Positioning abnormity capturing method for intelligent humanoid robot with body

Abstract

The invention discloses a method for capturing positioning abnormity of an intelligent humanoid robot with a body. The method comprises the following steps: acquiring visual information of a scene around the robot; the obtained visual information is input into a pre-trained model, a plurality of features in the scene are extracted, a scene semantic map is constructed according to the extracted feature information, and the scene semantic map comprises each feature and the pose of the feature in the map; acquiring a current positioning pose of the robot, and determining a target feature theoretically recognized by the robot at the current position from the scene semantic map; whether the features which are not recognized at the current positioning posture of the robot exist or not is judged, and the features which are not recognized at the moment under the current positioning posture of the robot exist or not; if yes, determining that positioning is abnormal and executing a positioning correction process; and if not, determining that the positioning is normal. The robot can quickly and accurately capture and position abnormity, and can cope with complex and changeable scenes with high dynamic performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robots, in particular to a body intelligent humanoid robot positioning anomaly capturing method. BACKGROUND

[0002] With the continuous development of technology, the demand for unmanned automation in airport scenarios is increasingly strong. In this context, it is gradually possible to deploy service robots to airport scenarios. Airport scenarios have their own distinctive characteristics. They are large in scale, typical representatives of large scenarios. The movement state of internal personnel and objects is complex and variable, with high dynamics. The scene condition may change at any time, with the property of easy change. Moreover, the safety requirements of airports are extremely high, belonging to high-security scenarios. These characteristics bring great technical challenges (positioning easy to lose) to the automatic driving of service robots, which is the current bottleneck that needs to be broken through. The existing robot positioning adopts sensor data and map matching to determine whether positioning anomalies occur through matching scores. This method cannot be applied to high-dynamic and variable scenarios. SUMMARY

[0003] Therefore, it is necessary to provide a body intelligent humanoid robot positioning anomaly capturing method to solve the problem of easy loss of existing robot positioning.

[0004] To achieve the above purpose, the inventors provide a body intelligent humanoid robot positioning anomaly capturing method, comprising the following steps:

[0005] Obtain visual information of the scene around the robot;

[0006] Input the obtained visual information into a pre-trained model to extract a plurality of features in the scene, and construct a scene semantic map according to the extracted plurality of feature information. The scene semantic map includes each feature and its pose in the map.

[0007] Obtain the current positioning pose of the robot, and determine the target feature that the robot should theoretically recognize at the current position from the scene semantic map;

[0008] Determine whether there are features that should not be recognized at the current positioning pose of the robot, or features that should be recognized at the current positioning pose of the robot but are not recognized at this time;

[0009] If so, determine that the positioning is abnormal and execute a positioning correction process;

[0010] If not, determine that the positioning is normal.

[0011] Further, in the step of extracting a plurality of features in the scene, the following steps are further included:

[0012] Determine whether the number of extracted features is greater than a number threshold;

[0013] If yes, a step of constructing a scene semantic map according to the extracted feature information is entered;

[0014] If no, it is determined whether the human flow is greater than a flow threshold value, if yes, the robot is parked and waits, and after timeout, an exception is reported, if no, a depth technology is used to identify features around the robot, and a single-line laser point cloud density within a certain distance around the robot is calculated;

[0015] It is determined whether the laser point cloud density exceeds a density threshold value;

[0016] If yes, it is judged that the robot is surrounded, and the robot is parked and waits, and after timeout, an exception is reported;

[0017] If no, a step of constructing a scene semantic map according to the extracted feature information is entered.

[0018] Further, when the step of determining whether the human flow is greater than the flow threshold value, the following steps are further included:

[0019] Obtaining pose information of the personnel relative to the robot;

[0020] The field of view range photographed by the camera on the robot is processed by layering according to a preset level;

[0021] In each level, the total width of the box corresponding to the personnel is calculated, and the personnel ratio is calculated in combination with the theoretical maximum field of view width of the camera in this level, and whether the human flow is greater than the flow threshold value is determined according to the personnel ratio.

[0022] Further, in the step of obtaining the pose information of the personnel relative to the robot, the following steps are further included:

[0023] Personnel detection is performed by using a yolov11 algorithm.

