Robot positioning method and device, computer device, storage medium and program product

Abstract

Embodiments of the present application provide a robot positioning method and device, computer equipment, storage medium and computer program product, and relate to the technical field of computer vision. The method comprises: obtaining a current temperature and point cloud fusion feature matrix; encoding the temperature and point cloud fusion feature matrix to obtain a current encoding feature vector; obtaining a relative position change matrix of a target robot between a previous time step and a current time step and a temperature feature difference matrix of the target robot between the previous time step and the current time step according to a historical encoding feature vector of the previous time step adjacent to the current time step and the current encoding feature vector; and obtaining target current pose information of the target robot at the current time step according to the relative position change matrix and the temperature feature difference matrix. The method improves positioning accuracy, path planning efficiency and accuracy, and map updating accuracy of the corresponding space.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a robot positioning method, apparatus, computer equipment, storage medium, and computer program product. Background Technology

[0002] In some scenarios, such as data center environments, robots are needed to conduct inspections and comprehensively perceive the environmental conditions of the data center, including information such as equipment operating temperature and server room layout. Among these, the positioning of the inspection robot is crucial.

[0003] Current positioning technology for inspection robots mainly relies on traditional laser vision technology, using a single data source, such as point cloud data, to locate the inspection robot. In environments with similar data center layouts, the "false loopback" phenomenon is prone to occur, leading to positioning errors and low positioning accuracy. Summary of the Invention

[0004] Therefore, it is necessary to provide a robot positioning method, device, computer equipment, storage medium, and computer program product to address the aforementioned technical problems.

[0005] Firstly, this application provides a robot localization method. The method includes:

[0006] Obtain the fusion feature matrix of the current temperature and point cloud of the target robot in the target space at the current time step;

[0007] The temperature and point cloud fusion feature matrix is ​​coded to obtain the current coded feature vector of the temperature and point cloud fusion feature matrix;

[0008] Based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, obtain the relative position change matrix of the target robot between the previous time step and the current time step, and obtain the temperature feature difference matrix of the target robot between the previous time step and the current time step;

[0009] Based on the relative position change matrix and the temperature feature difference matrix, the target robot's current pose information at the current time step is obtained.

[0010] In one embodiment, the step of feature encoding the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix includes: constructing an initial state space model for the target space; obtaining feature association parameters of the initial state space model based on the current temperature and point cloud fusion feature matrix; obtaining a target state space model for the target space based on the feature association parameters and the initial state space model; and using the target state space model to feature encode the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix.

[0011] In one embodiment, obtaining the relative position change matrix of the target robot between the previous time step and the current time step includes: determining a first point cloud subset in the historical point cloud dataset corresponding to the previous time step; filtering out a second point cloud subset that matches the first point cloud subset in the current point cloud dataset corresponding to the current time step based on the historical encoded feature vector and the current encoded feature vector; a correspondence exists between the multiple second point cloud data contained in the second point cloud subset and the multiple first point cloud data contained in the first point cloud subset; and obtaining the relative position change matrix based on the difference information between the second point cloud subset and the first point cloud subset.

[0012] In one embodiment, obtaining the temperature feature difference matrix of the target robot between the previous time step and the current time step includes: obtaining temperature change information of the current temperature feature matrix corresponding to the current time step and the historical temperature feature matrix corresponding to the previous time step based on the historical encoded feature vector and the current encoded feature vector; and obtaining the temperature feature difference matrix based on the temperature change information.

[0013] In one embodiment, obtaining the target robot's current pose information at the current time step based on the relative position change matrix and the temperature feature difference matrix includes: constructing an initial pose prediction function and an initial prediction error covariance function for the target robot in the target space; obtaining a target pose prediction function corresponding to the initial pose prediction function and a target prediction error covariance function corresponding to the initial prediction error covariance function based on the relative position change matrix and the temperature feature difference matrix; and using the target pose prediction function and the target prediction error covariance function to obtain the target robot's current pose information at the current time step.

[0014] In one embodiment, obtaining the target pose prediction function corresponding to the initial pose prediction function based on the relative position change matrix and the temperature feature difference matrix includes: obtaining the temperature observation residual of the temperature feature difference matrix for the initial pose prediction function based on the temperature feature difference matrix; obtaining a first optimized pose prediction function corresponding to the initial pose prediction function based on the temperature observation residual; obtaining the spatial observation residual of the relative position change matrix for the initial pose prediction function based on the relative position change matrix; obtaining a second optimized pose prediction function corresponding to the initial pose prediction function based on the spatial observation residual; and obtaining the target pose prediction function corresponding to the initial pose prediction function based on the first optimized pose prediction function and the second optimized pose prediction function.

[0015] In one embodiment, obtaining the first optimized pose prediction function corresponding to the initial pose prediction function based on the temperature observation residual includes: obtaining the temperature Kalman gain of the temperature feature difference matrix for the initial pose prediction function; and obtaining the first optimized pose prediction function corresponding to the initial pose prediction function based on the temperature observation residual and the temperature Kalman gain.

[0016] In one embodiment, obtaining the second optimized pose prediction function corresponding to the initial pose prediction function based on the spatial observation residual includes: obtaining the spatial Kalman gain of the relative position change matrix with respect to the initial pose prediction function; and obtaining the second optimized pose prediction function corresponding to the initial pose prediction function based on the spatial observation residual and the spatial Kalman gain.

[0017] In one embodiment, obtaining the target prediction error covariance function corresponding to the initial prediction error covariance function includes: obtaining the temperature Kalman gain of the temperature feature difference matrix for the initial pose prediction function; obtaining a first optimized prediction error covariance function corresponding to the initial prediction error covariance function based on the temperature Kalman gain; obtaining the spatial Kalman gain of the relative position change matrix for the initial pose prediction function; obtaining a second optimized prediction error covariance function corresponding to the initial prediction error covariance function based on the spatial Kalman gain; and obtaining the target prediction error covariance function corresponding to the initial prediction error covariance function based on the first optimized prediction error covariance function and the second optimized prediction error covariance function.

