Integrated positioning and navigation planning method among buildings

Abstract

The embodiment of the invention provides an inter-building integrated positioning and navigation planning method, relates to the technical field of navigation and positioning, is applied to a server in an inter-building navigation planning system, and is characterized in that the server receives an article distribution request which is sent by terminal equipment and is generated based on information of an article to be distributed, an article position and a target position, and sends the article distribution request to the terminal equipment; and determining a target delivery robot. And path planning is carried out according to the current position of the target distribution robot, the article position and the target position, so that the target distribution robot completes article distribution. And meanwhile, the target delivery robot uploads image data obtained in the driving process of the target delivery robot in real time, so that the server dynamically optimizes the driving path, and smooth proceeding of the delivery process is ensured. Through the method, the inter-building navigation planning system can still realize high-precision positioning and reliable navigation under the condition that the GNSS signal is shielded or interfered, so that the distribution efficiency and accuracy are effectively improved, and the requirements of a high-density building environment in a modern city are met.

Description

Technical Field

[0001] The present application relates to the field of navigation and positioning technology, and in particular to a method for planning combined positioning and navigation between buildings. Background Art

[0002] With the acceleration of urbanization and the continuous increase in building density, the role of combined positioning and navigation technology between buildings in modern urban environments is becoming increasingly important. At present, the Global Navigation Satellite System (GNSS) technology is very mature, but in complex environments such as urban buildings and trees, GNSS signals are blocked, making it difficult to obtain high-precision positioning results.

[0003] Among the existing navigation technology optimization methods, combined navigation is booming, such as the combined navigation of inertial navigation system and global positioning system (GPS), the combined navigation of geomagnetism and GPS, the combined navigation of machine vision and GPS, the combined navigation of lidar and QR code landmarks, etc., all of which overcome the positioning limitations in complex environments by combining corresponding sensors from unique directions.

[0004] However, in areas such as dense urban buildings where GPS satellite signals are weak or where GPS positioning errors are subject to large interference, GNSS signals are easily destroyed, resulting in low positioning accuracy. Therefore, providing a technical solution that can obtain high-precision positioning results in complex environments, such as when GNSS signals are blocked, is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The embodiment of the present application provides a combined positioning and navigation planning method between buildings, so as to achieve high-precision positioning and reliable navigation in a complex urban environment even when the GNSS signal is blocked or interfered.

[0006] In a first aspect, an embodiment of the present application provides a combined positioning and navigation planning method between buildings, which is applied to a server in a navigation planning system between buildings, wherein the navigation planning system between buildings also includes a terminal device and a plurality of delivery robots, and the method includes: The server receives an item delivery request sent by the terminal device, wherein the item delivery request is generated based on information of the item to be delivered, the location of the item, and the target location; After receiving the item delivery request, the server determines a target delivery robot from a plurality of delivery robots based on the item location; The server performs path planning based on the location of the target delivery robot, the location of the item, and the target location to obtain a driving path, wherein the driving path includes a first-stage path for loading items and a second-stage path for delivering items; The server sends the driving path to the target delivery robot; The server receives image data uploaded in real time by the target delivery robot during the process of loading and delivering items according to the driving path; The server optimizes the driving path based on the image data and sends the optimized driving path to the target delivery robot.

[0007] In a possible implementation, the server performs path planning based on the location of the target delivery robot, the location of the item, and the target location to obtain a driving path, including: The server performs path planning for the target delivery robot to pick up goods based on the architectural design of the current building, the location of the target delivery robot and the location of the item to obtain the first stage path; The server performs path planning for the delivery process of the target delivery robot based on the architectural design drawing of the current building, the location of the item and the target location to obtain the second stage path; The architectural design drawing includes the location and shape of each building in the building, the entrance and exit of each building, and the internal layout.

[0008] In a possible implementation, the server optimizes the driving path based on the image data, and sends the optimized driving path to the target delivery robot, including: The server processes the image data using a pre-acquired obstacle classification model to determine whether there is an obstacle in front of the target delivery robot, and the type of obstacle and the position information of the obstacle when there is an obstacle; If it is determined that there is an obstacle in front of the target delivery robot, the server uses an obstacle avoidance path planning algorithm to plan an obstacle avoidance path to avoid the obstacle according to the type of obstacle, the position information of the obstacle and the current position information of the target delivery robot; The server optimizes the driving path of the target delivery robot according to the obstacle avoidance path to obtain an optimized driving path.

[0009] In a possible implementation manner, the obstacle classification model includes a first sub-model and a second sub-model; The server processes the image data using a pre-acquired obstacle classification model to determine whether there is an obstacle in front of the target delivery robot, and the obstacle type and position information of the obstacle when there is an obstacle, including: The server inputs the image data into the first sub-model, extracts local features, and optimizes the extracted local features to obtain an optimized feature image; The server inputs the optimized feature image into the second sub-model, performs global feature extraction, and identifies and classifies obstacles based on the extracted global features to obtain identification and classification results; The identification and classification results include: A static obstacle, and position information of the static obstacle; or A dynamic obstacle, and position information of the dynamic obstacle; or No obstacles.

[0010] In a possible implementation, the first sub-model includes a convolution module and a pooling module, the server inputs the image data into the first sub-model, extracts local features, and optimizes the extracted local features to obtain an optimized feature image, including: The server inputs the image data into the convolution module of the first sub-model to extract local features, thereby obtaining local features in the image data; The server inputs the local features into the pooling module of the first sub-model to perform dimension reduction and retain spatial feature processing to obtain the optimized feature image.

