertial recursion.

[0003] During pure inertial recursive degradation, the robot inevitably accumulates position and heading errors. When the robot leaves the skylight area and lighting returns, visual feature points reappear stably. Once the system detects that the visual observation is valid, it switches back to the normal visual-inertial fusion mode. At this point, there will be a significant difference between the initial absolute pose estimate given by the visual odometry and the current inertial recursive prediction, causing the residual to exceed the test threshold set by the filter. Existing conventional switching logic often only focuses on whether the visual observation is continuous and stable, directly changing the filter covariance matrix and state vector structure, injecting the visual residual into the filter with full weight, without considering the impact of accumulated errors during degradation on the state transition.

[0004] This approach interprets the visual residual as a large measurement update, causing the fusion filter to directly output a step-like positioning jump, forcibly pulling the robot's pose to the position given by the vision. This jump causes the navigation module to perceive a sudden change in the robot's pose, leading to the calculation of a large correction command. This results in the robot suddenly turning on smooth ground, ultimately causing inaccurate inspection and docking judgments. Therefore, when the mobile robot recovers its visual-inertial tightly coupled fusion positioning after leaving an area with drastic lighting changes, how to avoid the step conflict between the pose drift accumulated by the degraded inertial recursion and the absolute pose given by the instantaneous visual observation at the fusion filter residual check threshold, and how to prevent the filter from directly outputting a positioning jump that interferes with navigation control accuracy and inspection and docking tasks, is a specific technical problem that urgently needs to be solved in this scenario. Summary of the Invention

[0005] This application discloses a high-precision robot positioning method and system based on environmental perception. It aims to solve the problem that when a mobile robot recovers its visual-inertial tightly coupled fusion positioning in an area with drastic changes in lighting, the pose drift accumulated by the degraded inertial recursion and the absolute pose given by the visual observation at the moment of recovery cause a step conflict at the residual test threshold of the fusion filter, resulting in the filter directly outputting a positioning jump that interferes with the accuracy of navigation and control and the inspection and fixed-point docking tasks.

[0006] In a first aspect, this application discloses a high-precision robot positioning method based on environmental perception, comprising the following steps: The robot acquires the first pose information provided by its visual observation module and the second pose information provided by its inertial recursion module. Calculate the difference between the first pose information and the second pose information; The system acquires the first uncertainty information of the state of the inertial recursion module and the second uncertainty information of the visual observation module, and generates the uncertainty information of the difference information based on the first and second uncertainty information. The first uncertainty information reflects the error range accumulated by the inertial recursion module under no correction. The second uncertainty information reflects the measurement noise level of the visual observation module itself in the current environment. Based on the difference information, the uncertainty information of the difference information, and the duration of the system in the pure inertial recursive mode, determine whether the source type of the difference information is the inertial drift accumulation type. When the source type of the difference information is inertial drift accumulation, determine the load-bearing capacity parameter of the kinematic parameters output by the robot chassis; Based on the difference information and the carrying capacity parameters, the adjustment parameters of the robot's motion state are determined; Based on the adjusted parameters, a gradual state correction process is formed.

[0007] Optionally, the difference between the first pose information and the second pose information is calculated, including: Calculate the 3D translation vector difference between the first pose information and the second pose information, as well as the rotation vector difference obtained through quaternion operations; The three-dimensional translation vector difference and rotation vector difference are combined to form a residual vector, and the residual vector is used as the difference information.

[0008] Optionally, based on the difference information, the uncertainty information of the difference information, and the duration of the system in pure inertial recursion mode, it can be determined whether the source type of the difference information is inertial drift accumulation, including: The Mahalanobis distance is determined based on the difference information and the uncertainty information of the difference information. Under the condition that the first preset condition is met, the source type of the difference information is determined to be the inertial drift accumulation type; the first preset condition is that the Mahalanobis distance is within a preset distance interval, and the ratio of the residual vector to the duration of the system in the pure inertial recursive mode is within a preset ratio range. If the first preset condition is not met, determine that the source type of the difference information is not the inertial drift accumulation type.

[0009] Optionally, based on the difference information and the uncertainty information of the difference information, the Mahalanobis distance is determined, including: Based on the difference information and the uncertainty information of the difference information, the prediction covariance matrix of the residual vector is determined; The Mahalanobis distance is obtained by calculating the transpose of the residual vector, the inverse of the prediction covariance matrix, and the product of the residual vector.

[0010] Optionally, based on the adjusted parameters, a progressive state correction process is formed, including: Based on the second uncertainty information and the adjustment parameters, the adjusted second uncertainty information is generated; Based on the first uncertainty information and the adjusted second uncertainty information, the updated gain parameters are calculated; Based on the updated gain parameters and difference information, a correction amount for the motion state is generated, and the robot's motion state is updated. The robot's motion state is repeatedly updated over multiple processing cycles to form a progressive state correction process.

[0011] Optionally, the method also includes: The updated state sequence during the progressive state correction process is obtained, and kinematic parameters are generated based on the updated state sequence. The kinematic parameters are then verified to meet the preset physical constraints. When kinematic parameters do not conform to physical constraints, the magnitude of motion state updates is limited; When both the difference information and the second uncertainty information meet the preset convergence conditions, the gradual state correction process ends.

[0012] Optionally, based on the second uncertainty information and the adjustment parameters, adjusted second uncertainty information is generated, including: Perform illumination change analysis on local areas of the image output by the visual observation module to identify areas of local illumination change; Evaluate the impact of localized illumination variations on the pose calculation accuracy of the visual observation module; Based on the evaluation results and adjusted parameters, the adjusted second uncertainty information is generated.

[0013] Optionally, the load-bearing capacity parameters of the kinematic parameters output by the robot chassis are determined, including: Obtain the load information of the robot chassis, the friction information between the wheels of the robot chassis and the ground, and the energy state information of the robot chassis battery; Based on load information, friction information, and energy state information, determine the load-bearing capacity parameters.

[0014] Optionally, based on load information, frictional force information, and energy state information, the load-bearing capacity parameters are determined, including: Obtain the first mapping relationship; the first mapping relationship includes the correspondence between different load information, different friction force information, different energy state information, and different load-bearing capacity parameters. Based on the first mapping relationship, load information, friction information, and energy state information, the load-bearing capacity parameters are determined.

[0015] Secondly, this application also discloses a high-precision robot positioning system based on environmental perception, the system comprising: The information acquisition module is used to acquire the first pose information provided by the robot's visual observation module and the second pose information provided by the inertial recursion module; The difference calculation module is used to calculate the difference information between the first pose information and the second pose information. The uncertainty generation module is used to acquire the first uncertainty information of the state of the inertial recursion module and the second uncertainty information of the visual observation module, and generate the uncertainty information of the difference information based on the first uncertainty information and the second uncertainty information; the first uncertainty information reflects the error range accumulated by the inertial recursion module under no correction; the second uncertainty information reflects the measurement noise level of the visual observation module itself in the current environment. The source determination module is used to determine whether the source type of the difference information is the inertial drift accumulation type based on the difference information, the uncertainty information of the difference information, and the duration of the system in the pure inertial recursion mode. The determination module is used to determine the load-bearing capacity parameters of the kinematic parameters output by the robot chassis when the source type of the difference information is inertial drift accumulation. The determination module is also used to determine the adjustment parameters of the robot's motion state based on the difference information and the carrying capacity parameters; The progressive correction module is used to form a progressive state correction process based on the adjusted parameters.