[0018] Secondly, this application provides a robot positioning device. The device includes:

[0019] The fusion module is used to obtain the fusion feature matrix of the current temperature and point cloud of the target robot in the target space at the current time step;

[0020] The encoding module is used to perform feature encoding on the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix;

[0021] The first calculation module is used to obtain, based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, the relative position change matrix of the target robot between the previous time step and the current time step, and the temperature feature difference matrix of the target robot between the previous time step and the current time step;

[0022] The second calculation module is used to obtain the target robot's current pose information at the current time step based on the relative position change matrix and the temperature feature difference matrix.

[0023] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0024] Obtain the fusion feature matrix of the current temperature and point cloud of the target robot in the target space at the current time step;

[0025] The temperature and point cloud fusion feature matrix is ​​coded to obtain the current coded feature vector of the temperature and point cloud fusion feature matrix;

[0026] Based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, obtain the relative position change matrix of the target robot between the previous time step and the current time step, and obtain the temperature feature difference matrix of the target robot between the previous time step and the current time step;

[0027] Based on the relative position change matrix and the temperature feature difference matrix, the target robot's current pose information at the current time step is obtained.

[0028] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0029] Obtain the fusion feature matrix of the current temperature and point cloud of the target robot in the target space at the current time step;

[0030] The temperature and point cloud fusion feature matrix is ​​coded to obtain the current coded feature vector of the temperature and point cloud fusion feature matrix;

[0031] Based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, obtain the relative position change matrix of the target robot between the previous time step and the current time step, and obtain the temperature feature difference matrix of the target robot between the previous time step and the current time step;

[0032] Based on the relative position change matrix and the temperature feature difference matrix, the target robot's current pose information at the current time step is obtained.

[0033] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0034] Obtain the fusion feature matrix of the current temperature and point cloud of the target robot in the target space at the current time step;

[0035] The temperature and point cloud fusion feature matrix is ​​coded to obtain the current coded feature vector of the temperature and point cloud fusion feature matrix;

[0036] Based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, obtain the relative position change matrix of the target robot between the previous time step and the current time step, and obtain the temperature feature difference matrix of the target robot between the previous time step and the current time step;

[0037] Based on the relative position change matrix and the temperature feature difference matrix, the target robot's current pose information at the current time step is obtained.

[0038] In the aforementioned robot localization method, apparatus, computer equipment, storage medium, and computer program products, firstly, the current temperature and point cloud fusion feature matrix of the target robot in the target space at the current time step can be obtained; next, the temperature and point cloud fusion feature matrix can be feature-encoded to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix; furthermore, based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, the relative position change matrix of the target robot between the previous time step and the current time step, as well as the temperature feature difference matrix of the target robot between the previous time step and the current time step, can be obtained; finally, based on the relative position change matrix and the temperature feature difference matrix, the target robot's current pose information at the current time step can be obtained. The method provided in this application embodiment can combine temperature data and point cloud data to generate a temperature and point cloud fusion feature matrix containing temperature features and spatial features. This can improve the distinguishability of the target robot to the environment and the accuracy of its localization. Furthermore, feature encoding can be performed on the temperature and point cloud fusion feature matrix to improve the accuracy of feature extraction, thereby improving localization accuracy. In addition, the iterative nearest point algorithm and Kalman filter can be used to predict the pose of the target robot, improving the accuracy of pose prediction. This can further improve localization accuracy, the efficiency and accuracy of path planning, and the accuracy of map updates in the corresponding space. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A flowchart illustrating a robot localization method provided in an embodiment of this application;

[0041] Figure 2 A schematic diagram of a Mamba network encoding process provided for an embodiment of this application;

[0042] Figure 3 A schematic diagram illustrating a process for obtaining the current encoded feature vector, provided in an embodiment of this application;

[0043] Figure 4 A schematic diagram of a process for obtaining a relative position change matrix provided in an embodiment of this application;

[0044] Figure 5A schematic diagram of a process for obtaining a temperature feature difference matrix provided in an embodiment of this application;

[0045] Figure 6 This application provides a schematic diagram of a process for obtaining the current pose information of a target, as illustrated in an embodiment of the present application.

[0046] Figure 7 A flowchart illustrating the process of obtaining a target pose prediction function is provided in an embodiment of this application.

[0047] Figure 8 A structural block diagram of a robot positioning device provided in an embodiment of this application;

[0048] Figure 9 This is an internal structural diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] In one exemplary embodiment, such as Figure 1 As shown, a robot localization method is provided. This embodiment illustrates the method applied to a server. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0051] Step 102: Obtain the current temperature and point cloud fusion feature matrix of the target robot in the target space at the current time step.

[0052] The target space can be any space requiring inspection, such as a server room or data center. The target robot can be equipped with multiple sensors, such as a high-precision infrared temperature measurement camera and LiDAR. In this step, the target robot simultaneously collects temperature matrix and point cloud data during its movement. It can utilize the onboard high-precision infrared temperature measurement camera to collect temperature matrix data in real time. The camera measures temperature by receiving infrared radiation emitted from the surface of objects in the environment and generates a two-dimensional temperature matrix. Simultaneously, the target robot can continuously capture spatial point cloud data using LiDAR. This process requires ensuring strict synchronization between the temperature matrix and the point cloud data acquisition timestamps. At the same time, i.e., the same time step, the acquired temperature and point cloud data must correspond to each other for effective subsequent fusion processing. For example, a high-precision clock synchronization module can be used to calibrate the acquisition times of the infrared temperature measurement camera and the lidar, ensuring they are consistent in time. Next, environmental compensation and spatial correction can be performed on the acquired temperature and point cloud data to ensure alignment in a unified coordinate system. Environmental compensation mainly considers the impact of environmental factors (e.g., light, reflection) on temperature measurement and point cloud acquisition, correcting them using appropriate algorithms. Spatial correction transforms the point cloud data and temperature matrix to the same spatial coordinate system, typically requiring coordinate transformation and calibration operations. For example, determining the transformation relationship between the lidar coordinate system and the infrared camera coordinate system ensures accurate matching, laying the foundation for subsequent fusion operations. Additionally, data standardization can be performed on the acquired temperature and point cloud data, and the temperature matrix can be normalized to map temperature values ​​to a standard range, facilitating subsequent fusion. Point cloud data is processed through filtering and downsampling to retain key points and reduce computational load.