[0024] Further, in the step of performing the positioning correction process, the following steps are further included:

[0025] Obtaining at least two features recognized in the current visual information and their observation angles relative to the robot;

[0026] It is determined whether the at least two features are all invariant features;

[0027] If yes, the angle deviation of each feature is calculated according to the current robot pose, the observation angle of each feature, and the standard pose and standard observation angle corresponding to each feature recorded in the scene semantic map, and a weighted average is performed based on the angle deviation to obtain a pose correction amount, and the current robot positioning pose is revised;

[0028] If not, the real-time laser point cloud corresponding to the currently recognized non-invariant feature is extracted, and the laser profile point cloud of the corresponding feature stored in the scene semantic map is compared in similarity, and it is judged whether the similarity of the comparison is greater than or equal to a comparison threshold, if yes, the two pieces of laser point cloud are registered based on the current robot pose and the standard pose of the feature in the scene semantic map, the pose error is obtained, and the current robot positioning pose is corrected by taking the mean value combined with the pose error.

[0029] Further, the scene is an indoor airport;

[0030] The feature that should not be recognized at the current positioning pose of the robot includes a car;

[0031] The target feature that should be recognized at the current positioning pose of the robot is not recognized, and the feature includes a top-hanging sign.

[0032] Further, the several features in the scene include semi-static features, and the semi-static features include stores,

[0033] It is judged whether the distance between the robot and the semi-static feature is less than a distance threshold, if yes, it is determined that the positioning is abnormal and the positioning correction process is executed.

[0034] Further, after the step of determining that the positioning is normal if not, the following steps are further included:

[0035] It is judged whether a new feature is captured;

[0036] If yes, the captured new feature is added to the temporary storage area;

[0037] If not, the running continues;

[0038] It is judged whether a new feature is recognized for more than a preset number of times in multiple tasks:

[0039] If yes, the new feature is updated in the scene semantic map.

[0040] Further, the pre-trained model is a Segment Anything Model or a DeepSeek-VL2.

[0041] To achieve the above-mentioned purposes, the inventors also provide a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the robot positioning abnormality capturing method of any one of the above-mentioned embodiments.

[0042] Unlike the prior art, the above technical solution has the following beneficial effects:

[0043] The robot can quickly and accurately capture positioning abnormalities, trigger correction processes, prevent the robot from losing positioning and moving in the wrong direction, and cope with complex and dynamic scenarios.

[0044] The above summary of the invention is only a summary of the technical solutions of the present application. In order to enable those skilled in the art to more clearly understand the technical solutions of the present application, and then implement the content recorded in the specification and drawings, and in order to enable the above and other purposes, characteristics and advantages of the present application to be more easily understood, the following will be described in combination with the specific embodiments of the present application and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0045] The accompanying drawings are only used to show the principles, implementation manners, applications, characteristics and effects of the specific embodiments of the present application and other related contents, and cannot be considered as a limitation of the present application.

[0046] Figure 1 One of the flowcharts of the robot positioning abnormality capturing method in the present embodiment;

[0047] Figure 2 The second flowchart of the robot positioning abnormality capturing method in the present embodiment;

[0048] Figure 3 The third flowchart of the robot positioning abnormality capturing method in the present embodiment;

[0049] Figure 4 The schematic diagram of the field of view of the photographing in the present embodiment is hierarchically processed according to the preset level;

[0050] Figure 5 The schematic diagram of the signboard in the present embodiment.

[0051] Explanation of reference signs:

[0052] 1, box width,

[0053] 2, signboard. DETAILED DESCRIPTION

[0054] In order to more clearly explain the possible application scenarios, technical principles, specific schemes that can be implemented, and the purposes and effects that can be achieved of the present application, the following will be described in combination with the specific embodiments listed and the accompanying drawings. The embodiments recorded in this paper are only used to more clearly explain the technical solutions of the present application, and therefore only serve as examples, and cannot limit the protection scope of the present application.

[0055] The term "embodiment" is mentioned in this document means that the specific features, structures or characteristics described in connection with the embodiment can be included in at least one embodiment of the present application. The term "embodiment" appearing in various places in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or association with other embodiments. In principle, in this application, as long as there is no technical contradiction or conflict, the technical features mentioned in each embodiment can be combined in any way to form a corresponding implementable technical solution.