[0016] Beneficial Effects: This application discloses a high-precision robot positioning method based on environmental perception. It acquires first pose information provided by the robot's visual observation module and second pose information provided by its inertial recursion module, and calculates the difference between them. Simultaneously, it acquires uncertainty information from both the inertial recursion module and the visual observation module, and generates uncertainty information for the difference based on this uncertainty information. Based on this, it combines the difference information, the uncertainty information of the difference information, and the duration of the system in pure inertial recursion mode to determine whether the source type of the difference information is inertial drift accumulation. When it is determined to be inertial drift accumulation, it further determines the carrying capacity parameter of the kinematic parameters output by the robot chassis, and determines the adjustment parameters of the robot's motion state based on the difference information and the carrying capacity parameter. Finally, it forms a progressive state correction process based on the adjustment parameters.

[0017] Through the above technical solution, this application effectively solves the problem in the prior art where, after a robot leaves an area with drastic changes in lighting, when restoring visual-inertial tightly coupled fusion localization, the pose drift accumulated by degraded inertial recursion and the absolute pose given by visual observation at the instant of restoration cause a step conflict at the residual test threshold of the fusion filter, leading to a direct output positioning jump from the filter that interferes with navigation control accuracy and inspection and docking tasks. This application, by introducing a judgment on the source type of the difference information and combining it with the robot chassis's load-bearing capacity parameters, achieves progressive correction of the robot's motion state, avoiding the step positioning jump caused by directly injecting the visual residual with full weight into the filter in traditional methods. This progressive correction process makes the robot's pose adjustment smoother and more stable, effectively preventing interference with navigation control, thereby improving the robot's positioning accuracy and navigation stability. Especially in complex lighting environments, it can ensure the accuracy of inspection and docking tasks, demonstrating significant and superior technical effects. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of a high-precision robot positioning method based on environmental perception provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of another high-precision robot positioning method based on environmental perception provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a high-precision robot positioning system based on environmental perception provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] To better understand the technical solution proposed in this application, some key terms involved will be explained first.

[0022] "First pose information" refers to the robot pose data estimated by the robot's visual observation module (e.g., based on visual odometry or visual SLAM system) at the current moment, which typically includes the robot's three-dimensional position and orientation (e.g., represented by rotation matrix or quaternion).

[0023] "Second pose information" refers to the robot pose data predicted at the current moment by the robot's inertial recursion module (e.g., based on the integration of the inertial measurement unit IMU).

[0024] "Difference information" refers to the deviation between the first pose information and the second pose information, reflecting the inconsistency between two different sensors or estimation methods in pose estimation.

[0025] "First uncertainty information" refers to the range of pose estimation error accumulated over time by the inertial recursion module without external correction, and is usually represented by the covariance matrix.

[0026] "Second uncertainty information" refers to the level of measurement noise of the visual observation module under the current environmental conditions, and is usually represented by the covariance matrix.

[0027] "Uncertainty information of difference information" refers to the statistical uncertainty of the difference information itself, which combines the first uncertainty information and the second uncertainty information.

[0028] "Duration of pure inertial recursion mode" refers to the length of time during which the system relies entirely on inertial recursion for pose estimation, without any effective visual observations for correction.

[0029] "Accumulated inertial drift type" refers to the fact that the main source of discrepancy information is the error accumulated due to inertial recursion over a long period of time without correction.

[0030] "Acceptance capability parameters" refer to the robot chassis's ability to smoothly absorb and perform pose adjustments in its current state, such as the chassis's maximum acceleration and angular velocity limits.

[0031] "Adjustment parameters" refer to specific values ​​used to correct the robot's motion state, such as pose correction and velocity correction.

[0032] "Progressive state correction process" refers to correcting the robot's state through a series of small, gradual pose updates, rather than a one-time large jump.

[0033] The following specific embodiments will provide a detailed introduction and explanation of a robot high-precision positioning method based on environmental perception provided in this application.

[0034] Reference Figure 1 This invention provides a high-precision robot positioning method based on environmental perception, comprising the following steps: S1: Obtain the first pose information provided by the robot's visual observation module and the second pose information provided by the inertial recursion module.

[0035] Obtaining the first pose information can be achieved in several ways. For example, the visual observation module can be a visual odometry system based on feature point matching, calculating the robot's relative motion by tracking and matching feature points between consecutive image frames and accumulating them to obtain the global pose. Another approach is for the visual observation module to be a visual odometry based on a direct method, estimating camera motion by minimizing image pixel intensity errors. Furthermore, deep learning-based visual pose estimation methods can also be employed, directly predicting the robot's pose from images using neural networks.

[0036] Acquiring the second pose information is typically accomplished by an inertial recursion module. This module can integrate an inertial measurement unit (IMU) containing accelerometers and gyroscopes. By integrating the measurements from the accelerometers and gyroscopes, the robot's velocity and pose changes can be obtained. For example, quaternion integration can be used to update the pose, combined with accelerometer integration to update the position. Without external correction, the inertial recursion module continuously updates the robot's pose based on the IMU data.

[0037] S2. Calculate the difference between the first pose information and the second pose information.

[0038] One way to calculate the difference information is to directly calculate the translation vector difference and rotation vector difference between the first pose information and the second pose information in three-dimensional space. For example, if the pose information is represented in the form of (position vector P, quaternion Q), the translation vector difference can be simply obtained by subtracting the two position vectors, while the rotation vector difference can be obtained by quaternion operations (e.g., Q1 * Q2_inverse) to obtain a quaternion representing the relative rotation, and then converting it into a rotation vector.

[0039] Another way to calculate the difference information is to transform both the first and second pose information into a unified coordinate system, then calculate the Euclidean distance and angular difference between them, and combine these differences into a vector as the difference information. For example, both poses can be represented as 4x4 homogeneous transformation matrices, and then the difference between the two matrices or the relative transformation matrix can be calculated to extract the translation and rotation components.

[0040] S3. Obtain the first uncertainty information of the state of the inertial recursion module and the second uncertainty information of the visual observation module, and generate the uncertainty information of the difference information based on the first uncertainty information and the second uncertainty information.

[0041] The first uncertainty information reflects the range of accumulated errors of the inertial recursion module under uncorrected conditions; the second uncertainty information reflects the measurement noise level of the visual observation module itself in the current environment.