[0053] Furthermore, based on the temperature matrix and point cloud data after data processing, the current temperature of the target robot in the target space at the current time step can be fused with the point cloud feature matrix. Specifically, this can include: First, projection calculation, using the intrinsic and extrinsic parameters of the infrared camera, projecting the point cloud data onto the two-dimensional pixel plane of the infrared camera, as shown in equations (1) and (2). The intrinsic parameters of the infrared camera include the focal length. Optical center coordinates External parameters describe the camera's position and orientation information in space.

[0054]

[0055] Among them, These are the projected pixel coordinates. The three-dimensional coordinates of the point cloud points. Focal length It is the light center.

[0056] The second step is temperature binding, which associates pixel coordinates with temperature values, based on the projected pixel coordinates. Find the corresponding temperature value in the temperature matrix. And attach it to the point cloud points to form corresponding extended points, as shown in equation (3). Since the correspondence between point cloud points and pixel coordinates is established in the projection calculation, the temperature value corresponding to each point cloud point can be accurately found in the temperature matrix through this correspondence. Thus, each extended point not only contains its own spatial coordinate information, but also carries the corresponding temperature information, realizing the fusion of temperature and spatial features at the point cloud point level, and obtaining the initial temperature and point cloud fusion feature matrix. For example, for a point cloud point on a rack in a data center, through projection and temperature binding operations, the point cloud point not only has coordinate information representing its position in space, but also is associated with the temperature value at that position, thus forming point cloud data containing temperature-spatial features.

[0057]

[0058] The third step involves filtering and downsampling the initial temperature and point cloud fusion feature matrix. Filtering aims to remove outliers and noise from the data. For example, statistical filtering can be used to remove outliers with densities significantly lower or higher than their surroundings based on the density distribution of points in the point cloud. Simultaneously, temperature data can be smoothed to remove abnormal temperature values ​​caused by sensor errors or environmental interference, ensuring data stability and reliability. Downsampling, on the other hand, reduces the amount of data and improves computational efficiency while preserving key data features. For example, a voxel lattice filtering method can be used to divide the point cloud space into voxel lattices of a certain size, selecting a representative point within each voxel lattice to reduce the number of points in the point cloud. Then, the current temperature and point cloud fusion feature matrix corresponding to the initial temperature and point cloud fusion feature matrix can be obtained, as shown in equation (4).

[0059]

[0060] Step 104: Perform feature encoding on the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix.

[0061] In this step, such as Figure 2As shown, the Mamba network can be used to encode the feature matrix of temperature and point cloud fusion. The encoded position and temperature features can be obtained by using the Mamba module based on the fused temperature-point cloud feature matrix. The core principle of this Mamba network is based on the selective state space model (SSM). By representing the parameters of the state space model as functions of the input features, dynamic modeling of the input features can be achieved. Specifically, it can include: First, an initial state space model for the target space can be constructed; Second, based on the current temperature and point cloud fusion feature matrix, the feature association parameters of the initial state space model can be obtained. The Mamba network can represent the parameters of the initial state space model as functions of the input features, that is, functions of the current temperature and point cloud fusion feature matrix, to obtain the feature association parameters. See equation (5):

[0062]

[0063] The third step is to obtain the target state space model for the target space based on the feature association parameters and the initial state space model. Then, update the initial state space model using the feature association parameters to obtain the target state space model, as shown in equation (6).

[0064]

[0065] in, and These are adaptive parameters calculated based on the current input. At each time step, the Mamba network dynamically adjusts its parameters based on the current temperature and the point cloud fusion feature matrix. and The value is used to update the state. This process enables the Mamba network to track changes in environmental characteristics. For example, when a robot moves to different areas of a data center, and the temperature and spatial structure change, the network accurately reflects these changes by adjusting parameters and updating its state, thus effectively modeling the current temperature and point cloud fusion feature matrix. In the fourth step, the Mamba network uses the target state space model to encode the temperature and point cloud fusion feature matrix, obtaining the current encoded feature vector. This current encoded feature vector is an abstract representation of the temperature and spatial features of the current frame; it integrates temperature and spatial information and can be compared with the encoded vectors of historical frames in subsequent processing.

[0066] Step 106: Based on the historical encoded feature vector and the current encoded feature vector of the target robot in the previous time step adjacent to the current time step, obtain the relative position change matrix of the target robot between the previous time step and the current time step, and obtain the temperature feature difference matrix of the target robot between the previous time step and the current time step.

[0067] The process involves obtaining the relative position change matrix of the target robot between the previous and current time steps based on the historical encoded feature vectors of the target robot in the previous time step and the current encoded feature vector. This can include: the point cloud data of the previous and current time steps are point cloud data from two adjacent frames. The Iterative Closest Point (ICP) algorithm can be used to match the historical point cloud data of the previous time step with the current point cloud data of the current time step. The basic idea of ​​the ICP algorithm is to find the optimal transformation (e.g., translation and rotation) between the two point clouds iteratively, so that one point cloud overlaps with the other point cloud as much as possible under this transformation. Specifically, firstly, a first subset of point clouds (e.g., randomly selecting a certain proportion of points) is selected from the historical frame point clouds of the previous time step. Then, in the current frame point cloud of the current time step, the corresponding points that are closest to the first subset of point cloud data (based on some distance metric, such as Euclidean distance) are found to obtain a second subset of point clouds. There is a one-to-one correspondence between the multiple second point cloud data contained in the second point cloud subset and the multiple first point cloud data contained in the first point cloud subset. Multiple matching point cloud pairs exist within these first and second point cloud subsets. Therefore, an initial transformation matrix can be calculated based on these multiple matching point cloud pairs, which can be understood as transforming the point cloud from one frame to the approximate position of the point cloud in another frame. Next, based on the differences between the two point clouds in the transformed matching point cloud pairs, the transformation matrix is ​​continuously adjusted. This process is repeated until convergence, i.e., the differences no longer decrease significantly, yielding the relative position change matrix of the target robot between the previous and current time steps. This relative position change matrix obtained through the ICP algorithm represents the relative pose transformation of the point cloud data between adjacent frames, thus providing an important basis for the pose estimation of the target robot.

