Positioning method, device and equipment of mobile robot and mobile robot

Abstract

Embodiments of the application disclose a positioning method and device of a mobile robot, equipment and the mobile robot. The method comprises: acquiring environment perception data and a corresponding local map of the mobile robot; adjusting a matching weight of a corresponding grid in pose matching based on a historical time sequence stability index of each grid in the local map; performing pose matching based on the environment perception data, the local map and the adjusted matching weight to obtain a current pose estimation result and an uncertainty measure of the pose matching; adaptively adjusting search parameters of a pose search space based on the uncertainty measure to update the pose search space; and outputting a target pose based on the current pose estimation result. The application realizes parameter adaptive control and time sequence stability weighting cooperative positioning, and improves the positioning robustness and computing efficiency in a dynamic environment.

Description

Technical Field

[0001] This application relates to the field of mobile robot navigation technology, and in particular to a positioning method, apparatus, device, and mobile robot for a mobile robot. Background Technology

[0002] In autonomous navigation systems for mobile robots, matching and localizing based on environmental perception data and prior maps is a core component for achieving accurate navigation. Existing technologies typically employ algorithms such as correlation scan matching or particle filtering to estimate the robot's pose by comparing sensor observation data with map features.

[0003] However, in practical applications, existing positioning schemes often employ fixed matching parameters such as search range, search step size, and sampling variance, which cannot be dynamically adjusted according to the complexity of the current environment and the matching confidence. This open-loop parameter configuration method leads to wasted computational resources in simple scenarios, while in scenarios with large attitude deviations or degradation, it is prone to convergence failure due to insufficient search space.

[0004] In addition, there are a large number of dynamic obstacles such as pedestrians and forklifts in industrial warehousing and other scenarios. Traditional positioning methods assume that the environment is static and unchanged. The map inconsistency caused by dynamic objects can easily produce false high likelihood regions, which seriously interferes with the accuracy and stability of pose matching. Summary of the Invention

[0005] This application provides a method, apparatus, device, and mobile robot for positioning a mobile robot, in order to solve the problems that fixed positioning parameters make it difficult to balance computational efficiency and convergence, and that dynamic obstacles can easily cause false matching, thereby achieving collaborative positioning based on parameter adaptive control and time-series stability weighting.

[0006] In a first aspect, embodiments of this application provide a method for locating a mobile robot, the method comprising: Acquire environmental perception data and corresponding local maps of the mobile robot; Based on the historical temporal stability index of each grid in the local map, the matching weight of the corresponding grid in pose matching is adjusted. Based on the environmental perception data, the local map, and the adjusted matching weights, the pose matching is performed to obtain the current pose estimation result and the uncertainty measure of the pose matching. The search parameters of the pose search space are adaptively adjusted based on the uncertainty metric to update the pose search space used for pose matching. The target pose of the mobile robot is output based on the current pose estimation result.

[0007] Secondly, embodiments of this application provide a positioning device for a mobile robot, the device comprising: The data acquisition module is used to acquire environmental perception data and corresponding local maps of the mobile robot. The weight adjustment module is used to adjust the matching weight of the corresponding grid in pose matching based on the historical temporal stability index of each grid in the local map. The matching and control module is used to perform pose matching based on the environmental perception data, the local map, and the adjusted matching weights to obtain the current pose estimation result and the uncertainty measure of the pose matching; and to adaptively adjust the search parameters of the pose search space based on the uncertainty measure to update the pose search space used for pose matching. The pose output module is used to output the target pose of the mobile robot based on the current pose estimation result.

[0008] Thirdly, embodiments of this application also provide an electronic device, which includes: One or more processors; Storage device for storing one or more programs. When one or more programs are executed by one or more processors, the one or more processors implement the positioning method for a mobile robot as provided in any embodiment of this application.

[0009] Fourthly, embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the positioning method for a mobile robot as provided in any embodiment of this application.

[0010] The technical solution of this application constructs a closed-loop mechanism of "perception-evaluation-control" by using the uncertainty measure of pose matching as a feedback quantity to adaptively adjust the search parameters of the pose search space. Compared with the open-loop matching mode with fixed parameters in the prior art, this mechanism can dynamically adjust the search range and sampling density according to the real-time matching quality: it automatically shrinks the search space to reduce the computational load when the positioning confidence is high, and automatically expands the search space to improve the global convergence ability when the uncertainty increases, thereby significantly optimizing the balance between computational efficiency and robustness while ensuring positioning accuracy.

[0011] Furthermore, by introducing the historical temporal stability index of local map grids to adjust the matching weights, the statistical characteristics in the time dimension are used to distinguish between static environments and dynamic interference. Grids with long-term stable hits receive higher weights, while the weights of frequently changing dynamic obstacle areas are automatically suppressed. This weighting strategy based on temporal consistency can effectively eliminate spurious high likelihood peaks generated by dynamic objects without relying on additional object detection hardware or complex semantic segmentation algorithms, significantly improving the positioning stability and anti-interference capability of mobile robots in highly dynamic and complex environments such as factories and warehouses.

[0012] Furthermore, by constructing a local map using a robust kernel function and combining it with uncertainty-weighted fusion of multi-source poses, the numerical stability of the gradient optimization process is ensured, avoiding divergence caused by outliers. On the other hand, a smooth transition between laser matching and external positioning sources is achieved, avoiding pose jumps caused by hard switching, and further enhancing the system's continuous and reliable positioning capability in degraded scenarios or when a single sensor fails. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the positioning method for a mobile robot provided in Embodiment 1 of this application; Figure 2 A flowchart illustrating a local map construction method provided in Embodiment 4 of this application; Figure 3 This is a schematic diagram of the structure of a positioning device for a mobile robot provided in Embodiment Six of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 7 of this application. Detailed Implementation

[0014] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present application, not the entire structure.