[0011] In a possible implementation manner, the second sub-model includes an image segmentation module, a linear embedding module, an encoding module, and a recognition and classification module. The server inputs the optimized feature image into the second sub-model, extracts global features, and recognizes and classifies obstacles based on the extracted global features to obtain recognition and classification results, including: The server inputs the optimized feature image into the image segmentation module of the second sub-model to perform image segmentation to obtain a plurality of image blocks; The server inputs the multiple image blocks into the linear embedding module of the second sub-model for position encoding to obtain an embedding vector; The server inputs the embedding vector into the encoding module of the second sub-model for feature extraction and representation learning to obtain a high-dimensional feature representation; The server inputs the high-dimensional feature representation into the recognition and classification module of the second sub-model for recognition and classification processing to obtain the classification result.

[0012] In a possible implementation, the server uses an obstacle avoidance path planning algorithm to plan an obstacle avoidance path that avoids the obstacle according to the obstacle type, the position information of the obstacle, and the current position information of the target delivery robot, including: The server determines the state space representation and the action space representation of the target delivery robot based on the obstacle type and the position information of the obstacle; The server adopts an obstacle avoidance path planning algorithm based on the state space representation, action space representation and position information of the target delivery robot to obtain the obstacle avoidance path that avoids the obstacle.

[0013] In a possible implementation manner, the server optimizes the driving path of the target delivery robot according to the obstacle avoidance path to obtain an optimized driving path, including: Based on the current position of the target delivery robot and the posture information of the obstacle, the server uses the obstacle avoidance path to replace the local path from the target delivery robot to the area where the obstacle is located in the current driving path, so as to obtain the optimized driving path.

[0014] In a possible implementation, the image data includes: the target delivery robot acquires images of the front area through a laser radar to obtain a three-dimensional point cloud image during driving, and acquires a two-dimensional image by photographing the front area through a camera.

[0015] In a second aspect, an embodiment of the present application provides a building navigation planning system, including: a server, a terminal device, and a plurality of delivery robots; The terminal device is used to obtain information about items to be delivered, item locations, and target locations, and to generate an item delivery request based on the item information, the item locations, and the target locations, and to send the item delivery request to the server; The server is used to execute the above first aspect and / or various possible implementation methods of the first aspect.

[0016] Each delivery robot is equipped with a laser radar and a camera. The laser radar is used to collect three-dimensional point cloud images in front of the vehicle, and the camera is used to collect two-dimensional images in front of the vehicle.

[0017] The combined positioning and navigation planning method between buildings provided in the embodiment of the present application is applied to the server in the navigation planning system between buildings. In this scheme, the server first receives the item delivery request generated by the terminal device based on the information of the items to be delivered, the location of the items and the target location, and determines the target delivery robot by analyzing the location of the items. Then, the path planning is performed according to the current position, the location of the items and the target location of the target delivery robot. The target delivery robot completes the delivery according to the planned path sent by the server, and uploads the image data obtained during its driving process in real time. The server uses these image data to dynamically optimize the driving path to ensure the smooth progress of the delivery process. Through the above method, the navigation planning system between buildings can still achieve high-precision positioning and reliable navigation when the GNSS signal is blocked or interfered, effectively improving the delivery efficiency and accuracy, and adapting to the needs of high-density building environments in modern cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0019] Figure 1 A schematic diagram of a scenario of a combined positioning and navigation planning method between buildings provided in this application; Figure 2 A schematic diagram of a combined positioning and navigation planning method between buildings provided in this application Figure 1 ; Figure 3 A schematic diagram of a combined positioning and navigation planning method between buildings provided in this application Figure 2 ; Figure 4 A schematic diagram of a combined positioning and navigation planning method between buildings provided in this application Figure 3 .

[0020] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0021] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0022] Before introducing the embodiments of the present application, the application background of the embodiments of the present application is first explained as follows: With the rise of spatial information technology and the fifth generation of mobile communication technology, the geographic information industry has developed rapidly under the integration of spatial information technology, such as smart cities, smart transportation, smart medical care, etc., and navigation and positioning are everywhere. At present, GNSS technology is widely used in transportation, surveying and mapping, high-precision time comparison, and earthquake monitoring. However, in complex environments such as urban canyons, tree occlusion, indoor and outdoor areas, GNSS signals are blocked, making it difficult to obtain high-precision positioning results.

[0023] Among the existing navigation technology optimization methods, integrated navigation technology is booming, such as the integrated navigation of inertial navigation system and GPS, the integrated navigation of geomagnetism and GPS, the integrated navigation of machine vision and GPS, the integrated navigation of automatic guided vehicles (AGV) based on magnetic nail technology and inertial navigation, the integrated navigation of AGV based on laser navigation and infrared navigation, the integrated navigation of lidar and QR code landmarks, the integrated navigation of AGV based on inertial and vision, the integrated navigation of multi-eye vision and laser, etc., all overcome the limitations of positioning in complex environments by combining corresponding sensors from a unique direction. However, in areas where GPS satellite signals are weak or GPS positioning errors are greatly interfered, such as dense urban building clusters, GNSS signals are easily destroyed, and the quality of satellite signal data is poor, resulting in low positioning accuracy, poor availability and continuity. Therefore, in complex environments such as road conditions between buildings, achieving accurate navigation and positioning is still challenging.

[0024] To sum up, providing a high-precision navigation and positioning method in complex environments, laying a foundation for further improving the intelligence and autonomy of path planning, and providing accurate and efficient positioning and navigation services for scenarios such as unmanned driving, smart warehousing, and smart parking are technical issues that need to be solved urgently.

[0025] Figure 1 A schematic diagram of a scenario of a combined positioning and navigation planning method between buildings provided in this application, such as Figure 1As shown, the specific application scenario of the present application at least includes: multiple delivery robots 101 for delivering goods between buildings, a server 102, a terminal device 103, etc. After receiving the item delivery request generated based on the information of the item to be delivered, the location of the item and the target location sent by the terminal device 103, the server 102 determines the target delivery robot 101 among the multiple delivery robots, and then plans the driving path according to the location of the target delivery robot 101, the location of the item and the target location and sends it to the target delivery robot 101. The target delivery robot 101 delivers the item according to the driving path sent by the server 102 and uploads the image data collected during the driving process to the server 102 in real time. The server 102 optimizes the driving path based on the image data, that is, whether there are obstacles, and sends the optimized driving path to the target delivery robot 101 to ensure the smooth progress of the delivery process.