[0056] Unless otherwise defined, the meaning of technical terms used in this document is the same as that generally understood by those skilled in the art to which the present application belongs; the use of related terms in this document is only for the purpose of describing specific embodiments, and is not intended to limit the present application.

[0057] In the description of the present application, the phrase "and / or" is a description of the logical relationship between the objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases: A exists, B exists, and A and B exist at the same time. In addition, the character " / " in this document generally represents that the associated objects before and after are a "or" logical relationship.

[0058] In this application, such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual quantity, primary and secondary or order relationship between them.

[0059] Without more limitations, in this application, the use of "include", "contain", "have" or other similar expressions in the sentence is intended to cover non-exclusive inclusion, and these expressions do not exclude the presence of other elements in the process, method or product including the described elements, so that the process, method or product including a series of elements can not only include those limited elements, but also include other elements not explicitly listed, or also include the elements inherent in such process, method or product.

[0060] As the same understanding as in the "Guidelines for Examination", in this application, "greater than", "less than", "exceed" and other expressions are understood as not including the number; "above", "below", "within" and other expressions are understood as including the number. In addition, the meaning of "multiple" in the description of the embodiments of the present application is more than two (including two), and similar expressions related to "multiple" are also understood in this way, for example, "multiple groups", "multiple times" and the like, unless otherwise explicitly limited.

[0061] In the description of the embodiments of the present application, the spatially-related expressions used, such as "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "perpendicular", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", and the like, indicate the orientation or positional relationship shown in the specific embodiments or the drawings, and are only for the convenience of describing the specific embodiments of the present application or for the reader to understand, and do not indicate or imply that the indicated device or component must have a specific position, a specific orientation, or be constructed or operated in a specific orientation, and therefore cannot be understood as a limitation on the embodiments of the present application.

[0062] Unless otherwise explicitly specified or limited, in the description of the embodiments of the present application, the terms "mount", "connect", "connect", "fix", "set", and the like should be understood broadly. For example, the "connection" can be a fixed connection, or a detachable connection, or an integral setting; it can be a mechanical connection, or an electrical connection, or a communication connection; it can be directly connected, or indirectly connected through an intermediate medium; it can be the internal communication of two elements or the interaction relationship between two elements. For those skilled in the art to which the present application belongs, the specific meaning of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0063] Please refer to Figure 1 The embodiment provides a body-aware intelligent humanoid robot positioning anomaly capturing method, comprising the following steps:

[0064] S101, acquiring visual information of a scene around the robot;

[0065] When the robot is executing a task, it continuously collects images of the surrounding environment at a frequency of 10-30 Hz through the camera equipped thereon as the visual information.

[0066] S102, inputting the acquired visual information into a pre-trained model, extracting a plurality of features in the scene, and constructing a scene semantic map according to the extracted feature information, wherein the scene semantic map comprises each feature and its pose in the map;

[0067] The recognized features can include overhead signs, columns, stores, specific billboards, fire hydrants, etc. Based on the extraction result, a scene semantic map is constructed, which not only records the spatial pose of each feature (i.e. its position and orientation in the global coordinate system), but also associates the visual bounding box width and the corresponding laser profile point cloud (collected during the period without personnel interference). The semantic map serves as a priori knowledge base and is stored in the local robot or the cloud server.

[0068] S103, obtaining a current positioning pose of the robot, and determining, from the scene semantic map, a target feature that the robot should theoretically recognize at the current position;

[0069] The robot can obtain a self-considered pose through its original positioning system (such as SLAM based on a laser radar). The robot queries a pre-constructed global scene semantic map to obtain, at the current self-considered pose, features that should theoretically be visible within the camera field of view.

[0070] S104, determining whether there is a feature that should not be recognized at the current positioning pose of the robot or that should be recognized at the current positioning pose of the robot but is not;

[0071] S105, if yes, determining that the positioning is abnormal and performing a positioning correction process;

[0072] S106, if no, determining that the positioning is normal.

[0073] The first type of abnormality: the robot compares and finds that a “top hanging sign” feature appears in the semantic map that it constructs in real time. However, according to the query result, the advertising board should not theoretically be seen at the current position (because it can be in another area 50 meters away). The “top hanging sign” is a “feature that should not be recognized”.

[0074] The second type of abnormality: the robot finds that a sign that should theoretically be seen is not detected in the real-time image, which should be recognized at this time but is not.