[0068] Based on the historical encoded feature vector and the current encoded feature vector of the target robot in the previous time step adjacent to the current time step, the temperature feature difference matrix between the target robot and the previous time step is obtained. This can include: obtaining the temperature change information of the current temperature feature matrix corresponding to the current time step and the historical temperature feature matrix corresponding to the previous time step based on the historical encoded feature vector and the current encoded feature vector; and obtaining the temperature feature difference matrix based on the temperature change information. Between adjacent frames, the temperature feature difference value is calculated by comparing the change in the temperature feature matrix. This can be achieved by calculating the difference in temperature values ​​at corresponding pixel positions, resulting in a difference matrix representing the temperature change. For example, for the temperature matrices of two adjacent time steps... and ,calculate The difference matrix is ​​obtained, where each element represents the temperature change at the corresponding location. This temperature feature difference value contains information about the change in ambient temperature during the target robot's movement, reflecting the target robot's pose changes from another perspective, because the target robot may perceive different temperature distributions at different locations, such as the temperature rising when close to a heat-generating device and decreasing when moving away.

[0069] Step 108: Based on the relative position change matrix and the temperature feature difference matrix, obtain the target robot's current pose information at the current time step.

[0070] In this step, the first step is to construct the initial pose prediction function and the initial prediction error covariance function of the target robot in the target space. Specifically, it is assumed that the target robot has a certain motion model from the previous time step to the current time step. This motion model can be established based on the kinematic principles or empirical data of the target robot. If the target robot is a wheeled inspection robot, its motion model can consider the relationship between factors such as wheel speed and steering angle and the changes in robot position and posture. Based on this motion model, the initial pose prediction function and the initial prediction error covariance function of the target robot in the target space are constructed, as shown in equations (7) and (8):

[0071]

[0072] in, It is a state transition matrix, which describes the transformation relationship of the target robot's pose between different time steps; It is process noise, used to account for uncertainties in the actual motion process, such as wheel slippage, motion deviation caused by uneven ground, etc. Process noise follows a Gaussian distribution; It is the covariance matrix of the process noise, which describes the distribution characteristics of the process noise; This is the initial predicted pose; This represents the initial prediction error covariance.

[0073] The second step is to obtain the temperature observation residual of the temperature feature difference matrix for the initial pose prediction function based on the temperature feature difference matrix; and to obtain the first optimized pose prediction function corresponding to the initial pose prediction function based on the temperature observation residual. Specifically, the main function of the temperature matrix data is to provide environmental feature localization. Based on the temperature feature difference matrix, the corresponding temperature observation value can be obtained. Then, based on the temperature observation value and the initial predicted pose, the temperature observation residual of the temperature feature difference matrix for the initial pose prediction function can be obtained, as shown in equation (9):

[0074]

[0075] in, The temperature matrix provides pose information (reflecting pose changes through differences in environmental characteristics). The observation matrix maps the state space (robot pose state) to the observation space (temperature feature space). The determination of the observation matrix depends on the specific relationship model between temperature features and robot pose. For example, if the temperature features are directly related to the robot's position in a certain direction, then... The corresponding element will reflect this relationship; This is the residual value observed at that temperature.

[0076] Next, based on the uncertainty of temperature in predicting the pose of the target robot, the temperature Kalman gain can be obtained using a Kalman filter, as shown in equation (10):

[0077]

[0078] in, This is the covariance matrix of the temperature observation noise, used to describe the error distribution characteristics of the temperature observation data; For temperature Kalman gain, This reflects the system's level of confidence in the temperature matrix observation at this step. Furthermore, the initial pose prediction function and the initial prediction error covariance function can be updated using the temperature observation residual and the temperature Kalman gain, respectively, to obtain the first optimized pose prediction function and the first optimized prediction error covariance function, as shown in equations (11) and (12):

[0079]

[0080] The third step is to obtain the spatial observation residual of the relative position change matrix for the initial pose prediction function based on the relative position change matrix; and to obtain the second optimized pose prediction function corresponding to the initial pose prediction function based on the spatial observation residual. Specifically, spatial observation values ​​can be obtained based on the relative position change matrix, and then the spatial observation residual can be obtained based on the spatial observation values, as shown in equation (13):

[0081]

[0082] in, This refers to the residual of the space observation; The relative pose information is calculated by ICP of the point cloud and used as an external pose observation for mapping updates. This step is similar to the calculation of the residual of the temperature matrix observation. By comparing the pose calculated by the point cloud ICP with the current pose estimate updated based on the temperature matrix, the difference between the two is obtained, providing a basis for subsequent fusion of point cloud observation information.

[0083] Next, based on the uncertainty of the pose prediction of the target robot using point cloud data, the spatial Kalman gain can be obtained using a Kalman filter, as shown in equation (14):

[0084]

[0085] in, Measurement noise for point cloud ICP is used to describe the error characteristics of point cloud ICP observation data. The spatial Kalman gain reflects the system's confidence in the point cloud ICP observation data under the current pose estimation. It is similar to the Kalman gain calculation when observing the temperature matrix, but takes into account the influence of point cloud ICP measurement noise.

[0086] Furthermore, in one possible implementation, the initial pose prediction function and the initial prediction error covariance function can be updated using the spatial observation residual and the spatial Kalman gain, respectively, to obtain the second optimized pose prediction function and the second optimized prediction error covariance function, as shown in equations (15) and (16); then, the target pose prediction function can be obtained based on the first optimized pose prediction function and the second optimized pose prediction function, and the target prediction error covariance function can be obtained based on the first optimized prediction error covariance function and the second optimized prediction error covariance function, as shown in equations (17) and (18). In another possible implementation, the first optimized pose prediction function and the first optimized prediction error covariance function can be updated using the spatial observation residual and the spatial Kalman gain, respectively, to obtain the target pose prediction function and the target prediction error covariance function, as shown in equations (19) and (20).