[0026] The terminal device 103 is used to send an item delivery request to the server, and can specifically be a local personal computer (PC), laptop, desktop, smart phone, tablet computer, or other electronic device with network connection and processing capabilities. The terminal device can access the relevant delivery service platform system through an installed application or through a web browser. The device is also equipped with a user interface for interacting with the user, and may include an input device (such as a keyboard or touch screen) and an output device (such as a display or speaker) to enhance the user experience.

[0027] The server 102 can be a cloud server, a local server, an edge server, a server cluster, or the like, which has sufficient data processing capabilities and supports efficient communication and data processing with the delivery robot and the terminal device.

[0028] The delivery robot 101 may be an autonomous mobile robot (AMR), a collaborative robot, or the like, equipped with multiple sensors such as a laser radar, a camera, a millimeter-wave radar, and having sufficient load-bearing capacity and reliable safety performance.

[0029] It should be noted that the physical devices involved in the above descriptions are exemplary in the figures and do not represent the only ones. The present disclosure does not specifically limit the specific forms and types of the physical devices involved.

[0030] Combined with the above scenarios, it can be seen that in the prior art, combined navigation technology still faces challenges in achieving accurate navigation and positioning in complex environments such as road conditions between buildings. The combined positioning and navigation planning method between buildings provided in this application is applied to the server in the navigation planning system between buildings. The system uses the server to receive and analyze delivery requests, determine the target delivery robot, and perform path planning. The robot uploads image data in real time during the delivery process, and the server dynamically optimizes the driving path based on the image data. Through the above method, the navigation planning system between buildings can achieve high-precision positioning and reliable navigation in complex urban environments, even when the GNSS signal is blocked or interfered with, effectively improving the delivery efficiency and accuracy, and adapting to the needs of high-density building environments in modern cities.

[0031] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0032] Figure 2 A schematic diagram of a combined positioning and navigation planning method between buildings provided in this application Figure 1 The method is applied to a server in a building navigation planning system, and the building navigation planning system also includes a terminal device and multiple delivery robots, such as Figure 2 As shown, the combined positioning and navigation planning method between buildings specifically includes: S201. The server receives an item delivery request sent by a terminal device. The item delivery request is generated based on information about items to be delivered, item locations, and target locations.

[0033] In this step, the terminal device managed by the distribution platform receives the item distribution request information from the user who has the item distribution demand. The item distribution request information includes: the information of the item to be distributed, the item location, the target location, the priority information, the security verification information and the special requirements. The information of the item to be distributed includes the name, quantity, size, weight, etc. of the item to be distributed; the item location refers to the current location of the item, usually expressed in the form of address or coordinates, which is used to determine the starting point of the distribution; the target location refers to the destination where the item needs to be delivered, usually expressed in the form of address or coordinates, which is used to determine the end point of the distribution; the priority information refers to the distribution platform supports different delivery priorities, such as expedited delivery or standard delivery; the security verification information is used to ensure the security of the request. The verification code or QR code; the special requirements refer to any special instructions or requirements related to the distribution, such as the need for special packaging or temperature control. Through detailed item distribution information, the building navigation planning system can achieve more accurate and efficient distribution services, improve user experience and platform efficiency.

[0034] Specifically, the terminal device managed by the distribution platform sends the item distribution request information of the user who has the item distribution demand to the server, and the server side receives the item distribution request sent by the terminal device. For example, in an actual application scenario between buildings, user A who works at the front desk on the 7th floor of the office area of ​​Building X submits an item distribution request through his terminal device in the mobile application of the distribution platform, and the terminal device managed by the distribution platform sends the item distribution request to the server. The item distribution request information is specifically as follows: the item to be distributed is: an A4 legal contract document, weighing 200 grams; the item location is the location of user A, that is, the front desk on the 7th floor of the office area of ​​Building X; the target location is the legal advisory office in the office area of ​​Building X: Room 1201 on the 12th floor of the office area of ​​Building X; the priority information is marked as expedited delivery; the security verification information indicates that the pickup code is 4321; the special requirements indicate that waterproof protection is required during transportation. The server receives the item distribution request sent by the terminal device and further parses it to provide timely and efficient distribution services.

[0035] S202: After receiving the item delivery request, the server determines a target delivery robot from multiple delivery robots based on the item location.

[0036] In this step, the server determines the target delivery robot from among the multiple delivery robots in the building based on factors such as the location of the item, the idle state of the delivery robot, the remaining power of the delivery robot, and the estimated mileage of the delivery robot. This method ensures efficient use of delivery robot resources and avoids unnecessary waiting and waste of resources.

[0037] Specifically, the server gives priority to robots that are currently idle and close to the item to ensure the fastest response. At the same time, robots with sufficient remaining power and expected mileage are given priority as target robots to ensure that they can complete the entire delivery task without the need for charging or replacement midway, thus avoiding wasting time. Still taking the example of user A who works at the front desk on the 7th floor of the office area of ​​Building X submitting an item delivery request through his terminal device on the mobile application of the delivery platform, after receiving the item delivery request, the server determines the delivery robot A located on the 1st floor of the shopping area, the delivery robot B located on the 13th floor, the delivery robot C located on the 1st floor of the office area, and the delivery robot D located on the 6th floor of the dining area based on the current idle state. Among them, the delivery robots C and D are closer to user A, 65 meters and 40 meters respectively, and the delivery robot C shows that the remaining power is 70%, and it is expected to travel 5 kilometers. The delivery robot D shows that the remaining power is 15%, and it is expected to travel 0.7 kilometers. The distance between user A and the legal advisory office is 1 kilometer. Based on the above factors, the server determines that the target robot is the delivery robot C. This priority selection strategy can ensure the efficient and reliable execution of the delivery task.