[0075] The robot can quickly and accurately capture the positioning abnormality and trigger the correction process, which can prevent the robot from losing positioning and moving in the wrong direction and getting lost, and can cope with complex and variable scenes with high dynamics.

[0076] See Figure 2 In some embodiments, when the step S102 of extracting a plurality of features in the scene is performed, the following steps are further included:

[0077] S1021, determining whether the number of the extracted plurality of features is greater than a number threshold;

[0078] S1022, if yes, proceeding to the step of constructing a scene semantic map according to the extracted plurality of feature information;

[0079] S1023, if no, determining whether the number of people is greater than a flow threshold, S1024, if yes, stopping the robot and waiting, and reporting an exception after timeout, S1025, if no, identifying features around the robot using deep technology and calculating a single-line laser point cloud density within a certain distance around the robot;

[0080] S1026, judging whether the laser point cloud density exceeds a density threshold value;

[0081] S1027, if yes, judging that the robot is surrounded, stopping the robot, waiting, and reporting an exception after timeout;

[0082] S1028, if no, entering a step of constructing a scene semantic map according to the extracted feature information.

[0083] Generally, the number threshold value can be set to 2, 3 or 4, and preferably 2. If the extracted features are too few, the robot can be surrounded. Therefore, the robot can evaluate the surrounding crowd through sensors (such as laser radar or camera).

[0084] If the crowd is very large, the robot is probably surrounded by the crowd, at which time the robot needs to stop and wait, and start a timer. After waiting for a preset 30 seconds (timeout), if the crowd is still very large, an exception can be reported to the server, and the content can be “positioning self-checking failure caused by continuous crowd obstruction”.

[0085] Some interference filtering mechanisms (such as crowd) are introduced to improve the robustness of positioning exception capture. After excluding the reason of large crowd, the robot uses deep technology to identify the features around the robot, and calculates the single-line laser point cloud density within a certain distance (for example, a region with a radius of 1 meter). It is found that the laser point cloud density is very high, exceeding the preset density threshold value. This indicates that the robot is likely to be surrounded by close and static obstacles (such as stacked barriers), resulting in limited vision. The robot judges that it is surrounded, and also executes the stop and wait strategy. After timeout, the robot rotates in place and performs surrounding identification several times. If the situation does not improve after timeout, an exception is reported, and the content can be “surrounded by static obstacles”.

[0086] It is found that the laser point cloud density is normal and does not exceed the threshold value. This indicates that the environment in which the robot is located is a feature sparse area (for example, an empty corridor) itself without obstructions. Although the number of features is small, it is still allowed to enter the step of constructing a scene semantic map according to the extracted feature information, and to attempt to use only a small number of features for positioning verification.

[0087] Please refer to Figure 3 In some embodiments, when judging whether the crowd is greater than the flow threshold value in step S1023, the following steps are further included:

[0088] Step S10231, obtaining pose information of the personnel relative to the robot;

[0089] Step S10232, layering the field of view range captured by the camera on the robot according to a preset level;

[0090] Step S10233, in each level, the staff corresponding to the box width sum, the box width can be Figure 4 The m is shown, and the staff proportion is calculated in combination with the theoretical maximum field of view width of the camera in the level. Whether the flow is greater than the flow threshold is determined according to the staff proportion.

[0091] The robot uses its mounted camera and built-in personnel detection algorithm (such as yolov11, SSD, etc.) to identify each person in the image in real time. For each identified person, the algorithm outputs a box (bounding box) and its approximate distance and direction (pose information) relative to the robot. This pose information can be obtained through stereo vision, personnel box size estimation, or fusion with laser radar data.

[0092] The two-dimensional image captured by the camera is divided into layers according to its depth of field (i.e. distance) according to the preset level. The interlayer spacing (which can be Figure 4 The n is shown) can be set to 0.5m, i.e. each level is 0.5m, 1m, 1.5m, 2m, 2.5m, 3m. Calculate the total length of the box width of each level, and calculate the proportion in combination with the theoretical maximum width of the camera field of view in each level. If it exceeds 70%, it is considered that the flow is large, at which time the navigation is notified to stop and wait, and after the timeout, the exception is reported and the positioning exception capture is not performed.