[0042] Obtaining the initial uncertainty information can be achieved through the error propagation model within the inertial recursion module. For example, in an inertial recursion system based on Kalman filtering or extended Kalman filtering, the filter maintains a state covariance matrix, whose diagonal elements reflect the uncertainties of each state variable (including position and attitude). Over time, without external observation correction, this covariance matrix gradually increases, reflecting the accumulation of inertial drift.

[0043] The second uncertainty information can be obtained through the measurement noise model of the visual observation module itself. For example, in visual odometry, the covariance matrix of the current visual observation can be estimated by analyzing factors such as the residuals of feature point matching, the richness of image texture, and lighting conditions. When feature points are sparse, lighting changes drastically, or motion is blurred, the covariance matrix of the visual observation will be large, indicating that its uncertainty is high.

[0044] The uncertainty information for generating the difference information can be achieved by combining the first uncertainty information and the second uncertainty information. For example, if the first pose information and the second pose information are independent, then the uncertainty information for the difference information can be simply obtained by adding the two uncertainty information (e.g., the covariance matrix). This reflects that the uncertainty of the difference information is the superposition of the uncertainties of the two pose estimations.

[0045] S4. Based on the difference information, the uncertainty information of the difference information, and the duration of the system in the pure inertial recursive mode, determine whether the source type of the difference information is the inertial drift accumulation type.

[0046] One way to determine this is to set a threshold. When a component of the difference information (such as position difference or angle difference) exceeds the threshold, and the duration of the system in pure inertial recursion mode exceeds another preset threshold, it is initially determined to be of the inertial drift accumulation type.

[0047] Another approach is to use statistical methods, such as calculating the Mahalanobis distance of the discrepancies. The Mahalanobis distance comprehensively considers both the magnitude and uncertainty of the discrepancies. If the Mahalanobis distance exceeds a certain statistical threshold, and the duration of the pure inertial recursive pattern is relatively long, it can be identified as inertial drift accumulation. For example, a time threshold can be set; only when the duration of the pure inertial recursive pattern exceeds this threshold should the possibility of inertial drift accumulation be considered.

[0048] S5. When the source type of the difference information is inertial drift accumulation, determine the load-bearing capacity parameter of the kinematic parameters output by the robot chassis.

[0049] One way to determine load-bearing capacity parameters is to pre-store a set of fixed load-bearing capacity parameters in the robot system. For example, based on the robot chassis design specifications, its maximum allowable acceleration, angular velocity, and maximum turning rate under specific road conditions can be set. When the problem is determined to be of the inertial drift accumulation type, these preset parameters are directly called as the load-bearing capacity parameters.

[0050] Another way to determine load-bearing capacity parameters is to dynamically adjust them based on the real-time status of the robot chassis. For example, information such as the robot chassis's current speed, load, and battery level can be obtained. If the robot's current speed is low, the load is light, and the battery level is sufficient, its load-bearing capacity can be considered strong, allowing for a larger adjustment range; conversely, if the robot's speed is high, the load is heavy, and the battery level is low, its load-bearing capacity is weak, and the adjustment range should be limited.

[0051] S6. Based on the difference information and the carrying capacity parameters, determine the adjustment parameters of the robot's motion state.

[0052] One way to determine the adjustment parameters is to use the difference information directly as the initial adjustment parameters, and then limit them according to the carrying capacity parameter. For example, if a pose component (such as position or attitude) in the difference information exceeds the maximum adjustment range allowed by the carrying capacity parameter, it is truncated to the maximum allowable value.

[0053] Another way to determine the adjustment parameters is to design a controller or optimizer that uses the difference information as the target and the carrying capacity parameter as a constraint to calculate the optimal adjustment parameters. For example, a proportional-integral-derivative (PID) controller can be used, taking the difference information as the error input, calculating the adjustment parameters through the PID algorithm, and ensuring that the adjustment parameters are within the range of the carrying capacity parameter.

[0054] S7. Based on the adjusted parameters, a gradual state correction process is formed.

[0055] One way to create a progressive state correction process is to break down the adjustment parameters into a series of small-step corrections and apply these corrections incrementally over multiple consecutive processing cycles. For example, if the adjustment parameters indicate that the robot needs to move 10 centimeters in a certain direction, this can be broken down into 10 corrections of 1 centimeter each, applied in each of the 10 consecutive control cycles.

[0056] Another way to create a gradual state correction process is to design a low-pass filter or smoothing function that takes the adjustment parameters as input and outputs a smoothly changing correction sequence. For example, exponential smoothing or moving average filters can be used to ensure that the robot's motion state transitions smoothly during the correction process, avoiding sudden jumps.

[0057] The visual observation module of this application provides a relatively accurate pose based on environmental features, while the inertial recursion module provides a pose based on inertial sensor integration, which accumulates drift errors over a long period. By calculating the difference between these two poses, the inconsistency of the current pose estimation can be quantified.

[0058] Simultaneously, the system acquires the first uncertainty information of the inertial recursion module's state and the second uncertainty information of the visual observation module. The first uncertainty information reflects the accumulated error range of the inertial recursion under uncorrected conditions, and its value increases with the increase of the pure inertial recursion time. The second uncertainty information reflects the measurement noise level of the visual observation module in the current environment; for example, its uncertainty will be relatively small after illumination is restored. Based on these two types of uncertainty information, the uncertainty information of the difference information is generated, which provides a statistical basis for subsequent judgment of the source of the difference.

[0059] The key is that this method determines whether the source of the difference information is inertial drift accumulation based on the difference information, the uncertainty of the difference information, and the duration of the system in pure inertial recursion mode. By comprehensively considering the magnitude of the difference, its statistical uncertainty, and the length of time of pure inertial recursion, it can more accurately identify pose deviations caused by inertial drift, rather than other types of errors (e.g., instantaneous anomalies in visual observation). For example, when the difference information is large, and its uncertainty indicates that the difference is statistically significant, and the system has been in pure inertial recursion mode for a long time, inertial drift accumulation is highly suspected.

[0060] When the source of the discrepancy information is determined to be inertial drift accumulation, this method determines the load-bearing capacity parameter of the kinematic parameters output by the robot chassis. This step is one of the core innovations of this application. The load-bearing capacity parameter reflects the robot chassis's ability to smoothly absorb and perform pose adjustments in its current state. For example, if the robot is traveling at high speed or in complex terrain, its load-bearing capacity may be weak, making it unsuitable for large-scale pose adjustments; conversely, if the robot is at low speed or stationary, its load-bearing capacity may be strong. By considering the actual kinematic constraints of the chassis, motion instability caused by excessively abrupt pose adjustments can be avoided.

[0061] Subsequently, based on the difference information and the carrying capacity parameters, the adjustment parameters for the robot's motion state are determined. This step combines the pose difference with the actual motion capability of the chassis to calculate an adjustment amount that effectively corrects the pose without exceeding the robot's kinematic limits. For example, if the difference information indicates that a large pose correction is needed, but the carrying capacity parameters show that the chassis cannot withstand such a large instantaneous adjustment, the adjustment parameters will be limited to a reasonable range.