[0015] Example 1 Figure 1 This is a flowchart illustrating the positioning method for a mobile robot provided in Embodiment 1 of this application, as shown below. Figure 1 As shown, the mobile robot positioning method provided in this embodiment can be applied to the positioning platform of a mobile robot mounted on a device with data processing capabilities, such as a computer. It can be used in conjunction with some application software to achieve a better user experience. Specifically, it can include the following steps: Step 101: Obtain the environmental perception data of the mobile robot and the corresponding local map.

[0016] Specifically, environmental perception data refers to the raw observation information obtained by the mobile robot through real-time detection of its surroundings. This information can take forms including, but is not limited to, two-dimensional laser scanning point clouds, three-dimensional depth images, or ultrasonic ranging data. The local map, on the other hand, is a map segment extracted from a pre-constructed global prior map, centered on the mobile robot's current estimated pose. By extracting a local map instead of using the global map for matching, the computational load of a single matching operation can be significantly reduced, while focusing on effective feature regions around the robot, providing a spatial reference for subsequent pose calculation.

[0017] Step 102: Based on the historical temporal stability index of each grid in the local map, adjust the matching weight of the corresponding grid in pose matching.

[0018] This step aims to address the interference resistance problem in complex dynamic environments. In real-world applications, the environment includes both static structures such as walls and pillars that remain stationary for extended periods, and dynamic obstacles such as pedestrians, forklifts, and temporarily stacked goods that move or appear frequently. If all grids are treated equally, the map inconsistencies caused by dynamic objects can easily generate false high-likelihood matching peaks, leading to positioning drift.

[0019] This embodiment introduces a historical time-series stability index, utilizing statistical characteristics over time to distinguish between static and dynamic features. Specifically, grids that maintain a stable hit state over a long period are assigned a higher matching weight; while grids whose hit state changes frequently and exhibit obvious dynamic characteristics are automatically assigned a lower matching weight.

[0020] This mechanism does not rely on expensive semantic segmentation hardware or complex target detection algorithms. It can effectively suppress dynamic interference based solely on spatiotemporal statistical laws, thereby improving the robustness of the positioning system in unstructured environments.

[0021] Step 103: Perform pose matching based on environmental perception data, local map and adjusted matching weights to obtain the current pose estimation result and the uncertainty measure of pose matching.

[0022] During pose matching, the algorithm projects environmental perception data onto a weighted local map and calculates the robot's current pose by searching for the optimal alignment position.

[0023] It is important to emphasize that the output of this step includes not only the current pose estimation result, but also a crucial intermediate quantity: the uncertainty metric for pose matching. This uncertainty metric is a quantitative representation of the reliability of the current matching result, reflecting the richness of the current environmental features, the sensor noise level, and the convergence quality of the matching algorithm itself.

[0024] For example, in scenarios with degraded features such as long corridors, or when dynamic disturbances are severe, even if the algorithm outputs a mathematically optimal solution, its corresponding uncertainty metric will increase significantly. This metric provides direct feedback for subsequent adaptive control, enabling the system to "sense" its own positioning status.

[0025] Step 104: Adaptively adjust the search parameters of the pose search space based on the uncertainty metric to update the pose search space used for pose matching.

[0026] This is the core feature that distinguishes this embodiment from the traditional open-loop positioning scheme, as it constructs a closed-loop feedback mechanism of "perception-evaluation-control".

[0027] Existing technologies typically employ fixed search ranges, step sizes, and sampling variances. This static configuration is difficult to adapt to changing environments: it can lead to wasted computing power in simple scenarios, while in scenarios with large deviations or degradation, it can easily result in lost positioning due to insufficient search space.

[0028] This embodiment uses the uncertainty measure obtained in step 103 as a feedback quantity to dynamically adjust the search parameters in the next moment or the current iteration process. When the uncertainty measure indicates that the current positioning is highly reliable, the system automatically shrinks the search space to reduce unnecessary candidate pose traversal, thereby improving computational efficiency; conversely, when the uncertainty measure indicates that the positioning is risky, the system automatically expands the search space, increasing the search density and breadth to ensure that the correct pose is recaptured globally.

[0029] This adaptive adjustment strategy achieves a dynamic balance between computing resources and positioning reliability, ensuring the continuous and stable operation of the system under various operating conditions.

[0030] Step 105: Output the target pose of the mobile robot based on the current pose estimation result.

[0031] The final output target pose can be directly used as the localization input for a mobile robot navigation system for path planning, motion control, or map updating. It should be understood that in some implementations, the target pose may be directly equal to the current pose estimation result; while in implementations involving multi-source fusion, the target pose may also be the result of fusing the current pose estimation result with poses from other sensors. Regardless of the output format, the underlying layer benefits from the dynamic disturbance resistance provided by the temporal stability weighting in the aforementioned steps, and the adaptive convergence capability provided by uncertainty feedback control, thereby ensuring high accuracy and high reliability of the output pose in complex environments.

[0032] Example 2 Based on the mobile robot localization method provided in Embodiment 1, this embodiment further elaborates on the uncertainty feedback control mechanism and multi-resolution search strategy.

[0033] As an optional implementation, the uncertainty measure includes at least one of a matching covariance matrix or a matching residual index; the search parameters include at least one of a search range, a search step size, or a sampling variance.

[0034] Specifically, the matched covariance matrix can comprehensively characterize the confidence distribution of pose estimation in each degree of freedom of translation and rotation. Its diagonal elements directly reflect the variance of each dimension, while the off-diagonal elements reveal the coupling relationship between dimensions. The matched residual index (such as weighted root mean square error or robust kernel function value) provides a scalar quality evaluation from the perspective of data fit goodness of fit.

[0035] Using these two types of indicators as feedback quantities enables the system to perceive both the orientation uncertainty caused by geometric degradation and the decline in overall matching quality.

[0036] Correspondingly, the search range determines the upper limit of the algorithm's ability to cope with large deviations, the search step size affects the balance between positioning accuracy and computational cost, and the sampling variance controls the exploration breadth of probabilistic search or optimization initialization. These three types of parameters constitute the core dimensions of the pose search space, and by dynamically adjusting them, fine-grained control over the algorithm's behavior can be achieved.