[0093] In some embodiments, in the steps of performing the positioning correction process, the following steps are further included:

[0094] Obtain at least two features identified in the current visual information and their observation angles relative to the robot;

[0095] Determine whether the at least two features are invariant features, which represent features that do not move for a long time, such as signs, fire sprinklers on the ceiling, and such features are stable in the scene for a long time;

[0096] If so, calculate the angle deviation of each feature according to the current robot pose, the observation angle of each feature, and the standard pose and standard observation angle corresponding to each feature recorded in the scene semantic map, and perform weighted averaging based on the angle deviation to obtain a pose correction amount, and revise the current robot positioning pose;

[0097] If not, the real-time laser point cloud corresponding to the current recognized non-invariant feature is extracted, and similarity comparison is performed between the laser profile point cloud of the corresponding feature stored in the scene semantic map and the extracted laser profile point cloud. It is judged whether the similarity of the comparison is greater than or equal to a comparison threshold (for example, which can be set to 60%, 70% or 80%). If yes, the two pieces of laser point cloud are registered based on the current robot pose and the standard pose of the feature in the scene semantic map, the pose error is obtained, and the current robot positioning pose is corrected by taking the mean value of the pose error.

[0098] The robot positioning pose is sensitive to angle error. At least two features are recognized, the current robot pose and the recognized angle are combined, and the pose and angle of the recognized feature recorded in the scene semantic map are combined to judge the deviation. If the feature is an invariant object, i.e. an invariant feature, the mean value can be revised with a weight. If the feature is not an invariant object, the laser point cloud of the feature is extracted for processing.

[0099] When the laser point cloud of the feature is extracted, the laser point cloud of the feature stored in the semantic map is compared. If the comparison exceeds the comparison threshold (for example, 70%), the two pieces of laser point cloud are registered based on the current pose and the stored pose in the semantic map, the changed pose (error) is obtained, and the pose is corrected by taking the mean value of the pose errors of multiple features.

[0100] The current robot pose, i.e. the robot positioning pose, visually recognizes the pose of the scene feature relative to the robot (Lc);

[0101] When the semantic map is created, the global positioning pose of the feature on the map (map pose Gp) and the pose of the feature relative to the robot (local pose Lp) are recorded. Combined with the current robot positioning pose (Gc), the pose of the current feature relative to the robot is calculated (Lctmp = Lp*Gc / Gp through simple proportional conversion Gc / Gp = Lctmp / Lp);

[0102] The obtained pose is compared (Lc == Lctmp?). If the deviation is large (for example, the direction angle is greater than 15°, which is sensitive to angle), it indicates that the positioning is abnormal.

[0103] If it is judged that the positioning is not abnormal:

[0104] If the feature is an invariant object, such as an airport sign, a fire hydrant, etc., which is basically invariant, the relative pose of the feature and the robot and the calculated angle can be used to obtain the deviation, which is calculated by capturing multiple times (for example, 10 times) to obtain a correction value, and the positioning is revised.

[0105] If the feature is a semi-static feature, such as a shop on both sides of the airport, it is possible to do activities and appear feature changes, at this time further laser point cloud feature matching is needed, and the matching rate exceeds the threshold (such as 70%), which can be used for angle correction; otherwise, no correction is made until the positioning exception state is reached or the feature recognized has error correction.

[0106] In order to improve robustness, when the positioning exception is recognized, a three-level fault is reported first (the robot continues to execute the task), at this time the motion area of the robot is limited, and the specific running speed is controlled at 0.5 m / s, and the single rotation amount does not exceed 5°, and the robot is allowed to run within a window (such as 5m*5m) to perform positioning self-correction (correction includes existing positioning algorithm logic, laser and map matching, and through 2,3 strategies), if the window cannot be corrected successfully, a secondary fault is reported, and manual intervention is needed.

[0107] In some embodiments, the scene is an indoor airport, and the robot is deployed in the indoor terminal area of the airport to perform passenger guidance, cleaning or luggage transportation tasks. The features that should not be recognized at the current positioning pose of the robot include cars, and if the robot is indeed in the indoor motion line area, the car is an absolute feature that should not appear, and is directly determined as a positioning exception - the robot may have misjudged that it is located in an outdoor parking lot due to positioning drift. The target features that should be recognized at the current positioning pose of the robot include top hanging signs 1 (such as shown in FIG. 1), flight information signs, fire hydrants, and service counters. Figure 5

[0108] In some embodiments, a number of features in the scene are extracted, including semi-static features, such as shops, and it is determined whether the distance between the robot and the semi-static feature is less than a distance threshold, and if so, a positioning exception is determined and a positioning correction process is performed.