[0062] Finally, based on the adjusted parameters, a gradual state correction process is formed. This means that pose correction is not completed all at once, but is achieved through a series of small, gradual updates. This gradual correction method can effectively avoid the positioning jumps caused by directly injecting large visual residuals in traditional methods. By gradually correcting the robot's motion state over multiple processing cycles, the navigation module can perceive smooth pose changes, thereby calculating stable correction commands, avoiding sudden robot turns or positioning inaccuracies, and ensuring the accuracy of inspection and docking tasks.

[0063] Compared with existing technologies, the advantage of this application lies in its refined management of the robot pose correction process. Traditional methods often directly inject the visual residual into the filter with full weight after visual observation recovery, leading to positioning jumps. This approach ignores the impact of accumulated errors during pure inertial recursion on state transitions, as well as the kinematic support capacity of the robot chassis. This application effectively solves this problem by introducing judgments on the source type of difference information, determination of robot chassis support capacity parameters, and a progressive state correction process.

[0064] Specifically, the core innovation of this application lies in: 1. Intelligent Judgment of the Source of Difference: By comprehensively considering the difference information, its uncertainty, and the duration of pure inertial recursion, the pose deviation caused by the accumulation of inertial drift can be accurately identified, avoiding misjudging other types of errors as inertial drift. This makes the correction strategy more targeted.

[0065] 2. Kinematic load-bearing capacity considerations: By introducing load-bearing capacity parameters of the robot chassis, pose adjustment is combined with the robot's actual kinematic constraints. This ensures that the pose correction process does not lead to robot instability, improving the system's robustness and safety.

[0066] 3. Progressive State Correction: A progressive correction process is adopted, breaking down large pose corrections into a series of small-step updates. This smooth transition avoids abrupt changes in positioning, enabling the navigation module to output stable control commands, thereby ensuring the robot's navigation accuracy and the accuracy of its inspection and docking after regaining its localization.

[0067] Through the above improvements, the method of this application can significantly improve the accuracy and stability of mobile robots when recovering visual-inertial fusion localization in complex lighting environments, effectively avoid the localization jump problem existing in traditional methods, and provide a more reliable localization solution for application scenarios such as industrial inspection.

[0068] like Figure 2 As shown, the difference information between the first pose information and the second pose information is calculated, including: S101. Calculate the difference in three-dimensional translation vector between the first pose information and the second pose information, as well as the difference in rotation vector obtained through quaternion operations.

[0069] Specifically, both the first and second pose information contain the robot's position and orientation in three-dimensional space. Position information is typically represented by a three-dimensional coordinate vector, while orientation information can be represented using rotation matrices, Euler angles, or quaternions. To accurately quantify the difference between the two poses, their independent deviations in translation and rotation must first be calculated. The three-dimensional translation vector difference refers to the direct vector difference between the two poses in spatial position, reflecting the robot's displacement from one position to another. The rotation vector difference obtained through quaternion operations compares the rotational components of the two poses using quaternions to obtain their angular deviation. Quaternions, as a non-singularity-based rotation representation method, can more stably and accurately describe three-dimensional rotation, avoiding the gimbaling problem that may occur with Euler angles.

[0070] S102. Combine the three-dimensional translation vector difference and rotation vector difference to form a residual vector, and use the residual vector as difference information.

[0071] Furthermore, the calculated 3D translation and rotation vector differences are combined to form a unified residual vector. This residual vector comprehensively reflects all inconsistencies in position and orientation between the two poses. For example, the 3D translation vector difference can be used as the first three components of the residual vector, and the rotation vector difference as the last three components (or the dimensions can be adjusted according to the specific representation), thereby constructing a high-dimensional vector to comprehensively characterize the pose differences. This residual vector is then used as difference information, providing basic data for subsequent error analysis and correction.

[0072] The proposed solution decomposes pose differences into three-dimensional translation vector differences and rotation vector differences, and uses quaternions to calculate the rotation difference. This allows for precise and singular-free quantification of the inconsistencies between the pose information provided by the robot's vision observation module and inertial recursion module. This decomposition method enables a more detailed and comprehensive analysis of pose errors. Subsequently, these independent difference quantities are combined into a residual vector, allowing subsequent error processing and state correction to be performed within a unified mathematical framework. This simplifies algorithm design and improves processing efficiency and accuracy. The residual vector, as difference information, provides a structured input for subsequent determination of the difference source type and adjustment parameters.

[0073] In some embodiments described above, this application proposes determining whether the source type of the difference information is inertial drift accumulation based on the difference information, the uncertainty information of the difference information, and the duration of the system in a pure inertial recursive mode. However, in actual implementation, relying solely on this information may lead to inaccurate or insufficiently robust judgments due to a lack of precise quantitative standards and statistical basis. For example, when the difference information is affected by environmental noise or occasional interference, it may be misjudged as inertial drift accumulation, or the true inertial drift accumulation may be ignored due to its inconspicuous characteristics, thereby affecting the accuracy of robot localization and the effectiveness of correction strategies.

[0074] In response, this application further proposes determining whether the source type of the aforementioned difference information is an inertial drift accumulation type based on the aforementioned difference information, the uncertainty information of the aforementioned difference information, and the duration of the system in a pure inertial recursive mode, including: S201. Determine the Mahalanobis distance based on the difference information and the uncertainty information of the difference information.

[0075] Specifically, the aforementioned difference information can be understood as the residual vector formed by combining the difference in the three-dimensional translation vector between the first pose information provided by the visual observation module and the second pose information provided by the inertial recursion module, and the difference in the rotation vector obtained through quaternion operations. The uncertainty information of the aforementioned difference information reflects the statistical distribution characteristics of the residual vector, such as its covariance matrix. Mahalanobis distance is a measure of the distance from a point to a distribution center. It considers the covariance of the data and can effectively handle the correlation between multidimensional data. When determining Mahalanobis distance, it is usually calculated based on the aforementioned difference information (i.e., the residual vector) and the uncertainty information of the aforementioned difference information (i.e., the covariance matrix of the residual vector). For example, the transpose matrix of the residual vector, the inverse matrix of the predicted covariance matrix of the residual vector, and the product of the residual vector can be calculated to obtain the Mahalanobis distance.

[0076] S202. Under the condition that the first preset condition is met, determine that the source type of the difference information is the inertial drift accumulation type.

[0077] The first preset condition is that the Mahalanobis distance is within a preset distance interval, and the ratio of the residual vector to the duration of the system in pure inertial recursive mode is within a preset ratio range.

[0078] The preset distance interval is used to limit the reasonable range of Mahalanobis distance, ensuring that the difference information is neither statistically negligible noise nor an extreme outlier, but rather conforms to the expected characteristics of inertial drift accumulation. The preset ratio range is used to limit the growth rate of the residual vector, that is, the amount of inertial drift accumulation per unit time, to distinguish between slowly accumulating inertial drift and sudden sensor failure or drastic environmental changes. Only when both of these sub-conditions are met is the source type of the difference information considered to be inertial drift accumulation.