[0037] It should be understood that although the specific parameter forms listed above are given in this embodiment, other indicators that can characterize the matching confidence, such as the information matrix and the Hessian matrix condition number, or other parameters that can characterize the search space characteristics, such as the resolution level and the number of candidate points, can also be used in other embodiments, as long as the adaptive adjustment function based on uncertainty can be achieved.

[0038] Based on this, this embodiment provides a specific adaptive adjustment logic. As an optional implementation, the search parameters of the pose search space are adaptively adjusted based on the uncertainty measure, including: narrowing the search range and sampling variance in response to the uncertainty measure satisfying a first preset condition; and expanding the search range and sampling variance in response to the uncertainty measure satisfying a second preset condition.

[0039] The first preset condition includes: the eigenvalues ​​of the matching covariance matrix are less than a preset eigenvalue threshold, and the matching score of the pose matching is greater than a preset score threshold; the second preset condition includes: the eigenvalues ​​are greater than or equal to the preset eigenvalue threshold, or the matching score is less than or equal to the preset score threshold. This design constructs a complete "perception-evaluation-control" closed loop.

[0040] When the largest eigenvalue of the matching covariance matrix is ​​small and the matching score is high, it indicates that the current environment features are rich, the sensor observations are reliable, and the algorithm converges well. At this time, the system is in a high confidence state, automatically narrowing the search range and sampling variance, which can significantly reduce the number of candidate pose traversals and optimization iterations, thereby greatly improving computational efficiency while ensuring accuracy.

[0041] Conversely, when the feature value increases (indicating geometric degradation or noise interference in a certain direction) or the matching score decreases (indicating environmental changes or excessive initial deviation), the system determines that the current positioning is at risk and immediately expands the search range and sampling variance to increase the coverage and exploration density of the search space, preventing it from getting stuck in local minima or losing the correct pose due to an excessively narrow search space.

[0042] To more clearly illustrate the necessity of this mechanism, we can compare it with traditional schemes that use fixed search parameters: if the fixed parameters are set too small, the robot is prone to localization divergence when experiencing severe shaking, degradation in long corridors, or relocalization scenarios, as the actual pose may exceed the search boundary; if the fixed parameters are set too large, a large amount of redundant computation will occur during normal stable operation, wasting the limited computing resources of the embedded platform. This embodiment, through real-time feedback adjustment, enables the algorithm to dynamically switch between "efficient mode" and "robust mode," fundamentally solving the contradiction between efficiency and convergence that is difficult to achieve with fixed parameters.

[0043] Furthermore, to complement the aforementioned adaptive parameter adjustment and improve matching performance, this embodiment also introduces a multi-resolution search strategy. As an optional implementation, pose matching is performed based on environmental perception data, a local map, and adjusted matching weights, including: in a low-resolution local map, performing a coarse search based on the search range to obtain a coarse matching pose; in a high-resolution local map, performing a fine search centered on the coarse matching pose and based on the search step size to obtain the current pose estimation result.

[0044] This strategy decomposes pose calculation into two cascaded stages. In the coarse search stage, a low-resolution map is used to abstract and generalize environmental features. Its larger grid size naturally provides stronger noise resistance and a larger capture basin. Combined with the potentially expanded search range after adaptive adjustment, it can quickly lock in an approximately correct initial pose value globally, effectively avoiding mismatches caused by multi-peak likelihood fields.

[0045] In the fine search phase, starting from the coarse matching pose, we switch to a high-resolution map for precise alignment. Since the initial value is close to the true value, even if the search step size is reduced after adaptive adjustment, it can still ensure that the high-precision solution is quickly converged in a local minima.

[0046] This coarse-to-fine layered architecture not only forms a good synergy with the aforementioned uncertainty feedback control, where uncertainty mainly drives the expansion and contraction of the coarse search range, while the fine search step size is finely adjusted and optimized based on the coarse search results, but also achieves an organic unity of global exploration and local development on the same set of distance field data through the joint use of multi-resolution maps, significantly enhancing the overall performance of the positioning system in complex dynamic environments.

[0047] Example 3 This embodiment further elaborates on the matching weight adjustment mechanism based on historical temporal stability. As an optional implementation, the historical temporal stability index is determined by: statistically analyzing the hit rate of each grid in the local map during historical multi-frame positioning; and calculating the historical temporal stability index of each grid based on the hit rate using exponential moving average or time window statistics.

[0048] Specifically, the "hit condition" here refers to whether the mobile robot's environmental perception data (such as laser-scanned point clouds), after being projected onto the local map coordinate system, falls into or is associated with a specific grid cell. To accurately characterize the grid's state stability, the system does not rely solely on the observation results of the current single frame, but maintains a statistical quantity spanning multiple historical moments.

[0049] For example, an exponential moving average algorithm can be used to recursively update the hit status. This algorithm introduces a forgetting factor, making the influence of recent observation data on the indicator higher than that of long-term data. This allows it to both retain the memory of the long-term structure of the environment and respond promptly to real changes in the environment. Alternatively, a fixed-length sliding time window can be used to statistically analyze the frequency or continuity of grid hits within the window. This time-dimensional statistical processing transforms discrete instantaneous observations into continuous state assessments, providing a reliable data foundation for subsequently distinguishing between static background and dynamic disturbances.

[0050] Based on this, the matching weights of the corresponding grid cells in pose matching are adjusted, including: using historical temporal stability indicators as weighting factors and combining them with the initial likelihood values ​​of each grid cell to obtain the adjusted matching weights. The initial likelihood value usually reflects the degree of geometric fit between the sensor observation and map features at the current moment, but in dynamic environments, relying solely on geometric fit can easily lead to misjudgments.