[0109] Specifically, when the robot extracts a feature, the robot's current positioning pose, camera field of view and scene semantic map are combined to perform the following processing:

[0110] (1) If an unchanging feature (such as a top hanging sign) that should not be recognized appears, a positioning exception is determined;

[0111] (2) If an unchanging feature (such as a top hanging sign) that should be recognized at the current positioning pose is not recognized, it is considered that there may be a positioning exception, at which time the navigation task is paused, and the robot rotates a certain tolerance angle left and right (to prevent being at the edge of the camera field of view), and if it is still not recognized, a positioning exception is determined;

[0112] (3) If a feature (such as a private car, a scene danger zone or an outdoor feature) that should not exist is recognized, a positioning exception is determined (close to an outdoor danger zone), and the model is trained to filter out possible interference (such as electronic screen pictures) in the indoor scene.​

[0113] (4) Recognize the semi-static feature (such as a store), combine the robot's dynamic line area and laser radar data, and assess whether the robot is close to the semi-static feature (such as a store). If so, it is determined that the positioning is abnormal (the robot thinks it is in the dynamic line area, but the visual captures the feature, and the laser radar data captures the distance of the obstacle in the corresponding direction to be very close, so it is determined to be abnormal).

[0114] In some embodiments, after the step S106 of determining that the positioning is normal, the following steps are further included:

[0115] Determine whether a new feature is captured;

[0116] If so, the captured new feature is added to the temporary storage area;

[0117] If not, continue running;

[0118] Determine whether a new feature is recognized for more than a preset number of times in multiple tasks:

[0119] If so, the new feature is updated in the scene semantic map.

[0120] In the case of determining that the pose is normal, a new feature (preferably of a specific type, such as a store or a sign, which can extract the store name or sign identification through OCR technology) is captured. The new feature is added to the temporary storage area of the semantic map. When the feature is recognized for more than a certain number of times in multiple tasks, the map (grid map + scene semantic map) is updated when the robot has no task.

[0121] In some embodiments, the pre-trained model is an artificial intelligence model, preferably a Segment Anything Model or DeepSeek-VL2. The introduction of a large model VLM and scene semantics enables the robot to have human-like eyes and brain, recognize, distinguish, and think about scene information, and identify whether the positioning is abnormal. After the pre-trained model is completed, the picture data (visual information) of the actual application scene is imported for further training. Through training, the model weight is adjusted to make the model more suitable for the scene (e.g., the success rate before adjustment is 70%, and the success rate after adjustment is 90%, but the success rate may decrease in other scenes). Intuitively, fine-tuning of the artificial intelligence model is to "input" more information to the model, and to "optimize" the specific functions of the model. By inputting a data set in a specific field, the model learns the knowledge in that field, thereby optimizing the performance of the large model in NLP tasks in a specific field, such as sentiment analysis, entity recognition, text classification, and dialogue generation.

[0122] The application provides a body intelligent humanoid robot positioning anomaly capturing method. By introducing advanced large artificial intelligence models such as SegmentAnything Model (SAM) and DeepSeek-VL2, various features in the scene can be accurately identified, segmented and extracted, thereby constructing a detailed scene semantic map. At the same time, these large models also give the robot the ability to think intelligently, so that it can capture surrounding environment information in real time through the visual system during operation, combine the decision output of the large model with its own positioning data and semantic map information, effectively identify positioning anomaly conditions and update the map in a timely manner, and ensure the accuracy and reliability of navigation and operation.

[0123] The embodiment also provides a computer readable storage medium storing a computer program, and the computer program is executed by a processor to implement the robot positioning anomaly capturing method of any one of the above embodiments.

[0124] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of the application, the patent protection scope of the application should not be limited. Any technical solutions obtained by replacing or modifying the equivalent structure or equivalent process based on the essential concept of the application, using the content described in the specification and drawings, or directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are all included in the patent protection scope of the application.