[0079] S203. If the first preset condition is not met, determine that the source type of the difference information is not the inertial drift accumulation type.

[0080] This application's solution, by introducing Mahalanobis distance and composite preset conditions, provides a more accurate and statistically reliable basis for determining the source type of differential information. Mahalanobis distance can comprehensively consider the magnitude and uncertainty of differential information, effectively distinguishing between normal fluctuations, accumulated inertial drift, and abnormal errors. Simultaneously, by combining the duration of the pure inertial recursive mode and using ratio judgment, it can further verify whether the differential information conforms to the characteristics of inertial drift accumulating over time, avoiding misjudging instantaneous errors or short-term failures of the vision module as inertial drift. Therefore, it can more accurately identify scenarios requiring inertial drift correction, thereby triggering subsequent motion state adjustment processes.

[0081] In some preferred embodiments, it is assumed that the robot is in a purely inertial recursive mode for a period of time, and its visual observation module cannot provide reliable pose information due to changes in ambient lighting or missing textures. During this period, the inertial recursive module accumulates a certain amount of error, causing a deviation between the second pose information it provides and the actual pose. When the visual observation module returns to normal and provides the first pose information, the system calculates the residual vector (i.e., difference information) between the first pose information and the second pose information, and generates uncertainty information for the residual vector.

[0082] Specifically, the system calculates the Mahalanobis distance based on the residual vector and its uncertainty information. For example, if the residual vector is [0.1m, 0.05m, 0.02m, 0.5deg, 0.3deg, 0.1deg] (containing three-dimensional translation and rotation components), its covariance matrix reflects the uncertainty of these components. The calculated Mahalanobis distance might be 1.8.

[0083] Simultaneously, the system checks the first preset condition. Assume the preset distance interval is set to [0.5, 3.0], and the ratio of the residual vector to the duration of the pure inertial recursive mode is (e.g., the translation component ratio ranges from [0.01 m / s to 0.05 m / s], and the rotation component ratio ranges from [0.1 deg / s to 0.5 deg / s]). If the robot is in pure inertial recursive mode for 10 seconds, the average translational rate of the residual vector is 0.01 m / s, and the average rotational rate is 0.05 deg / s.

[0084] At this point, the Mahalanobis distance of 1.8 is within the preset distance range [0.5, 3.0], and the calculated average translational and rotational rates are also within their preset ratio ranges. Since both sub-conditions are met, the system determines that the source of this difference information is of the inertial drift accumulation type. If the Mahalanobis distance calculation result is 0.2 (too small, possibly noise) or 4.0 (too large, possibly a sudden malfunction), or the ratio exceeds the preset range, it will be determined as a non-inertial drift accumulation type. In this way, the system can accurately identify scenarios that require inertial drift correction.

[0085] Specifically, the process of determining the Mahalanobis distance based on the difference information and the uncertainty information of the difference information includes: S301. Based on the difference information and the uncertainty information of the difference information, determine the prediction covariance matrix of the residual vector.

[0086] The difference information is typically represented as a residual vector, which contains the 3D translation and rotation vector differences between the first and second pose information. The uncertainty information reflects the statistical distribution characteristics of these differences, usually expressed as a covariance matrix. Specifically, the predicted covariance matrix of the residual vector is predicted using Kalman filtering or other state estimation methods based on the first uncertainty information of the inertial recursion module's state and the second uncertainty information of the visual observation module, as well as their correlation. Its purpose is to quantify the statistical uncertainty of the residual vector.

[0087] S302. Calculate the transpose of the residual vector, the inverse of the prediction covariance matrix, and the product of the residual vector to obtain the Mahalanobis distance.

[0088] The proposed method yields Mahalanobis distance by multiplying the transpose of the residual vector, the inverse of the prediction covariance matrix, and the residual vector. This calculation method fully considers the correlation between the components of the residual vector and their respective variances, thus providing a more statistically reasonable distance metric. Therefore, this method can effectively distinguish between fluctuations caused by normal measurement noise and significant differences caused by the accumulation of actual inertial drift, avoiding potential misjudgments that may arise from simple Euclidean distance.

[0089] Specifically, the above-mentioned gradual state correction process based on adjusted parameters may include the following steps: S401. Based on the second uncertainty information and the adjustment parameters, generate the adjusted second uncertainty information.

[0090] The generation of adjusted second uncertainty information refers to dynamically correcting the measurement noise level of the visual observation module based on its second uncertainty information and the adjustment parameters determined by the system. These adjustment parameters reflect the robot chassis's ability to withstand changes in kinematic parameters under its current state. Therefore, by combining these adjustment parameters, the actual reliability of the visual observation module under the current environment and robot motion state can be more accurately assessed, resulting in adjusted second uncertainty information that better reflects the actual situation.

[0091] S402. Based on the first uncertainty information and the adjusted second uncertainty information, calculate the updated gain parameter.

[0092] The calculated updated gain parameter refers to a gain parameter used to fuse visual observations and inertial recursion data, based on the first uncertainty information of the inertial recursion module state and the adjusted second uncertainty information mentioned above, through Kalman filtering or other state fusion algorithms. This gain parameter dynamically weighs the relative reliability of the two information sources to determine how much weight should be given to the differing information when correcting the robot's motion state.

[0093] S403. Based on the updated gain parameters and difference information, generate a correction amount for the motion state and update the robot's motion state.

[0094] Specifically, generating a correction value for the motion state and updating the robot's motion state involves multiplying the calculated updated gain parameter by the difference between the first pose information and the second pose information to obtain a value used to correct the robot's current motion state. This correction value is then applied to the robot's pose, velocity, and other motion state parameters to gradually reduce the observed differences.

[0095] S404. Repeatedly execute the robot's motion state update within multiple processing cycles to form a progressive state correction process.

[0096] This means that the aforementioned correction of motion state is not completed all at once, but is broken down into a series of small corrections, which are gradually applied over multiple consecutive time steps or processing cycles. This gradual correction method aims to avoid system instability, oscillations, or control shocks that may result from a single large correction.

[0097] The proposed solution dynamically adjusts the uncertainty information of the visual observation module and, based on this, calculates updated gain parameters together with the uncertainty information of the inertial recursion module. This allows the correction amount for the robot's motion state to more accurately reflect the actual needs under the current environment and the robot's motion capabilities. By gradually applying this correction amount over multiple processing cycles, the system can smoothly guide the robot's estimated state to a more accurate true state, effectively avoiding system oscillations or control instability that may be caused by sudden, large-scale corrections.

[0098] In some embodiments described above, a progressive state correction process is proposed based on adjustment parameters. However, if the correction process is not effectively monitored and managed during its implementation, the update range of the robot's motion state may be too large, causing the robot to enter a state that does not conform to actual physical constraints, affecting the accuracy and safety of positioning. Furthermore, without a clear termination condition, the progressive correction process may continue indefinitely, resulting in unnecessary consumption of computational resources and reducing the system's operating efficiency. Therefore, this application further proposes a scheme to optimize and control the progressive state correction process to ensure the physical rationality and efficiency of the correction.