[0051] This embodiment constructs a spatiotemporal joint weight correction model by integrating stability information in the temporal dimension into the matching evaluation in the spatial dimension. It should be understood that this combination method is not limited to simple linear superposition, but can also be other mathematical operations that can reflect the synergistic effect of the two, as long as it can achieve the purpose of adjusting the proportion of the initial likelihood value in the final matching score using the stability index.

[0052] Furthermore, as a preferred implementation, the historical temporal stability index monotonically increases with the hit consistency of the raster. This means that if a raster is continuously hit by environmental perception data in consecutive historical frames, its corresponding stability index value will gradually increase; conversely, if the raster's hit status exhibits discontinuous, random, or frequently changing characteristics, its stability index will remain at a low level or decay rapidly. This characteristic mathematically guarantees a positive correlation between the index value and the physical properties of the environment, that is, the more physically stable the structure, the higher its corresponding numerical representation.

[0053] In the specific implementation of weight adjustment, the historical time series stability index is mapped through a preset weight function to obtain the weight factor. The weight function is a function that monotonically increases with the historical time series stability index. The weight factor is multiplied by the initial likelihood value to obtain the adjusted matching weight.

[0054] For example, the weighting function can be designed as a sigmoid curve or a piecewise linear function, with the stability index as input and the weighting factor normalized to a specific interval as output. Through multiplication, when the stability index of a grid is high, the weighting factor approaches 1 or even larger, ensuring that the initial likelihood value is fully preserved or enhanced; conversely, when the stability index is low, the weighting factor is significantly less than 1, directly suppressing the initial likelihood value. This nonlinear mapping mechanism ensures the sensitivity of weight adjustment and avoids matching performance oscillations caused by small fluctuations in the index.

[0055] Under this mechanism, the historical time-series stability index of rasters that have been in a stable hit state for a long time approaches the first preset value, while the historical time-series stability index of rasters with frequent changes in hit state approaches the second preset value. The first preset value is greater than the second preset value.

[0056] From a physical perspective, static infrastructure such as walls, columns, and fixed shelves are repeatedly and continuously observed during robot operation, and their stability index naturally converges to a higher first preset value, thus gaining a dominant weight contribution in pose matching. On the other hand, dynamic obstacles such as pedestrians, forklifts, and temporarily stacked pallets have constantly changing positions over time. Even if they happen to coincide with a certain place on the map in the current frame, their hit records in historical time are sparse and discontinuous. Therefore, their stability index always hovers around a lower second preset value, resulting in their weight being automatically and significantly weakened in the matching calculation.

[0057] The core advantage of this design lies in the fact that it relies entirely on low-cost spatiotemporal statistical laws to distinguish between moving and static objects. It does not require additional deployment of visual sensors such as cameras, nor does it require running computationally intensive deep learning semantic segmentation models. It can effectively suppress dynamic interference on embedded platforms, perfectly demonstrating the balance between algorithm robustness and engineering implementation costs.

[0058] It should be understood that although this embodiment uses exponential moving average and monotonically increasing function as examples, other time-series state estimation algorithms such as Kalman filtering and Bayesian estimation can also be used in other embodiments, or other mapping methods such as table lookup and polynomial fitting can be used. As long as the function of dynamically adjusting the matching weight based on historical hit consistency can be achieved, they all fall within the protection scope of this application.

[0059] Example 4 This embodiment further specifies the method for constructing the local map. As an optional implementation, the method for constructing the local map includes: taking the estimated pose of the mobile robot as the center, extracting a local tile map from the globally occupied grid map; calculating the distance from each grid in the local tile map to the nearest obstacle; and mapping the distance to a likelihood value using a robust kernel function to construct the local map; wherein the first derivative of the robust kernel function with respect to the distance is continuous and bounded throughout the entire domain.

[0060] Specifically, please refer to Figure 2 , Figure 2 This is a flowchart illustrating a local map construction method provided in Embodiment 4 of this application. Figure 2 As shown, there are 5 modules: current pose acquisition module 201, local tile extraction module 202, distance field calculation module 203, robust kernel likelihood mapping module 204, and multi-resolution downsampling module 205.

[0061] The current pose acquisition module is used to acquire the current robot pose; the local tile extraction module uses the current pose as the center and crops local tiles of a fixed size in the global map; the distance field calculation module is used to initialize the grid distance to 0, scan forward / backward twice to obtain the distance d from the grid to the nearest obstacle; the robust kernel likelihood mapping module is used to apply a robust kernel such as Cauthy to each distance d to obtain the robust distance field likelihood; and the multi-resolution downsampling module performs multiple downsamplings on the high-resolution robust distance likelihood map.

[0062] Specifically, during the construction process, when performing local tile extraction, the system determines a rectangular region of a preset size as the local tile map based on the estimated pose of the mobile robot in the pre-stored global occupancy grid map. For example, a square region with a side length of 20 to 50 meters centered on the robot can be extracted.

[0063] Compared to directly using a global map, this dynamic capture strategy can limit the amount of data processing for subsequent distance field calculations and pose matching to a constant range, significantly reducing the memory usage and computing power requirements of the embedded computing platform, while ensuring that the algorithm only focuses on local environmental features around the robot that have immediate navigation value.

[0064] After acquiring the local tile map, the system calculates the distance from each grid cell to the nearest obstacle. In practice, the local tile map is first initialized by traversing it, setting the distance values ​​of all obstacle grid cells marked as occupied to 0, and setting the distance values ​​of all free grid cells to infinity or a sufficiently large preset maximum value.

[0065] Subsequently, a distance transformation algorithm is used to perform two passes on the map: a forward scan and a backward scan. During the forward scan, each grid cell updates its own distance based on the distance values ​​of its left, top, and top-left adjacent grid cells; during the backward scan, it updates its distance in reverse based on the distance values ​​of its right, bottom, and bottom-right adjacent grid cells.

[0066] Through these two linear-time scans, each free grid cell obtains its precise or approximate Euclidean distance to the nearest obstacle grid cell. This process transforms discrete occupied grid information into a continuous distance field representation, laying the data foundation for subsequent smooth matching.