Claims

1. A physical intelligence humanoid robot positioning abnormality capturing method characterized by comprising: The method comprises the following steps: acquiring visual information of a scene around the robot; inputting the acquired visual information into a pre-trained model to extract a plurality of features in the scene, and constructing a scene semantic map according to the extracted feature information, wherein the scene semantic map comprises each feature and a pose of each feature in the map; acquiring a current positioning pose of the robot, and determining target features that should be recognized by the robot in theory at the current position from the scene semantic map; determining whether there are features that should not be recognized at the current positioning pose of the robot or features that should be recognized at the current positioning pose of the robot but are not recognized at the current positioning pose of the robot; if yes, determining that the positioning is abnormal and performing a positioning correction process; if no, determining that the positioning is normal.

2. The embodied intelligent humanoid robot positioning anomaly capturing method of claim 1, wherein, In the step of extracting a plurality of features in the scene, the method further comprises the following steps: determining whether the number of the extracted plurality of features is greater than a number threshold; if yes, entering the step of constructing a scene semantic map according to the extracted feature information; if no, determining whether a human flow is greater than a flow threshold, if yes, stopping the robot and waiting, and reporting an exception after timeout, if no, identifying features around the robot by using a depth technology, and calculating a single-line laser point cloud density within a certain distance around the robot; determining whether the laser point cloud density exceeds a density threshold; if yes, judging that the robot is surrounded, stopping the robot and waiting, and reporting an exception after timeout; if no, entering the step of constructing a scene semantic map according to the extracted feature information.

3. The embodied intelligent humanoid robot positioning anomaly capturing method of claim 2, wherein, In the step of determining whether the human flow is greater than the flow threshold, the method further comprises the following steps: acquiring pose information of a person relative to the robot; layering a field of view range photographed by a camera on the robot according to a preset level; in each level, calculating a total width of a box corresponding to the person, and combining a theoretical maximum field of view width of the camera in the level, calculating a proportion of the person, and determining whether the human flow is greater than the flow threshold according to the proportion of the person.

4. The embodied intelligent humanoid robot positioning anomaly capturing method of claim 3, wherein, In the step of acquiring the pose information of the person relative to the robot, the method further comprises the following steps: detecting the person by using a yolov11 algorithm.

5. The embodied intelligent humanoid robot positioning anomaly capture method of claim 1, wherein, In the step of performing the positioning correction process, the method further comprises the following steps: acquiring at least two features recognized in current visual information and observation angles of the at least two features relative to the robot; determining whether the at least two features are all invariant features; if yes, calculating angle deviations of the features according to a current robot pose, the observation angles of the features, and standard poses and standard observation angles of the features recorded in the scene semantic map, and obtaining a pose correction amount by weighted average based on the angle deviations, and revising the current robot positioning pose; if no, extracting real-time laser point clouds corresponding to non-invariant features currently recognized, comparing the real-time laser point clouds with laser contour point clouds of corresponding features stored in the scene semantic map, determining whether a comparison similarity is greater than or equal to a comparison threshold, if yes, registering the two pieces of laser point clouds based on a current robot pose and a standard pose of the feature in the scene semantic map to obtain a pose error, and revising the current robot positioning pose by taking a mean value of the pose error. 6.The somatically intelligent humanoid robot positioning anomaly capture method of claim 1, wherein, The scene is an indoor airport. Features that should not be recognized at the current localization pose of the robot include cars; Features that should be recognized at the current localization pose of the robot but are not recognized include overhead signs. 7.The somatically intelligent humanoid robot positioning anomaly capturing method of claim 1, wherein, Extracting several features in the scene includes semi-static features, Semi-static features include stores, Determining whether the distance between the robot and the semi-static feature is less than a distance threshold, if so, determining that the localization is abnormal and executing a localization correction process. 8.The somatically intelligent humanoid robot positioning anomaly capture method of claim 1, wherein, After the step of determining whether the localization is normal, the method further comprises the following steps: Determining whether a new feature is captured; If so, adding the captured new feature to a temporary storage area; If not, continue running; Determining whether a new feature is recognized for more than a preset number of times in multiple tasks: If so, updating the new feature to the scene semantic map. 9.The somatically intelligent humanoid robot positioning anomaly capture method of claim 1, wherein, The pre-trained model is Segment Anything Model or DeepSeek-VL2.

Images