[0099] The method also includes: S501. Obtain the updated state sequence during the progressive state correction process, generate kinematic parameters based on the updated state sequence, and verify whether the kinematic parameters meet the preset physical constraints.

[0100] Specifically, in the progressive state correction process, the robot's motion state is updated in each processing cycle, forming an updated state sequence. This updated state sequence can contain a series of time-varying motion information such as the robot's position, attitude, velocity, and acceleration. Based on this updated state sequence, corresponding kinematic parameters can be generated, such as the robot's instantaneous linear velocity, angular velocity, linear acceleration, and angular acceleration. Subsequently, these generated kinematic parameters need to be verified to determine whether they conform to preset physical constraints. Preset physical constraints refer to the range of motion allowed by the robot's own structure, dynamic characteristics, and environmental conditions, such as the robot's maximum speed, maximum acceleration, maximum turning radius, and maximum tilt angle.

[0101] S502. When the kinematic parameters do not conform to the physical constraints, limit the magnitude of motion state updates.

[0102] When the verification results show that the kinematic parameters do not meet the preset physical constraints, such as the calculated robot speed exceeding its maximum design speed, it is necessary to limit the magnitude of motion state updates. Limiting the update magnitude can be achieved in several ways, such as adjusting the gain parameter used to generate the motion state correction, directly scaling the correction, or truncating the updated state when it exceeds the physical constraints, to ensure that the robot's motion state is always within the physically feasible range.

[0103] S503. When both the difference information and the second uncertainty information meet the preset convergence conditions, the gradual state correction process ends.

[0104] In practical applications, to improve the efficiency of the calibration process and avoid unnecessary computational resource consumption, the gradual state calibration process will terminate when both the difference information and the second uncertainty information meet the preset convergence conditions. The difference information reflects the pose inconsistency between the visual observation module and the inertial recursion module, while the second uncertainty information reflects the measurement noise level of the visual observation module itself in the current environment. The convergence condition can be set such that the values ​​of both the difference information and the second uncertainty information are less than a preset threshold, or that their changes are less than a threshold over multiple consecutive processing cycles, indicating that the system has reached a stable and high-precision positioning state.

[0105] This application's solution effectively solves the aforementioned problems by introducing a physical constraint verification and correction mechanism for kinematic parameters. Specifically, during the progressive state correction process, by continuously acquiring and updating the state sequence and generating kinematic parameters, the system can monitor in real time whether the robot's motion state is within a reasonable range. When it detects that the kinematic parameters do not conform to the preset physical constraints, such as when the robot is corrected to an excessively high speed or acceleration, the system immediately limits the magnitude of the motion state update, thereby preventing the robot from entering a physically impossible state due to over-correction or erroneous correction, ensuring the physical rationality and safety of the positioning results. Furthermore, by setting convergence conditions for difference information and second uncertainty information, this application can intelligently determine when the correction process has achieved the expected effect and terminate the process in a timely manner. This not only avoids unnecessary iterative calculations and saves computational resources but also ensures the efficiency of the correction process and the stability of the positioning results.

[0106] In some preferred embodiments, a specific example is given below. Assume that during progressive state correction, the robot's current estimated speed is 2.5 m / s, while its maximum design speed is 2.0 m / s. After acquiring the updated state sequence and generating kinematic parameters based on it, the system checks whether the 2.5 m / s speed conforms to preset physical constraints. Since 2.5 m / s exceeds the maximum speed limit of 2.0 m / s, the system determines that the kinematic parameters do not meet the physical constraints. In this case, the system limits the magnitude of the motion state update, for example, adjusting the speed correction to ensure the final speed does not exceed 2.0 m / s, or directly truncating the updated speed to 2.0 m / s, to ensure the physical rationality of the robot's motion state.

[0107] Furthermore, during the calibration process, the system continuously monitors the difference information between the visual observation module and the inertial recursion module, as well as the second uncertainty information of the visual observation module. For example, when the norm of the difference information is less than 0.01 meters and the trace of the covariance matrix of the second uncertainty information is less than 0.005, the system determines that both conditions have met the preset convergence conditions. At this point, the incremental state calibration process will end immediately, and the robot will adopt the current high-precision positioning result, thereby avoiding unnecessary subsequent iterations and improving positioning efficiency.

[0108] In some embodiments described above, an adjusted second uncertainty information is generated based on the second uncertainty information and adjustment parameters. However, during its implementation, the measurement noise level of the visual observation module (i.e., the second uncertainty information) may not adequately account for dynamic changes in the environment, particularly changes in local lighting conditions. When the robot is in an environment with drastic lighting changes, such as moving from a bright area to a shadow area, or being affected by a transient light source (such as a flash), the pose calculation accuracy of the visual observation module will be significantly interfered with, causing the second uncertainty information it provides to fail to accurately reflect the current measurement reliability. If the above problem is not addressed, the state correction process based on inaccurate second uncertainty information may not converge effectively, and may even introduce new positioning errors, thereby affecting the robustness and accuracy of the robot's high-precision positioning.

[0109] In this regard, this application further proposes the above-mentioned method of generating adjusted second uncertainty information based on the second uncertainty information and adjustment parameters, including: S601. Perform illumination change analysis on local areas of the image output by the visual observation module to identify areas of local illumination change.

[0110] Specifically, analyzing local illumination changes in images output by the visual observation module involves using image processing techniques, such as brightness histogram analysis, gradient change detection, or pixel value statistics for specific regions, to analyze image frames acquired in real-time by the visual observation module. The aim is to identify local areas in the image exhibiting significant changes in brightness, contrast, or color. These areas are typically caused by factors such as changes in external light sources, shadow movement, or alterations in the reflective properties of objects. Identifying areas of local illumination change can be achieved using techniques such as threshold segmentation, edge detection, or machine learning models to distinguish regions in the image that are significantly affected by illumination from stable regions.

[0111] S602. Evaluate the impact of local illumination changes on the pose calculation accuracy of the visual observation module.

[0112] Assessing the impact of localized illumination variations on the pose calculation accuracy of the visual observation module can be understood as quantifying the potential interference of these identified illumination variation regions with feature point extraction, descriptor matching, and pose calculation processes in visual odometry or visual SLAM (Simultaneous Localization and Mapping) algorithms. Specifically, this can be comprehensively evaluated based on factors such as the severity of the illumination variation, the size of the variation region, and its location in the image (e.g., whether it covers key feature point areas). For example, an impact model can be established to map the illumination variation parameters to the decrease in pose calculation accuracy, or a predictive model can be trained using experimental data. The aim is to provide a quantitative basis for adjusting subsequent second uncertainty information.

[0113] S603. Based on the evaluation results and adjustment parameters, generate the adjusted second uncertainty information.