[0038] S203. The server performs path planning based on the location of the target delivery robot, the location of the item, and the target location to obtain a driving path, wherein the driving path includes a first-stage path for loading items and a second-stage path for delivering items.

[0039] In this step, the first-stage path refers to the driving path of the delivery robot from the current position to the item location to complete the loading of items. The first-stage path is the path obtained by the server planning the path of the target robot's picking process based on the architectural design drawing of the current building stored in the storage unit, the location of the target delivery robot, and the location of the item, that is, the driving path of the delivery robot C from the current position to the location of user A.

[0040] The second-stage path refers to the driving path of the delivery robot from the location of the item to the target location to complete the delivery of the item. The second-stage path is the path obtained by the server through path planning for the delivery process of the target delivery robot based on the architectural design of the current building stored in the storage unit, the location of the item and the target location, that is, the driving path of the delivery robot C from the location of user A to Room 1201, 12th Floor, Office Area, Building X, Legal Advisory Office.

[0041] The architectural design includes the location and shape of each building in the building, the entrances and exits of each building, and the internal layout. Take the building X where the delivery robot is located as an example. Building X is a large commercial complex that includes multiple functional areas, such as office area, shopping area, and dining area. The architectural design describes the location and shape of multiple functional areas in detail. For example, the office area is located on the north side of the complex, occupying the 1st to 30th floors of the building. The shopping area is located on the south side of the complex, occupying the 1st to 4th floors of the building. The dining area is located on the 5th floor and the negative first floor, with an open layout and a large atrium in the center of the 1st floor. In the architectural design, the entrance and exit locations of each area are also clearly marked. For example, the main entrance and exit of the office area is located on the north side of the building, close to the parking lot, and the entrances and exits of the shopping area and the dining area are distributed on the east and south sides of the building. In addition, the architectural design also details the internal layout of each floor, including the location of corridors, stairs, elevators, and the number and purpose of each room or shop. The detailed layout information of the architectural design enables the server to perform accurate path planning so that delivery can be completed smoothly and efficiently.

[0042] Specifically, still taking the example of user A working at the front desk on the 7th floor of the office area of ​​Building X submitting an item delivery request through his terminal device on the mobile application of the delivery platform, after receiving the item delivery request, the server determines that the delivery robot C among the multiple delivery robots in Building X is the target delivery robot, and performs path planning based on the location of the target delivery robot C, the item location, and the target location to obtain the first-stage path and the second-stage path. In the first-stage path, the server instructs the delivery robot C to pass through the north entrance of the office area, take the No. 5 elevator to the front desk on the 7th floor to load the items. In the second-stage path, the server instructs the delivery robot C to drive along the designated corridor to the No. 3 elevator, take the elevator to the 12th floor, and then drive along the designated direction to Room 1201.

[0043] In a specific implementation of this solution, the server uses a path algorithm to obtain the driving path. In the algorithm, the distances of all nodes in the architectural design of Building X are first initialized to infinity, and the distance of the starting point is zero. Then, the starting point is added to a priority queue, and the node with the smallest current distance is selected for expansion in each step. For each neighbor node of the node, the distance from the starting point through the node to the neighbor node is calculated. If the distance is less than the currently recorded neighbor node distance, the distance of the neighbor node is updated and added to the priority queue. This process is repeated until all nodes have been visited or the shortest path to the target node is determined.

[0044] Specifically, after the target delivery robot C has loaded the items, it needs to reach Room 1201 on the 12th floor of the office area of ​​Building X from the front desk on the 7th floor of the office area of ​​Building X. The path inside the building of Building X can be represented as a weighted graph, where nodes represent different locations and the weight of the edge represents the distance or time of the path. The path planning algorithm first sets the distance of the front desk on the 7th floor of the office area of ​​Building X to zero and the distances of other nodes to infinity. Then, the node with the smallest distance, the front desk on the 7th floor, is selected for expansion, and the distances of its neighboring nodes are updated. For example, if the distance from the front desk on the 7th floor to point Y is 5, and the distance to node Z is 10, the distances between points Y and Z are updated. Next, the next node with the smallest distance is selected from the priority queue for expansion, and the distances of neighboring nodes are continued to be updated. In this way, the server gradually determines the shortest path from the front desk on the 7th floor of the office area of ​​Building X to Room 1201 on the 12th floor of the office area of ​​Building X, and finally instructs the delivery robot to travel along this path to ensure an efficient and smooth delivery process.

[0045] S204. The server sends the driving path to the target delivery robot.

[0046] The server sends the planned first-stage path and second-stage path to the target delivery robot.

[0047] S205. The server receives image data uploaded in real time by the target delivery robot during the loading and delivery process of items according to the driving path.

[0048] In this step, in order to ensure the safety, accuracy and efficiency of the delivery process, the inter-building navigation planning system can continuously monitor the driving environment of the delivery robot by collecting image data in real time and reporting it to the server, so as to timely identify and respond to potential obstacles or emergencies, thereby avoiding accidents. The image data collected by the delivery robot in real time and uploaded to the server include: the target delivery robot collects images of the front area through the laser radar to obtain a three-dimensional point cloud image during the driving process, and the front area is photographed by the camera to obtain a two-dimensional image.

[0049] The delivery robot uses laser radar to collect images of the area in front of it to generate a three-dimensional point cloud image, and uses a camera to capture the area in front of it to obtain a two-dimensional image. The combination of multimodal data provides rich environmental information. The three-dimensional point cloud image provides accurate spatial depth information, which helps to identify and locate obstacles, and the two-dimensional image provides visual details, which enhances the understanding and recognition of the environment. The navigation planning system between buildings uses this multimodal image data to continuously monitor the driving environment of the delivery robot in order to promptly identify and respond to potential obstacles or emergencies, improve the accuracy and response speed of navigation decisions, and effectively avoid accidents, ensuring the safe and efficient operation of the delivery robot.