[0087]

[0088] Finally, the target robot's current pose information at the current time step can be obtained based on the target pose prediction function and the target prediction error covariance function. This target current pose information can include the target robot's position information and attitude information, and can be expressed as equation (21):

[0089]

[0090] in, Indicates position coordinates, Indicates the direction (attitude) angle.

[0091] Furthermore, the target robot's current pose information at the current time step can be used for map updating and loop closure detection. Map updating involves Simultaneous Localization and Mapping (SLAM) based on the fused pose. If the temperature matrix and point cloud features indicate location repetition, loop closure locations are marked, and global coordinates are adjusted to reduce accumulated errors. Loop closure detection and map optimization involve matching loops using temperature matrix features. After detecting a loop, the map is optimized and adjusted in the backend to reduce drift.

[0092] In this embodiment, the method first obtains the current temperature and point cloud fusion feature matrix of the target robot in the target space at the current time step; next, it encodes the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix; further, it obtains the relative position change matrix of the target robot between the previous time step and the current time step, and the temperature feature difference matrix of the target robot between the previous time step and the current time step, based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector; finally, it obtains the target robot's current pose information at the current time step based on the relative position change matrix and the temperature feature difference matrix. The method provided in this application embodiment can combine temperature data and point cloud data to generate a temperature and point cloud fusion feature matrix containing temperature features and spatial features. This can improve the distinguishability of the target robot to the environment and the accuracy of its localization. Furthermore, feature encoding can be performed on the temperature and point cloud fusion feature matrix to improve the accuracy of feature extraction, thereby improving localization accuracy. In addition, the iterative nearest point algorithm and Kalman filter can be used to predict the pose of the target robot, improving the accuracy of pose prediction. This can further improve localization accuracy, the efficiency and accuracy of path planning, and the accuracy of map updates in the corresponding space.

[0093] In one exemplary embodiment, such as Figure 3 As shown, step 104 may include steps 302 to 308. Wherein:

[0094] Step 302: Construct an initial state space model for the target space.

[0095] Step 304: Obtain the feature association parameters of the initial state space model based on the current temperature and the point cloud fusion feature matrix.

[0096] The Mamba network can express the parameters of the initial state space model as a function of the input features, that is, a function of the current temperature and the point cloud fusion feature matrix, to obtain the feature association parameters, as shown in equation (5).

[0097] Step 306: Obtain the target state space model for the target space based on the feature association parameters and the initial state space model.

[0098] Specifically, based on the feature association parameters and the initial state space model, a target state space model for the target space is obtained. The initial state space model is then updated using these feature association parameters to obtain the target state space model, as shown in equation (6). and These are adaptive parameters calculated based on the current input. At each time step, the Mamba network dynamically adjusts its parameters based on the current temperature and the point cloud fusion feature matrix. and The value is used to update the state. This process enables the Mamba network to track changes in environmental characteristics. For example, when a robot moves to different areas of a data center and the temperature and spatial structure change, the network accurately reflects these changes by adjusting parameters and updating the state, thereby effectively modeling the current temperature and point cloud fusion feature matrix.

[0099] Step 308: Using the target state space model, perform feature encoding on the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix.

[0100] The Mamba network utilizes a target state-space model to encode the fused temperature and point cloud feature matrix, obtaining the current encoded feature vector. This current encoded feature vector is an abstract representation of the temperature and spatial features of the current frame, fusing temperature and spatial information, and can be compared with the encoded vectors of historical frames in subsequent processing.

[0101] In the method of this embodiment, a Mamba network can be introduced to encode the feature matrix of temperature and point cloud fusion, thereby improving the accuracy of feature extraction.

[0102] In one exemplary embodiment, such as Figure 4 As shown, obtaining the relative position change matrix of the target robot between the previous time step and the current time step in step 106 can include steps 402 to 406. Wherein:

[0103] Step 402: Determine the first subset of point clouds in the historical point cloud dataset corresponding to the previous time step.

[0104] Step 404: Based on the historical encoding feature vector and the current encoding feature vector, select the second point cloud subset that matches the first point cloud subset from the current point cloud dataset corresponding to the current time step.

[0105] There is a correspondence between the multiple second point cloud data contained in the second point cloud subset and the multiple first point cloud data contained in the first point cloud subset.

[0106] Step 406: Obtain the relative position change matrix based on the difference information between the second point cloud subset and the first point cloud subset.

[0107] In this context, the point cloud data from the previous time step and the current time step are point cloud data from two adjacent frames. The Iterative Closest Point (ICP) algorithm can be used to match the historical point cloud data from the previous time step with the current point cloud data from the current time step. The basic idea of ​​the ICP algorithm is to find the optimal transformation (e.g., translation and rotation) between the two point clouds iteratively, so that one point cloud overlaps with the other point cloud as much as possible under this transformation. Specifically, firstly, a first subset of point clouds is selected from the historical frame point clouds of the previous time step (e.g., a certain proportion of points are randomly selected). Then, in the current frame point cloud of the current time step, the corresponding points that are closest to the first subset of point cloud data (based on some distance metric, such as Euclidean distance) are found to obtain a second subset of point clouds. There is a one-to-one correspondence between the multiple second point cloud data contained in the second point cloud subset and the multiple first point cloud data contained in the first point cloud subset. Multiple matching point cloud pairs exist within these first and second point cloud subsets. Therefore, an initial transformation matrix can be calculated based on these multiple matching point cloud pairs, which can be understood as transforming the point cloud from one frame to the approximate position of the point cloud in another frame. Next, based on the differences between the two point clouds in the transformed matching point cloud pairs, the transformation matrix is ​​continuously adjusted. This process is repeated until convergence, i.e., the differences no longer decrease significantly, yielding the relative position change matrix of the target robot between the previous and current time steps. This relative position change matrix obtained through the ICP algorithm represents the relative pose transformation of the point cloud data between adjacent frames, thus providing an important basis for the pose estimation of the target robot.