[0067] Based on this, this embodiment maps distance to likelihood values ​​using a robust kernel function. As a preferred specific example, a Cauchy-type robust kernel function can be used for mapping, and its mathematical expression is as follows: w_{dist}(d) = \frac{1}{1 + \left( \frac{d}{\sigma} \right)^{2}}; Where d represents the distance from the grid to the nearest obstacle, and σ is a scale parameter used to control the rate at which the likelihood value decays with distance. Its value is usually related to the standard deviation of the ranging noise of the lidar or the map resolution. This function non-linearly maps the distance value to the interval [0, 1], with a higher likelihood value the closer the distance is to the nearest object and a lower likelihood value the farther the distance is to the nearest object.

[0068] It should be understood that although this embodiment uses the Cauchy kernel as an example for detailed description, in other embodiments, the Tukey double-weighted kernel, Huber kernel, or other robust kernel functions that satisfy the bounded derivative property can also be used, as long as they can realize the transformation of geometric distance into matching cost with good numerical properties.

[0069] The key technical reason for using the robust kernel function described above in this embodiment is that its first derivative with respect to distance is continuous and bounded throughout the entire domain. Taking the derivative of the Cauchy kernel function, its first derivative expression is: ; As can be seen from the derivative formula, when the distance d approaches 0 (i.e., close to the obstacle surface), the derivative value approaches 0 rather than infinity; as the distance d increases, the absolute value of the derivative first increases and then decreases, always remaining within a finite range. This mathematical property is crucial for gradient-based pose matching optimization.

[0070] In traditional Gaussian likelihood fields or binary occupied grid models, the gradient at the boundary of obstacles is often extremely steep or even discontinuous. This can cause optimization algorithms to easily overflow or oscillate when approaching the true value due to excessively large gradients, or fail to converge due to getting trapped in sharp local minima.

[0071] The robust distance field likelihood map constructed in this embodiment provides a gentle and bounded gradient near obstacles, effectively avoiding numerical instability caused by gradient spikes. At the same time, in regions far from obstacles, the function still retains some gradient information and will not decay to zero prematurely like a truncated kernel function.

[0072] This means that even if the initial pose estimation of the mobile robot has a large deviation, the optimization algorithm can still gradually approach the true pose along the effective gradient direction, thereby significantly expanding the convergence radius of the iterative optimization and enhancing the system's ability to resist interference from outlier observation data such as laser flying points and glass reflections.

[0073] Furthermore, to complement the multi-resolution search strategy in the aforementioned embodiments, this embodiment includes a multi-resolution downsampling step after constructing the high-resolution local map. Specifically, by performing multiple fixed-ratio downsampling operations on the high-resolution robust distance field likelihood map, a series of low-resolution distance field likelihood maps with progressively decreasing resolution are generated.

[0074] These maps of different resolutions collectively form an image pyramid structure. Low-resolution maps, due to their larger grid size, naturally exhibit stronger blurring effects and larger capture basins, making them suitable for rapidly performing coarse searches over large areas to determine initial pose values. High-resolution maps, on the other hand, retain rich environmental details, making them suitable for performing fine searches and gradient optimization based on the coarse search results to obtain high-precision poses. This multi-resolution joint use mechanism allows the same set of range field data to simultaneously meet the needs of global exploration and local refinement, further improving the adaptability and computational efficiency of the localization algorithm under different operating conditions.

[0075] Example 5 This embodiment further details the adaptive fusion mechanism for multi-source poses. As an optional implementation, the method also includes: acquiring the external pose output by an external positioning source and the corresponding external uncertainty.

[0076] Specifically, external positioning sources refer to positioning systems independent of the primary environmental perception sensors (such as LiDAR). These systems can take forms including, but are not limited to, ultra-wideband (UWB) positioning base stations, visual simultaneous localization and mapping (V-SLAM) systems, global navigation satellite systems (GNSS), or high-precision odometry fusion modules. External pose provides the robot's absolute or relative position observation in a global coordinate system or a specific reference frame, while external uncertainty quantifies the reliability of this observation.

[0077] In practical engineering implementation, external uncertainty can be directly output from the external sensor interface (such as the positioning accuracy factor returned by the UWB solution engine), or it can be modeled and estimated in real time by the local system based on auxiliary information such as the historical stability, signal strength, or Doppler frequency shift of the external positioning source.

[0078] In addition, since the data update frequency of external positioning sources is usually out of sync with that of environmental perception sensors, after acquiring the external pose, the system needs to perform short-term extrapolation or interpolation based on the timestamp difference and the robot's kinematic model to ensure that it is consistent with the current pose estimation result in terms of spatiotemporal reference.

[0079] Based on this, the first fusion weight and the second fusion weight are calculated using the uncertainty metric of pose matching and external uncertainty. The core of this step lies in transforming abstract uncertainty into quantifiable information contribution.

[0080] In practice, the uncertainty measure of pose matching (such as the matching covariance matrix mentioned in Example 2) and the external uncertainty (such as the external covariance matrix) can be converted into scalar confidence indices. For example, the trace or the largest eigenvalue of each covariance matrix can be calculated, and its reciprocal can be taken as the original confidence level.

[0081] Subsequently, the original confidence scores of both are normalized so that the sum of the first fusion weight and the second fusion weight is 1. When the uncertainty of pose matching is small (i.e., the laser matching quality is high), the first fusion weight automatically increases, and the system trusts the matching results of the environmental perception data more. Conversely, when environmental degradation causes a surge in matching uncertainty, while the external positioning source remains stable, the second fusion weight automatically increases, and the system smoothly transfers control to the external positioning source.

[0082] This dynamic weight allocation mechanism based on real-time uncertainty fundamentally overcomes the pose change problem that traditional fixed weight fusion or threshold hard switching strategies are prone to when sensor performance fluctuates.