[0050] S206. The server optimizes the driving path based on the image data and sends the optimized driving path to the target delivery robot.

[0051] In this step, path optimization refers to the process in which the target delivery robot completes loading of items according to the first-stage path issued by the server and completes delivery according to the second-stage path issued by the server. By collecting image data of the front area in real time, it is determined whether there are obstacles in the driving path. When encountering obstacles, the inter-building navigation planning system dynamically adjusts the original first-stage path and the second-stage path to ensure that the target delivery robot can reach the destination safely and efficiently.

[0052] Specifically, the process in which the server optimizes the driving path based on the image data and sends the optimized driving path to the target delivery robot is as follows: S2061: The server uses a pre-acquired obstacle classification model to process the image data to determine whether there is an obstacle in front of the target delivery robot, as well as the obstacle type and obstacle posture information when there is an obstacle.

[0053] In this step, the obstacle classification model is used to analyze and identify obstacles in the image data. Its main function is to determine whether there are obstacles in the area ahead of the target delivery robot by processing the image data collected by the sensors carried by the target delivery robot, and further identify the type and posture information of the obstacles. The types of obstacles include static obstacles and dynamic obstacle areas. Static obstacles refer to fixed objects, such as walls, furniture, or building structures; dynamic obstacles refer to movable objects, such as pedestrians, other vehicles, or animals. Posture information refers to the position and posture of the obstacle in space. The position refers to the coordinates of the obstacle in three-dimensional space, and the posture describes the direction and angle of the obstacle.

[0054] Through the obstacle classification model, the inter-building navigation planning system can analyze potential threats in the environment in real time so as to dynamically adjust the driving path of the target delivery robot to ensure that it can complete the loading and delivery tasks safely and efficiently.

[0055] S2062: If it is determined that there is an obstacle in front of the target delivery robot, the server uses an obstacle avoidance path planning algorithm to plan an obstacle avoidance path to avoid the obstacle based on the obstacle type, the obstacle's posture information, and the target delivery robot's current position information.

[0056] In this step, the obstacle avoidance path planning algorithm is used to calculate and generate a new driving path. When the server determines that there is an obstacle in the target delivery robot, the algorithm will be triggered. Its main purpose is to ensure that the target delivery robot can safely bypass the obstacle and continue to move in the target direction. The obstacle avoidance path is used for the target delivery robot to avoid obstacles, that is, in the current driving path, the local path from the target delivery robot to the area where the obstacle is located. The target delivery robot avoids obstacles through the obstacle avoidance path while maintaining driving efficiency and safety as much as possible. Compared with the driving path in the original obstacle area, the obstacle avoidance path will bypass or avoid the detected obstacles, which may slightly extend the driving distance, but safety is given priority.

[0057] By adopting an obstacle avoidance path planning algorithm, the inter-building navigation planning system can ensure that the target delivery robot can flexibly cope with obstacles in complex and changing environments, avoid collisions, and complete delivery tasks efficiently.

[0058] S2063: The server optimizes the driving path of the target delivery robot according to the obstacle avoidance path to obtain an optimized driving path.

[0059] In this step, the optimized path refers to a new driving path obtained by updating the local path in the area where the obstacle zone is located in the original driving path after the server obtains the obstacle avoidance path through the obstacle avoidance path planning algorithm.

[0060] Specifically, based on the current position of the target delivery robot and the position information of the obstacle, the server uses the obstacle avoidance path to replace the local path of the target delivery robot to the obstacle area in the current driving path, and obtains the optimized driving path. By optimizing the path, the target delivery robot can effectively avoid obstacles, ensure that it can continue to move forward safely and complete the delivery task. The path optimization process not only improves the safety of the target delivery robot, but also improves its adaptability and efficiency in dynamic environments.

[0061] The combined positioning navigation planning method between buildings provided in the embodiment of the present application is intended to improve the accuracy and efficiency of distribution services. The core process of this embodiment includes: the server receives the item distribution request sent by the terminal device, determines the target distribution robot based on the item location and the status of the distribution robot, and performs path planning to obtain the driving path for loading and delivering items, and sends the path information to the target distribution robot. During the distribution process, the server monitors the robot's driving environment through real-time collected image data, identifies obstacles and optimizes the path to ensure that the robot completes the task safely and efficiently. Through the above method, the navigation planning system between buildings can effectively cope with complex and changing environments, ensuring that the robot safely avoids obstacles and successfully completes the distribution task. It not only improves the utilization efficiency of robot resources, reduces waiting time and resource waste, but also enhances the system's adaptability in a dynamic environment, improves the accuracy and efficiency of distribution services, and improves user experience and the overall efficiency of the platform.

[0062] Figure 3 A schematic diagram of a combined positioning and navigation planning method between buildings provided in this application Figure 2 ,exist Figure 2 Based on the embodiment, the obstacle classification model includes a first sub-model and a second sub-model, such as Figure 3 As shown, the server uses a pre-acquired obstacle classification model to process the image data to determine whether there is an obstacle in front of the target delivery robot, as well as the obstacle type and obstacle posture information when there is an obstacle, specifically including the following steps: S301: The server inputs the image data into the first sub-model, extracts local features, and optimizes the extracted local features to obtain an optimized feature image.

[0063] In this step, the first sub-model is an algorithm model used to process and analyze image data. Its main function is to extract local features from the input image data. The features include the shape, edge, texture and other information of obstacles in the image, so as to help the navigation planning system between buildings identify and classify obstacles, and provide a data basis for the analysis and decision-making of the target delivery robot.

[0064] Specifically, the first sub-model includes a convolution module and a pooling module. The server inputs the image data into the first sub-model, extracts local features, and optimizes the extracted local features to obtain an optimized feature image, which specifically includes: S3011: The server inputs the image data into the convolution module of the first sub-model to extract local features and obtain local features in the image data.