[0114] In practical applications, adjusted second uncertainty information is generated based on the evaluation results and adjustment parameters. Specifically, the original second uncertainty information is corrected based on the degree of impact of local illumination changes on pose calculation accuracy obtained from the evaluation, combined with the system's preset adjustment parameters. For example, when the evaluation results show that illumination changes have a significant impact on pose calculation accuracy, the value of the second uncertainty information can be increased accordingly to reflect the increased measurement noise level of the visual observation module in the current environment. The adjustment parameter can be a weighting factor, an offset, or a more complex function, used to control the magnitude and method of adjustment. The purpose is to enable the second uncertainty information to more accurately reflect the actual measurement uncertainty of the visual observation module under dynamic lighting conditions.

[0115] This application's solution introduces local illumination change analysis of the output image of the visual observation module, enabling real-time perception and identification of environmental factors that may affect the accuracy of visual pose calculation. It is precisely because these local illumination change areas are evaluated that the system can quantify their specific impact on the pose calculation accuracy of the visual observation module. Based on this, and combined with preset adjustment parameters, the original second uncertainty information is dynamically adjusted, allowing the adjusted second uncertainty information to more accurately and robustly reflect the actual measurement uncertainty of the visual observation module under current complex lighting conditions. This mechanism ensures that the system can more reasonably assess the reliability of visual observation under unstable lighting conditions, avoiding the accumulation of positioning errors caused by visual information distortion.

[0116] In some preferred embodiments, a specific example is given below. Suppose a robot is positioning itself in an indoor environment that includes windows, causing localized areas to be exposed to direct sunlight throughout the day, creating dynamically changing shadows and highlights.

[0117] When the robot moves near the window, the image captured by the vision observation module shows significant changes in illumination in the window area and the area of ​​the ground directly exposed to sunlight. At this point, the system performs illumination change analysis on local areas of the image output by the vision observation module. For example, by analyzing the average pixel brightness and variance of specific grid areas in the image, it identifies areas where the brightness increases or decreases sharply and marks them as areas of local illumination change.

[0118] The system then evaluates the impact of these localized illumination variations on the pose calculation accuracy of the visual observation module. For example, if the illumination variation covers key texture areas used for feature matching in the image, or causes a large number of feature points to be lost or mismatched, the evaluation results will indicate a significant negative impact on pose calculation accuracy.

[0119] Based on the evaluation results, for example, if the evaluation shows that changes in illumination cause a 20% decrease in the accuracy of visual pose calculation, and combined with preset adjustment parameters (e.g., a gain factor proportional to the degree of influence of illumination changes), the system will generate adjusted second uncertainty information. Specifically, the original second uncertainty information is multiplied by a factor greater than 1 to increase its value, thereby reducing the weight of the visual observation module in the subsequent fusion process and preventing its inaccurate observations from having an excessively negative impact on the overall localization result. In this way, even in areas with poor lighting conditions, the robot can maintain high localization accuracy and robustness.

[0120] Specifically, in the above method, the step of determining the load-bearing capacity parameter of the kinematic parameters output by the robot chassis can be further refined as follows.

[0121] The load-bearing capacity parameters for determining the kinematic parameters output by the robot chassis include: S701. Obtain the load information of the robot chassis, the friction information between the wheels of the robot chassis and the ground, and the energy state information of the battery of the robot chassis.

[0122] The load information of the robot chassis refers to the weight or load currently borne by the robot. This can include the weight of the robot's own structure, the payload it carries (such as goods, tools, etc.), and any instantaneous forces that may be exerted by the external environment.

[0123] Load information can be obtained through force sensors, load cells mounted on the robot chassis, or through preset load configuration parameters. The purpose is to assess the robot's current total mass, which directly affects its motion inertia, acceleration, and braking distance.

[0124] The frictional force between the robot's wheels and the ground reflects its gripping ability when the wheels are in contact with the ground. This is affected by various factors such as ground material (e.g., concrete, tile, carpet, slippery surfaces), wheel material, wheel wear, and vertical load. Frictional force information can be obtained by identifying the ground type and looking up a table through wheel slip ratio sensors, ground friction coefficient estimation modules, or environmental perception modules (e.g., vision or lidar). The purpose is to evaluate the robot's maximum driving torque and braking capacity, as well as its motion stability under different ground conditions.

[0125] The state of energy (SGE) information of a robot chassis's battery refers to the current charge, voltage, current, and health status of the battery. This directly affects the robot's continuous operating time and its peak power output capability. SGE information can be obtained in real time through a battery management system. Its purpose is to assess the maximum power output that the robot chassis can provide under current conditions, thereby influencing its acceleration performance and sustained motion capability.

[0126] S702. Determine the load-bearing capacity parameters based on load information, friction information, and energy state information.

[0127] The load-bearing capacity parameter is a comprehensive indicator used to quantify the robot chassis's ability to safely and stably perform kinematic parameter adjustments under the current environment and its own state. This parameter can be a scalar value, a vector, or a matrix, reflecting the robot's response limit to external corrections or internal motion commands. For example, a low load-bearing capacity parameter means that the robot chassis has limited responsiveness to large kinematic adjustments, requiring a more conservative adjustment strategy; conversely, a high load-bearing capacity parameter allows the robot chassis to withstand bolder kinematic adjustments.

[0128] Specifically, the first mapping relationship can be obtained, and the load-bearing capacity parameters can be determined based on the first mapping relationship, load information, friction information, and energy state information.

[0129] The first mapping relationship includes the correspondence between different load information, different friction force information, different energy state information, and different load-bearing capacity parameters; This application's solution comprehensively assesses the physical limits and dynamic characteristics of the robot chassis by acquiring load information from the robot chassis, friction information between the wheels and the ground, and energy state information from the battery. Specifically, load information directly affects the robot's inertia, friction information determines the robot's interaction capability with the ground, and energy state information limits the robot's power output. By combining this information, the range of kinematic parameter adjustments that the robot chassis can safely bear under current working conditions can be accurately determined. For example, when the robot has a heavy load, low ground friction, or insufficient battery power, its load-bearing capacity parameters will decrease accordingly. This means that the robot will exhibit greater instability or be unable to respond promptly to large-scale motion state adjustments. Conversely, under light load, high friction, and sufficient battery power, the robot has a stronger load-bearing capacity and can accept larger adjustment parameters. In this way, this application ensures that the subsequent motion state adjustment process is physically feasible and safe, avoiding robot instability or damage due to excessive adjustment ranges.

[0130] like Figure 3 As shown in the figure, this embodiment of the invention also provides a high-precision robot positioning system based on environmental perception. The system includes: The information acquisition module is used to acquire the first pose information provided by the robot's visual observation module and the second pose information provided by the inertial recursion module; The difference calculation module is used to calculate the difference information between the first pose information and the second pose information. The uncertainty generation module is used to acquire the first uncertainty information of the state of the inertial recursion module and the second uncertainty information of the visual observation module, and generate the uncertainty information of the difference information based on the first uncertainty information and the second uncertainty information; the first uncertainty information reflects the error range accumulated by the inertial recursion module under no correction; the second uncertainty information reflects the measurement noise level of the visual observation module itself in the current environment. The source determination module is used to determine whether the source type of the difference information is the inertial drift accumulation type based on the difference information, the uncertainty information of the difference information, and the duration of the system in the pure inertial recursion mode. The determination module is used to determine the load-bearing capacity parameters of the kinematic parameters output by the robot chassis when the source type of the difference information is inertial drift accumulation. The determination module is also used to determine the adjustment parameters of the robot's motion state based on the difference information and the carrying capacity parameters; The progressive correction module is used to form a progressive state correction process based on the adjusted parameters.