[0108] In one exemplary embodiment, such as Figure 5 As shown, obtaining the temperature feature difference matrix of the target robot between the previous time step and the current time step in step 106 can include steps 502 to 504. Wherein:

[0109] Step 502: Based on the historical coding feature vector and the current coding feature vector, obtain the temperature change information of the current temperature feature matrix corresponding to the current time step and the historical temperature feature matrix corresponding to the previous time step.

[0110] Step 504: Obtain the temperature feature difference matrix based on the temperature change information.

[0111] Specifically, based on historical and current encoded feature vectors, the temperature change information between the current temperature feature matrix at the current time step and the historical temperature feature matrix at the previous time step is obtained. Based on this temperature change information, a temperature feature difference matrix is ​​obtained. Between adjacent frames, the difference in temperature features is calculated by comparing the changes in the temperature feature matrix. This can be achieved by calculating the difference in temperature values ​​at corresponding pixel locations, resulting in a difference matrix representing the temperature change. For example, for the temperature matrices of two adjacent time steps... and ,calculate The difference matrix is ​​obtained, where each element represents the temperature change at the corresponding location. This temperature feature difference value contains information about the change in ambient temperature during the target robot's movement, reflecting the target robot's pose changes from another perspective, because the target robot may perceive different temperature distributions at different locations, such as the temperature rising when close to a heat-generating device and decreasing when moving away.

[0112] In the method of this embodiment, the relative position change matrix obtained by the ICP algorithm represents the relative pose transformation of point cloud data between adjacent frames, thereby obtaining the relative position change between point cloud data, providing an important basis for pose estimation of the target robot, and can calculate the temperature feature difference value to represent the information of environmental temperature change of the target robot during movement, thus improving the accuracy of pose prediction.

[0113] In one exemplary embodiment, such as Figure 6 As shown, step 108 may include steps 602 to 606:

[0114] Step 602: Construct the initial pose prediction function and the initial prediction error covariance function of the target robot in the target space.

[0115] In this context, it is assumed that the target robot has a certain motion model from the previous time step to the current time step. This motion model can be established based on the kinematic principles or empirical data of the target robot. If the target robot is a wheeled inspection robot, its motion model can consider the relationship between factors such as wheel speed and steering angle and the changes in robot position and posture. Based on this motion model, the initial pose prediction function and the initial prediction error covariance function of the target robot in the target space are constructed, as shown in equations (7) and (8). It is a state transition matrix, which describes the transformation relationship of the target robot's pose between different time steps; It is process noise, used to account for uncertainties in the actual motion process, such as wheel slippage, motion deviation caused by uneven ground, etc. Process noise follows a Gaussian distribution; It is the covariance matrix of the process noise, which describes the distribution characteristics of the process noise; This is the initial predicted pose; This represents the initial prediction error covariance.

[0116] Step 604: Based on the relative position change matrix and the temperature feature difference matrix, obtain the target pose prediction function corresponding to the initial pose prediction function, and obtain the target prediction error covariance function corresponding to the initial prediction error covariance function.

[0117] Step 606: Use the target pose prediction function and the target prediction error covariance function to obtain the target robot's current pose information at the current time step.

[0118] In one exemplary embodiment, such as Figure 7 As shown, step 604, obtaining the target pose prediction function corresponding to the initial pose prediction function based on the relative position change matrix and the temperature feature difference matrix, may include steps 702 to 710. Wherein:

[0119] Step 702: Based on the temperature feature difference matrix, obtain the temperature observation residual of the temperature feature difference matrix for the initial pose prediction function.

[0120] In this step, the main function of the temperature matrix data is to provide environmental feature localization. Based on the temperature feature difference matrix, the corresponding temperature observation values ​​can be obtained. Then, based on the temperature observation values ​​and the initial predicted pose, the temperature observation residual of the temperature feature difference matrix with respect to the initial pose prediction function can be obtained, as shown in equation (9). The temperature matrix provides pose information (reflecting pose changes through differences in environmental characteristics). The observation matrix maps the state space (robot pose state) to the observation space (temperature feature space). The determination of the observation matrix depends on the specific relationship model between temperature features and robot pose. For example, if the temperature features are directly related to the robot's position in a certain direction, then... The corresponding element will reflect this relationship; This is the temperature observation residual. Furthermore, based on the temperature observation residual, the first optimized pose prediction function corresponding to the initial pose prediction function can be obtained.

[0121] Step 704: Based on the temperature observation residual, obtain the first optimized pose prediction function corresponding to the initial pose prediction function.

[0122] Step 706: Based on the relative position change matrix, obtain the spatial observation residual of the relative position change matrix for the initial pose prediction function.

[0123] In this step, spatial observations can be obtained based on the relative position change matrix, and then the spatial observation residuals can be obtained based on the spatial observations, as shown in equation (13):

[0124]

[0125] in, This refers to the residual of the space observation; The relative pose information is calculated by ICP of the point cloud and used as an external pose observation for mapping updates. This step is similar to the calculation of the residual of the temperature matrix observation. By comparing the pose calculated by the point cloud ICP with the current pose estimate updated based on the temperature matrix, the difference between the two is obtained, providing a basis for subsequent fusion of point cloud observation information.

[0126] Step 708: Based on the spatial observation residual, obtain the second optimized pose prediction function corresponding to the initial pose prediction function.

[0127] Step 710: Obtain the target pose prediction function corresponding to the initial pose prediction function based on the first optimized pose prediction function and the second optimized pose prediction function.

[0128] In one exemplary embodiment, step 704 may include:

[0129] Obtain the temperature Kalman gain of the temperature feature difference matrix for the initial pose prediction function; based on the temperature observation residual and the temperature Kalman gain, obtain the first optimized pose prediction function corresponding to the initial pose prediction function.

[0130] Specifically, based on the uncertainty of temperature in predicting the pose of the target robot, a Kalman filter can be used to obtain the temperature Kalman gain, as shown in equation (10), where, This is the covariance matrix of the temperature observation noise, used to describe the error distribution characteristics of the temperature observation data; For temperature Kalman gain, This reflects the system's level of confidence in the temperature matrix observation at this step. Furthermore, the initial pose prediction function can be updated using the temperature observation residual and the temperature Kalman gain to obtain the first optimized pose prediction function, as shown in equation (11).