[0065] In this step, the convolution module is a computing structure specially used to process image data. Its main function is to perform local extraction. The convolution operation slides on the image through the filter to gradually extract the features of different regions. By using the convolution module for local feature extraction, the accuracy of image feature extraction is ensured, the accuracy of obstacle detection is improved, and reliable data support is provided for subsequent path planning and decision-making.

[0066] In one specific implementation of this scheme, the convolution module extracts features by sliding multiple filters (also called convolution kernels) over the input image. Each filter is a small matrix that is moved (or convolved) stepwise over the image, performing a dot product operation at each location. Specifically, the filter is element-wise multiplied with a local area of ​​the image, and the results are added to obtain a single value, which constitutes a pixel in the feature map. As the filter slides over the image, that is, this process is repeated at different locations, each filter can recognize a specific feature pattern, such as edges, corners, or textures. By using multiple filters, the convolution layer is able to extract a variety of different features from the input image.

[0067] Specifically, the convolution module contains three layers, two three-dimensional convolution layers and one two-dimensional convolution layer. The convolution kernel sizes of the three-dimensional convolution layer are and, and the number of convolution kernels is 8 and 16 respectively; the convolution kernel size of the two-dimensional convolution layer is and, and the number of convolution kernels is 64. The three-dimensional convolution layer with a convolution kernel size of uses 8 convolution kernels to perform convolution operations in the spatial dimension and depth dimension. Each convolution kernel performs a dot product operation in a local area of ​​the input image data to extract specific feature patterns, such as edges or textures. Then, the three-dimensional convolution layer with a convolution kernel size of uses 16 convolution kernels in the spatial dimension and depth dimension to further process the features output by the previous three-dimensional convolution layer, and no longer expands in the depth dimension. Then the two-dimensional convolution layer with a convolution kernel size of receives the output of the second three-dimensional convolution layer as input, and only uses 64 convolution kernels to perform convolution operations in the spatial dimension, and finally obtains a variety of different local features of obstacles in the input image data, providing strong support for subsequent classification tasks.

[0068] S3012: The server inputs the local features into the pooling module of the first sub-model to perform dimensionality reduction and retain spatial feature processing to obtain an optimized feature image.

[0069] In this step, the main function of the pooling module is to reduce the dimension of local features. By reducing the size of local features, the complexity and amount of calculation of the data are reduced, and the operational efficiency of the model is improved. The pooling module is usually implemented through maximum pooling or average pooling. Maximum pooling selects the maximum value of the pooling window and retains the most significant features; average pooling calculates the average value in the pooling window, effectively reducing the feature size while retaining key spatial information. The use of the pooling module helps to reduce overfitting and improve the generalization ability of the model.

[0070] In a specific implementation of this solution, the pooling module uses the maximum pooling window to obtain the first optimized feature, uses the average pooling window to obtain the second optimized feature, and averages the first optimized feature and the second optimized feature to obtain the optimized feature image. By combining the maximum pooling and average pooling methods, both the significant features are retained and the local features are smoothed, providing a more comprehensive and stable feature representation. The optimized feature image can better support subsequent obstacle detection and path planning, and enhance the adaptability and generalization performance of the model in different environments.

[0071] S302: The server inputs the optimized feature image into the second sub-model, extracts global features, and identifies and classifies obstacles based on the extracted global features to obtain identification and classification results. The identification and classification results include: static obstacles and the position and posture information of static obstacles; or, dynamic obstacles and the position and posture information of dynamic obstacles; or, no obstacles.

[0072] In this step, the second sub-model is used to process the optimized feature image output by the first sub-model. Its main function is to perform global feature extraction and identify and classify obstacles based on these features. Among them, the identification and classification results include: static obstacles, the position and posture information of static obstacles, such as stairs, garages, walls, decorative potted plants, etc., as well as the position, size and other information of the obstacles; or dynamic obstacles, the position and posture information of dynamic obstacles, such as pedestrians, other delivery robots, animals, etc., as well as the position, size and other information of the obstacles; or no obstacles. The second sub-model can effectively identify and classify different types of obstacles by performing global feature extraction on the optimized feature image, and provide their position and posture information, which helps the target delivery robot to navigate safely in complex environments and avoid collisions.

[0073] Specifically, the second sub-model includes an image segmentation module, a linear embedding module, a coding module, and a recognition and classification module. The server inputs the optimized feature image into the second sub-model, extracts global features, and recognizes and classifies obstacles based on the extracted global features to obtain recognition and classification results, which specifically include: S3021: The server inputs the optimized feature image into the image segmentation module of the second sub-model for image segmentation to obtain multiple image blocks.

[0074] In this step, the role of image segmentation is to divide the input optimized feature image into image blocks of fixed size so as to convert it into sequence data for subsequent encoding processing.

[0075] In a specific implementation of this solution, the image segmentation module records the input optimized feature image as, where and are the height and width of the image respectively, and is the number of channels. If the image is a grayscale image, the number of channels is 1; if the image is a color image, the number of channels is 3; if the image is a four-channel format image (Red Green Blue Alpha, RGBA), the number of channels is 4. The Alpha channel is used to represent the transparency or opacity of each pixel in the image. The Alpha value is usually between 0 and 255, where 0 represents complete transparency and 255 represents complete opacity. The image segmentation module divides the entire image into image blocks of size to obtain image blocks, where. It provides a basis for the subsequent conversion of image data into sequence data for encoding processing.

[0076] Specifically, before the server inputs the image data into the first sub-model, the image data is preprocessed, including random cropping, denoising, contrast enhancement, transparency increase, and normalization, which improves the adaptability of the second sub-model in different environments and enhances the accuracy and reliability of obstacle recognition and classification. In a specific implementation of this solution, the image segmentation module records the input optimized feature image as, and divides the entire image into image blocks of size, obtaining 16 image blocks of size.

[0077] S3022: The server inputs the multiple image blocks into the linear embedding module of the second sub-model for position encoding to obtain an embedding vector.