[0131] This application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by a computer program instructing related hardware. This program can be stored in the computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be an internal storage unit of the task execution device (including a data sending end and / or a data receiving end) of any of the foregoing embodiments, such as the hard disk or memory of the task execution device. The computer-readable storage medium can also be an external storage device of the terminal device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the terminal device. Further, the computer-readable storage medium can include both the internal storage unit of the task execution device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the task execution device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0132] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0133] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, essentially, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0134] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A high-precision robot positioning method based on environmental perception, characterized in that, Includes the following steps: The robot acquires the first pose information provided by its visual observation module and the second pose information provided by its inertial recursion module. Calculate the difference between the first pose information and the second pose information; The system acquires first uncertainty information about the state of the inertial recursion module and second uncertainty information about the visual observation module, and generates uncertainty information about the difference information based on the first and second uncertainty information. The first uncertainty information reflects the error range accumulated by the inertial recursion module under uncorrected conditions. The second uncertainty information reflects the measurement noise level of the visual observation module itself in the current environment. Based on the difference information, the uncertainty information of the difference information, and the duration of the system in pure inertial recursion mode, determine whether the source type of the difference information is inertial drift accumulation type; When the source type of the difference information is inertial drift accumulation, determine the load-bearing capacity parameter of the kinematic parameters output by the robot chassis; Based on the difference information and the bearing capacity parameters, the adjustment parameters of the robot's motion state are determined; Based on the adjustment parameters, a gradual state correction process is formed.

2. The high-precision robot positioning method based on environmental perception according to claim 1, characterized in that, The calculation of the difference information between the first pose information and the second pose information includes: Calculate the three-dimensional translation vector difference between the first pose information and the second pose information, as well as the rotation vector difference obtained through quaternion operations; The three-dimensional translation vector difference and the rotation vector difference are combined to form a residual vector, and the residual vector is used as the difference information.

3. The high-precision robot positioning method based on environmental perception according to claim 2, characterized in that, The determination of whether the source type of the difference information is an inertial drift accumulation type based on the difference information, the uncertainty information of the difference information, and the duration of the system in pure inertial recursion mode includes: Based on the difference information and the uncertainty information of the difference information, the Mahalanobis distance is determined; Under the condition that the first preset condition is met, the source type of the difference information is determined to be inertial drift accumulation type; the first preset condition is that the Mahalanobis distance is within a preset distance interval, and the ratio of the residual vector to the duration of the system in pure inertial recursion mode is within a preset ratio range. If the first preset condition is not met, it is determined that the source type of the difference information is not the inertial drift accumulation type.

4. The high-precision robot positioning method based on environmental perception according to claim 3, characterized in that, The determination of the Mahalanobis distance based on the difference information and the uncertainty information of the difference information includes: Based on the difference information and the uncertainty information of the difference information, the prediction covariance matrix of the residual vector is determined; The Mahalanobis distance is obtained by multiplying the transpose of the residual vector, the inverse of the prediction covariance matrix, and the residual vector.

5. The high-precision robot positioning method based on environmental perception according to claim 1, characterized in that, The gradual state correction process based on the adjustment parameters includes: Based on the second uncertainty information and the adjustment parameters, the adjusted second uncertainty information is generated; Based on the first uncertainty information and the adjusted second uncertainty information, the updated gain parameter is calculated; Based on the updated gain parameters and the difference information, a correction amount for the motion state is generated, and the robot's motion state is updated. The robot's motion state is repeatedly updated over multiple processing cycles to form a progressive state correction process.

6. The high-precision robot positioning method based on environmental perception according to claim 5, characterized in that, The method further includes: The updated state sequence during the progressive state correction process is obtained, and kinematic parameters are generated based on the updated state sequence. The kinematic parameters are then verified to meet the preset physical constraints. When the kinematic parameters do not conform to the physical constraints, the magnitude of motion state updates is limited; When both the difference information and the second uncertainty information meet the preset convergence conditions, the gradual state correction process ends.

7. A high-precision robot positioning method based on environmental perception according to claim 5, characterized in that, The step of generating adjusted second uncertainty information based on the second uncertainty information and the adjustment parameters includes: Perform illumination change analysis on local areas of the image output by the visual observation module to identify areas of local illumination change; Evaluate the impact of the localized illumination variation region on the pose calculation accuracy of the visual observation module; Based on the evaluation results and the adjustment parameters, the adjusted second uncertainty information is generated.

8. The high-precision robot positioning method based on environmental perception according to claim 1, characterized in that, The load-bearing capacity parameters for determining the kinematic parameters output by the robot chassis include: Obtain the load information of the robot chassis, the friction information between the wheels of the robot chassis and the ground, and the energy state information of the robot chassis battery; The load-bearing capacity parameter is determined based on the load information, the friction force information, and the energy state information.

9. A high-precision robot positioning method based on environmental perception according to claim 8, characterized in that, The step of determining the load-bearing capacity parameter based on the load information, the frictional force information, and the energy state information includes: Obtain the first mapping relationship; the first mapping relationship includes the correspondence between different load information, different friction force information, different energy state information, and different load-bearing capacity parameters. The load-bearing capacity parameter is determined based on the first mapping relationship, the load information, the frictional force information, and the energy state information.

10. A high-precision robot positioning system based on environmental perception, characterized in that, The system includes: The information acquisition module is used to acquire the first pose information provided by the robot's visual observation module and the second pose information provided by the inertial recursion module; The difference calculation module is used to calculate the difference information between the first pose information and the second pose information; An uncertainty generation module is used to acquire first uncertainty information of the state of the inertial recursion module and second uncertainty information of the visual observation module, and generate uncertainty information of the difference information based on the first uncertainty information and the second uncertainty information; the first uncertainty information reflects the error range accumulated by the inertial recursion module under uncorrected conditions; the second uncertainty information reflects the measurement noise level of the visual observation module itself in the current environment; The source determination module is used to determine whether the source type of the difference information is an inertial drift accumulation type based on the difference information, the uncertainty information of the difference information, and the duration of the system being in a pure inertial recursive mode. The determination module is used to determine the load-bearing capacity parameter of the kinematic parameters output by the robot chassis when the source type of the difference information is inertial drift accumulation type; The determining module is also used to determine the adjustment parameters of the robot's motion state based on the difference information and the bearing capacity parameters; The progressive correction module is used to form a progressive state correction process based on the adjustment parameters.

Images