[0131] In one exemplary embodiment, step 708 may include:

[0132] Obtain the spatial Kalman gain of the relative position change matrix for the initial pose prediction function; based on the spatial observation residual and the spatial Kalman gain, obtain the second optimized pose prediction function corresponding to the initial pose prediction function.

[0133] Specifically, based on the uncertainty of the pose prediction of the target robot using point cloud data, a Kalman filter can be used to obtain the spatial Kalman gain, as shown in equation (14), where, Measurement noise for point cloud ICP is used to describe the error characteristics of point cloud ICP observation data. The spatial Kalman gain reflects the system's confidence in the point cloud ICP observation data under the current pose estimation. It is similar to the Kalman gain calculation when observing the temperature matrix, but takes into account the influence of point cloud ICP measurement noise.

[0134] Furthermore, in one possible implementation, the initial pose prediction function can be updated using the spatial observation residual and the spatial Kalman gain to obtain the second optimized pose prediction function, as shown in equation (15); subsequently, the target pose prediction function can be obtained based on the first and second optimized pose prediction functions, as shown in equation (17). In another possible implementation, the first optimized pose prediction function can be updated using the spatial observation residual and the spatial Kalman gain to obtain the target pose prediction function, as shown in equation (19).

[0135] In an exemplary embodiment, obtaining the target prediction error covariance function corresponding to the initial prediction error covariance function in step 604 may include:

[0136] Obtain the temperature Kalman gain of the temperature feature difference matrix for the initial pose prediction function; based on the temperature Kalman gain, obtain the first optimized prediction error covariance function corresponding to the initial prediction error covariance function; obtain the spatial Kalman gain of the relative position change matrix for the initial pose prediction function; based on the spatial Kalman gain, obtain the second optimized prediction error covariance function corresponding to the initial prediction error covariance function; based on the first optimized prediction error covariance function and the second optimized prediction error covariance function, obtain the target prediction error covariance function corresponding to the initial prediction error covariance function.

[0137] In the method of this embodiment, the iterative nearest point algorithm and Kalman filter can be used to predict the pose of the target robot, which improves the accuracy of pose prediction. In turn, it can further improve the accuracy of localization, the efficiency and accuracy of path planning, and the accuracy of map updates in the corresponding space.

[0138] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0139] Based on the same inventive concept, this application also provides a robot positioning device for implementing the robot positioning method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more robot positioning device embodiments provided below can be found in the limitations of the robot positioning method described above, and will not be repeated here.

[0140] In one embodiment, such as Figure 8 As shown, a robot positioning device is provided, including: a fusion module 802, an encoding module 804, a first calculation module 806, and a second calculation module 808, wherein:

[0141] The fusion module 802 is used to obtain the fusion feature matrix of the current temperature and point cloud of the target robot in the target space at the current time step;

[0142] The encoding module 804 is used to perform feature encoding on the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix.

[0143] The first calculation module 806 is used to obtain the relative position change matrix of the target robot between the previous time step and the current time step based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, and to obtain the temperature feature difference matrix of the target robot between the previous time step and the current time step.

[0144] The second calculation module 808 is used to obtain the target robot's current pose information at the current time step based on the relative position change matrix and the temperature feature difference matrix.

[0145] In one embodiment, the encoding module 804 is further configured to: construct an initial state space model for the target space; obtain feature association parameters of the initial state space model based on the current temperature and point cloud fusion feature matrix; obtain a target state space model for the target space based on the feature association parameters and the initial state space model; and use the target state space model to perform feature encoding on the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix.

[0146] In one embodiment, the first calculation module 806 is further configured to: determine a first point cloud subset in the historical point cloud dataset corresponding to the previous time step; filter out a second point cloud subset that matches the first point cloud subset in the current point cloud dataset corresponding to the current time step based on the historical encoding feature vector and the current encoding feature vector; there is a correspondence between the multiple second point cloud data contained in the second point cloud subset and the multiple first point cloud data contained in the first point cloud subset; and obtain the relative position change matrix based on the difference information between the second point cloud subset and the first point cloud subset.

[0147] In one embodiment, the first calculation module 806 is further configured to: obtain temperature change information of the current temperature feature matrix corresponding to the current time step and the historical temperature feature matrix corresponding to the previous time step based on the historical coding feature vector and the current coding feature vector; and obtain the temperature feature difference matrix based on the temperature change information.

[0148] In one embodiment, the second calculation module 808 is further configured to: construct an initial pose prediction function and an initial prediction error covariance function for the target robot in the target space; obtain a target pose prediction function corresponding to the initial pose prediction function and a target prediction error covariance function corresponding to the initial prediction error covariance function based on the relative position change matrix and the temperature feature difference matrix; and use the target pose prediction function and the target prediction error covariance function to obtain the target current pose information of the target robot at the current time step.

[0149] In one embodiment, the second calculation module 808 is further configured to: obtain the temperature observation residual of the temperature feature difference matrix for the initial pose prediction function based on the temperature feature difference matrix; obtain a first optimized pose prediction function corresponding to the initial pose prediction function based on the temperature observation residual; obtain the spatial observation residual of the relative position change matrix for the initial pose prediction function based on the relative position change matrix; obtain a second optimized pose prediction function corresponding to the initial pose prediction function based on the spatial observation residual; and obtain the target pose prediction function corresponding to the initial pose prediction function based on the first optimized pose prediction function and the second optimized pose prediction function.

[0150] In one embodiment, the second calculation module 808 is further configured to: obtain the temperature Kalman gain of the temperature feature difference matrix for the initial pose prediction function; and obtain the first optimized pose prediction function corresponding to the initial pose prediction function based on the temperature observation residual and the temperature Kalman gain.

[0151] In one embodiment, the second calculation module 808 is further configured to: obtain the spatial Kalman gain of the relative position change matrix for the initial pose prediction function; and obtain the second optimized pose prediction function corresponding to the initial pose prediction function based on the spatial observation residual and the spatial Kalman gain.