[0083] Furthermore, based on the first and second fusion weights, the current pose estimation result is weighted and fused with the external pose to obtain the fused pose; and the fused pose is output as the target pose. In the specific execution of the weighted fusion operation, differentiated interpolation strategies need to be adopted for different components of the pose. For the translation component (x, y coordinates), since it is a vector in Euclidean space, linear interpolation can be directly used, i.e., the fused translation is equal to the weighted sum of the two input translations with their corresponding fusion weights as coefficients. However, for the rotation component (heading angle), due to its periodicity and non-Euclidean nature, simple numerical weighting may lead to angle jumps or non-shortest path problems.

[0084] Therefore, this embodiment preferably employs spherical linear interpolation (Slerp) or shortest angle interpolation algorithms to ensure that the fused rotational posture always lies on the geodesic path between the two input postures, thereby guaranteeing the kinematic continuity and physical rationality of the output posture. The final fused posture not only integrates the advantages of multi-source information but also achieves noise suppression through uncertainty weighting, and can be directly used as the target posture of the mobile robot navigation system for subsequent path tracking and control.

[0085] It is important to emphasize that although the fusion weights calculated based on uncertainty in this embodiment and the search parameters adjusted based on uncertainty in Embodiment 2 both use "uncertainty" as the input variable, their scope and functional nature in the technical architecture are completely different.

[0086] In Example 2, uncertainty feedback control operates on the internal search process of the pose matching algorithm. Its purpose is to adjust the search range, step size, and sampling variance of the next frame or the current iteration step. This belongs to the execution layer adjustment in the "perception-evaluation-control" closed loop, aiming to balance computational efficiency and convergence capability. In contrast, the uncertainty weighted fusion in this example operates on the decision layer of multi-source data output. Its purpose is to allocate confidence levels among different sensor sources to generate the optimal estimation result. This belongs to the state optimization of the information fusion layer, aiming to improve the smoothness and robustness of the output result. One internally regulates the algorithm's behavior, while the other externally integrates observation information. They are independent yet collaborative, together forming a multi-layered defense system for the mobile robot positioning system in complex dynamic environments.

[0087] It should be understood that although this embodiment uses the reciprocal of the trace of the covariance matrix as an example to illustrate the weight calculation method, in other embodiments, information matrix, residual sum of squares or other statistical quantities that can characterize the reliability of data can also be used to construct the weight mapping relationship. As long as the adaptive fusion function based on uncertainty can be achieved, it falls within the protection scope of this application.

[0088] Example 6 Figure 3This is a schematic diagram of a positioning device for a mobile robot provided in Embodiment Six of this application. The positioning device for a mobile robot provided in this embodiment can execute the positioning method for a mobile robot provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the method. This device can be implemented in software and / or hardware, such as... Figure 3 As shown, the positioning device for the mobile robot specifically includes: The data acquisition module 301 is used to acquire environmental perception data of the mobile robot and the corresponding local map. The weight adjustment module 302 is used to adjust the matching weight of the corresponding grid in pose matching based on the historical temporal stability index of each grid in the local map. The matching and control module 303 is used to perform pose matching based on the environmental perception data, the local map and the adjusted matching weights, to obtain the current pose estimation result and the uncertainty measure of the pose matching; and to adaptively adjust the search parameters of the pose search space based on the uncertainty measure to update the pose search space used for pose matching. The pose output module 304 is used to output the target pose of the mobile robot based on the current pose estimation result.

[0089] Example 7 Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 7 of this application, as shown below. Figure 4 As shown, the electronic device includes a processor 410, a memory 420, an input device 430, and an output device 440; the number of processors 410 in the electronic device can be one or more. Figure 4 Taking a processor 410 as an example; the processor 410, memory 420, input device 430, and output device 440 in the electronic device can be connected via a bus or other means. Figure 4 Taking the example of a connection between China and Israel via a bus.

[0090] The memory 420, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the mobile robot positioning method in this embodiment of the invention. The processor 410 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 420, thereby implementing the aforementioned mobile robot positioning method. Acquire environmental perception data and corresponding local maps of the mobile robot; Based on the historical temporal stability index of each grid in the local map, the matching weight of the corresponding grid in pose matching is adjusted. Based on the environmental perception data, the local map, and the adjusted matching weights, the pose matching is performed to obtain the current pose estimation result and the uncertainty measure of the pose matching. The search parameters of the pose search space are adaptively adjusted based on the uncertainty metric to update the pose search space used for pose matching. The target pose of the mobile robot is output based on the current pose estimation result.

[0091] The memory 420 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on terminal usage. Furthermore, the memory 420 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 420 may further include memory remotely located relative to the processor 410, which can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0092] Example 8 This embodiment provides a mobile robot that innovatively integrates the localization algorithm described in the previous embodiments into a physical entity to achieve autonomous navigation in complex dynamic environments. The mobile robot includes a mobile chassis, environmental perception sensors, and electronic devices.

[0093] The mobile chassis serves as the motion carrier and physical foundation for mobile robots, supporting various sensors and computing units and providing mobility in the operating environment. Specifically, the form of the mobile chassis can be selected in various ways to meet the needs of the actual application scenario.

[0094] For example, in flat warehousing and logistics scenarios, mobile chassis can adopt differential drive or omnidirectional wheel drive wheeled chassis to achieve higher mobility and control precision; in unstructured terrain or outdoor scenarios, mobile chassis can also adopt tracked chassis or multi-legged bionic chassis to enhance obstacle crossing ability and environmental adaptability.

[0095] In addition, mobile chassis typically integrate onboard sensors such as motor encoders and inertial measurement units, which can provide real-time odometer increment information for the positioning system. This serves as an important auxiliary input for pose prediction and initial estimation, complementing external environment perception data and further enhancing the continuity and robustness of the positioning system.

[0096] Environmental perception sensors are mounted on the mobile chassis to collect environmental perception data. Specifically, the installation position of the environmental perception sensors should ensure that their field of view covers key environmental features in the robot's direction of travel, and avoid being obstructed by the chassis structure itself.