[0078] In this step, the linear embedding module is used to convert multiple image blocks into embedding vectors that the model can handle. First, the linear embedding module flattens each image block, converting it from a two-dimensional form to a one-dimensional vector, and then passes the one-dimensional vector through a linear projection layer so that each flattened image block is mapped to a feature space of fixed dimension. In addition, in order to enable the second sub-model to understand the position of each image block in the original image, the linear embedding module also adds position encoding. By adding position information to the embedding vector of each image block, the model can recognize and maintain the spatial structure of the image. Finally, after linear embedding and position encoding processing, the image block is converted into a series of embedding vectors, which not only contains the visual features of the image block, but also carries position information, ready to be input into the subsequent encoding module for global feature extraction. This process ensures that the model can effectively capture and utilize the global information and spatial relationships of the image, thereby improving the accuracy of obstacle recognition and classification.

[0079] Specifically, the linear embedding module converts the image block into a two-dimensional matrix, converts it into a one-dimensional vector through linear mapping and adds a category vector with the same length as the one-dimensional vector. In a specific implementation of this scheme, the linear embedding module connects the image blocks to obtain a two-dimensional matrix, which is converted into a one-dimensional vector through linear mapping. Therefore, the image of the dimension is converted into a one-dimensional vector. Then, a learnable classification vector is added to represent the global features of the image after encoding. Finally, the position encoding containing spatial information is added as the input of the encoder layer.

[0080] S3023: The server inputs the embedded vector into the encoding module of the second sub-model for feature extraction and representation learning to obtain a high-dimensional feature representation.

[0081] In this step, the encoding module is usually composed of multiple layers of neural networks. Through the encoding module, the input data is deeply analyzed and converted, and the original embedding vector is converted into a high-dimensional feature representation, which not only contains the basic information of the input data, but also captures more complex patterns and relationships, so that the model can understand the deep semantics and structure of the data and enhance the expressiveness of the model. It provides a basis for subsequent classification tasks, so that the model can make predictions and decisions more accurately and smoothly.

[0082] Specifically, in a specific implementation of this solution, a Transformer encoding module with a multi-head self-attention mechanism is used to calculate the correlation between each element in the input sequence and other elements. The attention weight is calculated by linear transformation of the query, key and value matrices, which is used to weighted sum the value matrix of the input sequence to generate the contextual representation of each element. The multi-head self-attention mechanism calculates multiple different attention heads in parallel, each head captures different feature patterns in different subspaces, and finally concatenates and linearly transforms the outputs of these heads to form the final attention output, that is, to obtain a high-dimensional feature representation.

[0083] S3024: The server inputs the high-dimensional feature representation into the recognition and classification module of the second sub-model for recognition and classification processing to obtain a classification result.

[0084] In this step, the recognition and classification module is usually composed of one or more fully connected layers, which uses the previously obtained high-dimensional features to perform specific recognition and classification tasks. Specifically, the probability distribution of each category is calculated, and the output is converted into a probability value used to represent the input data belonging to each category by using an activation function. Since the input is a high-dimensional feature representation extracted by the encoding module, the recognition and classification module can use this rich feature information for more accurate classification, which not only improves the accuracy of classification, but also enhances the robustness of the model in the face of noise and variant data, making it more reliable and adaptable in practical applications.

[0085] Specifically, in a specific implementation of this solution, the recognition and classification module consists of a fully connected layer and an activation function GELU. The recognition and classification module receives the output weights of the multi-head self-attention mechanism and compares them. Finally, like most classification networks, a softmax classifier is used to obtain the final classification result and determine the category and location information of each object.

[0086] The combined positioning navigation planning method between buildings provided in the embodiment of the present application is intended to process image data through an obstacle classification model to determine whether there is an obstacle in front of the target delivery robot and identify the type and posture information of the obstacle. The method achieves this goal through the collaborative work of two sub-models. First, the image data is input into the first sub-model to extract local features in the image, such as shape, edge and texture, and optimize the image to generate an optimized feature image. Then, the optimized feature image is input into the second sub-model, and the obstacles are identified and classified through global feature extraction and analysis, and the results including obstacle type and posture information are output. Through the above method, high-precision identification and classification of obstacles are achieved. Local feature extraction ensures the capture of detailed information, while global feature extraction enhances the model's adaptability to complex environments. Through multi-level feature extraction and optimization processing, the model can maintain efficient obstacle detection and path planning capabilities in different environments, significantly improving the navigation safety and decision accuracy of the delivery robot. Overall, the method shows excellent robustness and generalization performance in complex building environments, providing strong technical support for intelligent navigation systems.

[0087] Figure 4 A schematic diagram of a combined positioning and navigation planning method between buildings provided in this application Figure 3 Based on the above embodiment, the server uses an obstacle avoidance path planning algorithm to plan an obstacle avoidance path according to the obstacle type, the obstacle posture information and the current position information of the target delivery robot, which specifically includes the following steps: S401: The server determines the state space representation and action space representation of the target delivery robot based on the obstacle type and the position information of the obstacle.

[0088] In this step, the state space representation information of the target delivery robot is used to indicate the position, direction, speed, and size of the target delivery robot; the action space representation information is used to indicate the obstacle avoidance measures that the target delivery robot may take when there are obstacles in the front area of ​​the target delivery robot. The measures are used to control the transportation of the target delivery robot to avoid obstacles and successfully reach the target position.

[0089] Regarding the state space representation of the target delivery robot, specifically, the state space representation of the target delivery robot is, where is the position of the target delivery robot, is the speed of the target delivery robot, is the shape, size and position of the obstacles in the front area of ​​the target delivery robot, is the number of obstacles, and represents the relative distances from the obstacles to the target delivery robot in the horizontal and vertical directions, respectively.

[0090] The relative distance between the current position of the target delivery robot and the target position is expressed as; the relative distance between the current position of the target delivery robot and the item position is expressed as; the relative distance between the current position of the target delivery robot and the delivery robot charging station is expressed as. In summary, the state space of the target delivery robot is expressed as.