[0152] In one embodiment, the second calculation module 808 is further configured to: obtain the temperature Kalman gain of the temperature feature difference matrix for the initial pose prediction function; obtain a first optimized prediction error covariance function corresponding to the initial prediction error covariance function based on the temperature Kalman gain; obtain the spatial Kalman gain of the relative position change matrix for the initial pose prediction function; obtain a second optimized prediction error covariance function corresponding to the initial prediction error covariance function based on the spatial Kalman gain; and obtain the target prediction error covariance function corresponding to the initial prediction error covariance function based on the first optimized prediction error covariance function and the second optimized prediction error covariance function.

[0153] Each module in the aforementioned robot positioning device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0154] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores robot localization-related data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a robot localization method.

[0155] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0156] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0157] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0158] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0159] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0160] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0161] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0162] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A robot localization method, characterized in that, The method includes: Obtain the fusion feature matrix of the current temperature and point cloud of the target robot in the target space at the current time step; The temperature and point cloud fusion feature matrix is ​​coded to obtain the current coded feature vector of the temperature and point cloud fusion feature matrix; Based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, obtain the relative position change matrix of the target robot between the previous time step and the current time step, and obtain the temperature feature difference matrix of the target robot between the previous time step and the current time step; Construct the initial pose prediction function and the initial prediction error covariance function of the target robot in the target space; Based on the relative position change matrix and the temperature feature difference matrix, obtain the target pose prediction function corresponding to the initial pose prediction function, and obtain the target prediction error covariance function corresponding to the initial prediction error covariance function; Using the target pose prediction function and the target prediction error covariance function, the target robot's current pose information at the current time step is obtained.

2. The method according to claim 1, characterized in that, The step of performing feature encoding on the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix includes: Construct an initial state space model for the target space; Based on the current temperature and the point cloud fusion feature matrix, the feature association parameters of the initial state space model are obtained; Based on the feature association parameters and the initial state space model, obtain the target state space model for the target space; Using the target state space model, feature encoding is performed on the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix.

3. The method according to claim 1, characterized in that, The step of obtaining the relative position change matrix of the target robot between the previous time step and the current time step includes: Determine the first subset of point clouds in the historical point cloud dataset corresponding to the previous time step; Based on the historical encoding feature vector and the current encoding feature vector, a second point cloud subset that matches the first point cloud subset is selected from the current point cloud dataset corresponding to the current time step; there is a correspondence between the multiple second point cloud data contained in the second point cloud subset and the multiple first point cloud data contained in the first point cloud subset; The relative position change matrix is ​​obtained based on the difference information between the second point cloud subset and the first point cloud subset.

4. The method according to claim 1, characterized in that, The step of obtaining the temperature feature difference matrix of the target robot between the previous time step and the current time step includes: Based on the historical coding feature vector and the current coding feature vector, obtain the temperature change information of the current temperature feature matrix corresponding to the current time step and the historical temperature feature matrix corresponding to the previous time step; Based on the temperature change information, the temperature feature difference matrix is ​​obtained.

5. The method according to claim 1, characterized in that, The step of obtaining the target pose prediction function corresponding to the initial pose prediction function based on the relative position change matrix and the temperature feature difference matrix includes: Based on the temperature feature difference matrix, obtain the temperature observation residual of the temperature feature difference matrix with respect to the initial pose prediction function; Based on the temperature observation residual, obtain the first optimized pose prediction function corresponding to the initial pose prediction function; Based on the relative position change matrix, obtain the spatial observation residual of the relative position change matrix with respect to the initial pose prediction function; Based on the spatial observation residual, obtain the second optimized pose prediction function corresponding to the initial pose prediction function; Based on the first optimized pose prediction function and the second optimized pose prediction function, the target pose prediction function corresponding to the initial pose prediction function is obtained.

6. The method according to claim 5, characterized in that, The step of obtaining the first optimized pose prediction function corresponding to the initial pose prediction function based on the temperature observation residual includes: Obtain the temperature Kalman gain of the temperature feature difference matrix for the initial pose prediction function; Based on the temperature observation residual and the temperature Kalman gain, the first optimized pose prediction function corresponding to the initial pose prediction function is obtained.

7. The method according to claim 5, characterized in that, The step of obtaining the second optimized pose prediction function corresponding to the initial pose prediction function based on the spatial observation residual includes: Obtain the spatial Kalman gain of the relative position change matrix with respect to the initial pose prediction function; Based on the spatial observation residual and the spatial Kalman gain, the second optimized pose prediction function corresponding to the initial pose prediction function is obtained.

8. The method according to claim 1, characterized in that, The step of obtaining the target prediction error covariance function corresponding to the initial prediction error covariance function includes: Obtain the temperature Kalman gain of the temperature feature difference matrix for the initial pose prediction function; Based on the temperature Kalman gain, obtain the first optimized prediction error covariance function corresponding to the initial prediction error covariance function; Obtain the spatial Kalman gain of the relative position change matrix with respect to the initial pose prediction function; Based on the spatial Kalman gain, obtain the second optimized prediction error covariance function corresponding to the initial prediction error covariance function; Based on the first optimized prediction error covariance function and the second optimized prediction error covariance function, the target prediction error covariance function corresponding to the initial prediction error covariance function is obtained.

9. A robot positioning device, characterized in that, The device includes: The fusion module is used to obtain the fusion feature matrix of the current temperature and point cloud of the target robot in the target space at the current time step; The encoding module is used to perform feature encoding on the temperature and point cloud fusion feature matrix to obtain the current encoded feature vector of the temperature and point cloud fusion feature matrix; The first calculation module is used to obtain, based on the historical encoded feature vector of the target robot in the previous time step adjacent to the current time step and the current encoded feature vector, the relative position change matrix of the target robot between the previous time step and the current time step, and the temperature feature difference matrix of the target robot between the previous time step and the current time step; The second calculation module is used to construct the initial pose prediction function and the initial prediction error covariance function of the target robot in the target space; obtain the target pose prediction function corresponding to the initial pose prediction function and the target prediction error covariance function corresponding to the initial prediction error covariance function based on the relative position change matrix and the temperature feature difference matrix; and obtain the target current pose information of the target robot at the current time step using the target pose prediction function and the target prediction error covariance function.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1-8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.

Images