[0097] For example, environmental sensing sensors can be mounted on the top center or front upper side of the mobile chassis to obtain a wide horizontal scanning field of view. Regarding sensor type selection, environmental sensing sensors can include, but are not limited to, one or more combinations of 2D LiDAR, 3D LiDAR, depth cameras, or ultrasonic sensor arrays.

[0098] Taking a two-dimensional lidar as an example, it can scan the surrounding environment at a high frequency and output point cloud data containing distance and angle information, providing accurate geometric observation input for pose matching in the aforementioned embodiments. It should be understood that although this embodiment uses lidar as the main example, in other embodiments, visual sensors or other ranging sensors can also be used, as long as they can acquire environmental observation data for matching with the local map. The data acquisition frequency of the environmental perception sensor should match the processing power of the electronic device to ensure real-time positioning updates.

[0099] The electronic device is the same as described in the foregoing embodiments, and it is communicatively connected to the environmental perception sensor. Specifically, the electronic device here is the hardware entity that stores and executes the positioning program as described in Embodiment 7, which typically takes the form of an embedded industrial control computer, an edge computing box, or a dedicated AI computing module inside the mobile robot.

[0100] Communication connectivity is a crucial link ensuring the real-time operation of the positioning closed loop. Specific implementation methods can include Ethernet, CAN bus, USB serial port, or wireless communication protocols. Through this communication connection, raw environmental perception data collected by environmental perception sensors is transmitted with low latency to the electronic device. The processor then executes the complete positioning process, including timing stability weight adjustment, uncertainty feedback search parameter adaptive control, and ultimately outputs a high-precision target pose. This target pose can then be sent to the motion controller of the mobile chassis for path tracking and obstacle avoidance decisions.

[0101] This integrated hardware and software design enables the aforementioned abstract positioning method to be implemented in the physical world, giving mobile robots the practical ability to maintain stable positioning in complex working conditions such as human-machine interaction and long corridor degradation. It effectively solves the engineering pain point that traditional mobile robots are prone to losing positioning or drifting in dynamic environments.

[0102] Example 9 To more clearly illustrate the effectiveness of the technical solution in practical applications, a detailed explanation is provided below using a human-machine co-operation logistics warehouse as a typical application scenario. In this scenario, mobile robots need to perform autonomous handling tasks in a complex environment with dense shelving, narrow aisles, and frequent passage of forklifts and pedestrians.

[0103] This environment poses a severe challenge to the positioning system: on the one hand, the frequent appearance of dynamic obstacles can easily generate false high-likelihood areas in traditional static map matching, causing the positioning results to drift with the dynamic objects; on the other hand, the long straight corridor areas commonly found in warehouses are prone to laser matching degradation due to their simple and repetitive geometric features, resulting in positioning loss or significant jitter.

[0104] In addition, when the robot moves between different areas, external auxiliary positioning signals (such as UWB) often experience intermittent occlusion or recovery, which can easily cause sudden changes in the output pose if not handled properly.

[0105] To address the aforementioned dynamic interference problem, this application solves it through a weight adjustment mechanism based on historical temporal stability as described in the preceding embodiments. Specifically, when a forklift or pedestrian moves frequently along the robot's path, the hit status of the grid they occupy in historical frames exhibits a high degree of discontinuity.

[0106] According to the exponential moving average update strategy described in Example 3, the historical temporal stability indices of these grids rapidly decay and approach a lower second preset value. During pose matching, the system maps these low-stability indices to smaller weight factors using a monotonically increasing weight function and multiplies them by the initial likelihood value, thereby automatically reducing the contribution of dynamic regions to the overall matching score.

[0107] In contrast, static structures such as walls, columns, and fixed shelves, due to their long-term stable hits, maintain a high stability index near the first preset value, thus gaining dominant matching weights. This adaptive weighting method based on spatiotemporal statistical characteristics allows the positioning algorithm to fundamentally ignore short-term dynamic interference sources, maintaining stable and accurate positioning results in highly dynamic environments with mixed human and machine traffic without relying on additional visual semantic segmentation hardware.

[0108] For scenarios with degraded features, such as long corridors, this application utilizes an adaptive search parameter mechanism with uncertainty feedback control to ensure the continuity of localization. When a mobile robot enters a long, straight corridor, due to the severe insufficiency of environmental geometric constraints in the direction perpendicular to the corridor, the eigenvalues ​​of the covariance matrix obtained from pose matching will increase significantly in that direction, or the matching score will decrease due to a lack of effective feature support.

[0109] At this point, the system determines that the uncertainty measure meets the second preset condition and immediately triggers a defensive adjustment strategy, automatically expanding the search range and sampling variance for the next time step. This operation essentially expands the algorithm's global exploration capability, increases the probability of recapturing the correct pose in weak feature environments, and effectively avoids localization divergence or loss due to an excessively narrow fixed search space.

[0110] Once the robot leaves the corridor and re-enters the feature-rich area such as the shelving section, the matching quality improves, the uncertainty metric meets the first preset condition, and the system immediately narrows the search range and sampling variance, switching back to high-efficiency mode. This dynamic parameter adjustment closed loop driven by environmental feature quality perfectly balances robustness in degraded scenarios with computational efficiency in normal scenarios.

[0111] To address the issue of fluctuations in external positioning source signals, this application employs a multi-source pose adaptive fusion mechanism to achieve a smooth transition in the output results. For example, when the robot is deep within a shelf and the laser matching quality temporarily decreases due to obstruction, while the external UWB signal recovers and has high accuracy, the system does not simply perform a hard switch of the signal source. Instead, it calculates the relative relationship between the laser matching uncertainty and the external uncertainty in real time.

[0112] According to the fusion weight calculation logic described in Example 5, since the uncertainty of laser matching is large while the external uncertainty is small, the second fusion weight will automatically increase, and the first fusion weight will decrease accordingly. Subsequently, the system performs weighted fusion of the current pose estimation result and the external pose, wherein the translation component uses linear interpolation and the rotation component uses spherical linear interpolation.

[0113] This soft fusion strategy based on real-time confidence ensures that even with drastic fluctuations in the performance of a single sensor, the final output target pose remains kinematically continuous and smooth, completely eliminating the trajectory jump phenomenon caused by the traditional threshold switching method.

[0114] In summary, in the complex application scenario of human-machine co-operation in logistics warehouses, the localization method provided in this application constructs a multi-layered, comprehensive defense system through the coordinated operation of three core mechanisms: temporal stability weighting, uncertainty feedback search control, and multi-source adaptive fusion. This system not only effectively suppresses false matching caused by dynamic obstacles but also overcomes the risk of convergence failure due to feature degradation and ensures output smoothness during multi-source information switching.

[0115] Practical applications show that mobile robots using this solution have significantly reduced localization drift rate and greatly improved relocation success rate in such scenarios. They also maintain excellent real-time response performance on the same computing power platform, fully verifying the creative value and practicality of the technical solution in solving practical engineering pain points.

[0116] It is worth noting that in the embodiments of the above-mentioned device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of this application.

[0117] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

Claims

1. A positioning method for a mobile robot, characterized in that, The method includes: Acquire environmental perception data and corresponding local maps of the mobile robot; Based on the historical temporal stability index of each grid in the local map, the matching weight of the corresponding grid in pose matching is adjusted. Based on the environmental perception data, the local map, and the adjusted matching weights, the pose matching is performed to obtain the current pose estimation result and the uncertainty measure of the pose matching. The search parameters of the pose search space are adaptively adjusted based on the uncertainty metric to update the pose search space used for pose matching. The target pose of the mobile robot is output based on the current pose estimation result.

2. The method according to claim 1, characterized in that, The uncertainty measure includes at least one of a matched covariance matrix or a matched residual index; The search parameters include at least one of the following: search range, search step size, or sampling variance.

3. The method according to claim 2, characterized in that, The search parameters for adaptively adjusting the pose search space based on the uncertainty metric include: In response to the uncertainty measure satisfying a first preset condition, the search range and the sampling variance are narrowed; In response to the uncertainty measure satisfying the second preset condition, the search range and the sampling variance are expanded; The first preset condition includes: the eigenvalue of the matching covariance matrix is ​​less than a preset eigenvalue threshold, and the matching score of the pose matching is greater than a preset score threshold; the second preset condition includes: the eigenvalue is greater than or equal to the preset eigenvalue threshold, or the matching score is less than or equal to the preset score threshold. The step of performing pose matching based on the environmental perception data, the local map, and the adjusted matching weights includes: In the low-resolution local map, a coarse search is performed based on the search range to obtain a coarse matching pose; In the high-resolution local map, a fine search is performed centered on the coarse matching pose and based on the search step size to obtain the current pose estimation result.

4. The method according to claim 1, characterized in that, The methods for determining the historical time series stability index include: The hit status of each grid in the local map during the historical multi-frame localization process was statistically analyzed; Based on the hit situation, the historical time series stability index of each grid is calculated by exponential moving average or time window statistics. The adjustment of the matching weight of the corresponding grid in pose matching includes: The historical time series stability index is used as a weighting factor and combined with the initial likelihood value of each grid to obtain the adjusted matching weight.

5. The method according to claim 4, characterized in that, The historical time series stability index increases monotonically with the hit consistency of the grid. The step of combining the historical time-series stability index as a weighting factor with the initial likelihood value of each grid cell includes: The weighting factor is obtained by mapping the historical time series stability index through a preset weighting function, wherein the weighting function is a function that monotonically increases with the historical time series stability index. The adjusted matching weights are obtained by multiplying the weight factor by the initial likelihood value. Among them, the historical time-series stability index of the grid that has been in a stable hit state for a long time approaches a first preset value, and the historical time-series stability index of the grid that has a frequently changing hit state approaches a second preset value, wherein the first preset value is greater than the second preset value.

6. The method according to claim 1, characterized in that, The methods for constructing the local map include: Centered on the estimated pose of the mobile robot, a local tile map is extracted from the globally occupied grid map; Calculate the distance from each grid cell in the local tile map to the nearest obstacle; The distance is mapped to a likelihood value using a robust kernel function to construct the local map; The first derivative of the robust kernel function with respect to the distance is continuous and bounded throughout the entire domain.

7. The method according to claim 1, characterized in that, The method further includes: Obtain the external pose output by the external positioning source and the corresponding external uncertainty; Based on the uncertainty metric of the pose matching and the external uncertainty, the first fusion weight and the second fusion weight are calculated; Based on the first fusion weight and the second fusion weight, the current pose estimation result and the external pose are weighted and fused to obtain the fused pose; The step of outputting the target pose of the mobile robot based on the current pose estimation result includes: The fused pose is used as the target pose.

8. A positioning device for a mobile robot, characterized in that, The device includes: The data acquisition module is used to acquire environmental perception data and corresponding local maps of the mobile robot. The weight adjustment module is used to adjust the matching weight of the corresponding grid in pose matching based on the historical temporal stability index of each grid in the local map. The matching and control module is used to perform pose matching based on the environmental perception data, the local map, and the adjusted matching weights to obtain the current pose estimation result and the uncertainty measure of the pose matching; and to adaptively adjust the search parameters of the pose search space based on the uncertainty measure to update the pose search space used for pose matching. The pose output module is used to output the target pose of the mobile robot based on the current pose estimation result.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the positioning method of the mobile robot according to any one of claims 1 to 7.

10. A mobile robot, characterized in that, include: Mobile chassis; An environmental perception sensor, mounted on the mobile chassis, is used to collect environmental perception data; as well as, The electronic device of claim 9 is communicatively connected to the environmental sensing sensor.

Images