[0091] Regarding the motion space representation of the target delivery robot, specifically, the motion space representation of the target delivery robot is defined as, where is the motion change of the target delivery robot in the horizontal direction, is the motion change of the target delivery robot in the vertical direction, and is the speed change of the target delivery robot.

[0092] The state and action space representation enhances the target delivery robot's environmental perception, path planning, and dynamic obstacle avoidance capabilities, improving its autonomous navigation and task execution efficiency in complex environments.

[0093] S402: The server uses an obstacle avoidance path planning algorithm based on the state space representation, action space representation and position information of the target delivery robot to obtain an obstacle avoidance path that avoids the obstacles.

[0094] In this step, the state space representation of the target delivery robot determined in S401 provides key parameters such as the current position information, speed, direction and size of the target delivery robot and obstacles, and the action space representation provides a set of actions that the target delivery robot can perform in different situations, including detouring to the left, detouring to the right, stopping and waiting, etc. The obstacle avoidance path planning algorithm calculates a path that can safely avoid obstacles based on the state information of the robot and obstacles, taking into account multiple factors such as the shortest path, minimum energy consumption and minimum time, so as to ensure that the robot can successfully reach the target location in a complex environment.

[0095] Specifically, user A, who works at the front desk on the 7th floor of the office area of ​​Building X, submits an item delivery request through his terminal device on the mobile application of the delivery platform. After receiving the item delivery request, the server determines that the delivery robot C among the multiple delivery robots in Building X is the target delivery robot, and performs path planning based on the location of the target delivery robot C, the item location, and the target location to obtain the first-stage path and the second-stage path. In the second-stage path, the server instructs the delivery robot C to drive along the designated corridor to the No. 3 elevator, take the elevator to the 12th floor, and then drive along the designated direction to Room 1201.

[0096] When the target delivery robot is heading to Elevator No. 3, there is a cleaner in the area ahead who is moving a cart to clean. In order to achieve efficient path planning and obstacle avoidance, the system first obtains the robot's state space representation through sensors, including its current position, speed, direction, and the position information and dynamic characteristics of surrounding obstacles. The obstacle avoidance path planning algorithm uses this information to generate a series of possible actions through a policy network, including acceleration, deceleration, turning or stopping, to deal with possible obstacles ahead. The algorithm evaluates the value of different actions based on the current state space and action space information, calculates the expected return of each action, and selects the optimal obstacle avoidance strategy, such as deceleration or changing the path, to safely avoid the cleaner who is moving the cart to clean. And deliver the items to the target location in the shortest time. It not only improves the robot's autonomous navigation ability, but also significantly improves delivery efficiency and safety.

[0097] After obtaining the obstacle avoidance path, the server uses the obstacle avoidance path to replace the local path of the target delivery robot in the current driving path to the area where the obstacle is located, that is, to obtain the optimized driving path.

[0098] In a specific implementation, the building-to-building navigation planning system also uses visualization tools to monitor and display the driving process of the target delivery robot in real time, so that managers can understand the location, path and status of the delivery robot in real time, thereby improving the transparency and controllability of the entire delivery process. At the same time, real-time monitoring helps to quickly identify and respond to abnormal situations, such as the robot encountering obstacles, deviating from the predetermined path or malfunctioning, etc., to ensure the safety and timeliness of the delivery task. The data records of the visualization tool also provide a basis for subsequent system improvement and optimization.

[0099] The combined positioning navigation planning method between buildings provided in the embodiment of the present application realizes efficient obstacle avoidance path planning in complex environments by utilizing the state space and action space representation of the target delivery robot, combining the position information of obstacles, and adopting an obstacle avoidance path planning algorithm. Through the above method, the autonomous navigation capability and path planning efficiency of the delivery robot between buildings are improved, the safety and timeliness of the delivery task are ensured, and the intelligence level of the overall logistics system is improved.

[0100] The present application provides a building navigation planning system, including: a server, a terminal device and multiple delivery robots.

[0101] like Figure 1 As shown, the terminal device is used to obtain the information of the item to be delivered, the location of the item, and the target location, and generate an item delivery request based on the item information, the location of the item, and the target location, and send the item delivery request to the server.

[0102] Each delivery robot is equipped with a laser radar and a camera. The laser radar is used to collect three-dimensional point cloud images in front of it, and the camera is used to collect two-dimensional images in front of it.

[0103] The server is used to execute the method provided by the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be described in detail here.

[0104] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A combined positioning and navigation planning method between buildings, characterized in that: A server applied to a navigation planning system between buildings, wherein the navigation planning system between buildings also includes a terminal device and a plurality of delivery robots, and the method includes: The server receives an item delivery request sent by the terminal device, wherein the item delivery request is generated based on information of the item to be delivered, the location of the item, and the target location; After receiving the item delivery request, the server determines a target delivery robot from a plurality of delivery robots based on the item location; The server performs path planning based on the location of the target delivery robot, the location of the item, and the target location to obtain a driving path, wherein the driving path includes a first-stage path for loading items and a second-stage path for delivering items; The server sends the driving path to the target delivery robot; The server receives image data uploaded in real time by the target delivery robot during the process of loading and delivering items according to the driving path; The server optimizes the driving path based on the image data, and sends the optimized driving path to the target delivery robot; The server optimizes the driving path based on the image data and sends the optimized driving path to the target delivery robot, including: The server processes the image data using a pre-acquired obstacle classification model to determine whether there is an obstacle in front of the target delivery robot, and the type of obstacle and the position information of the obstacle when there is an obstacle; If it is determined that there is an obstacle in front of the target delivery robot, the server uses an obstacle avoidance path planning algorithm to plan an obstacle avoidance path to avoid the obstacle according to the type of obstacle, the position information of the obstacle and the current position information of the target delivery robot; The server optimizes the driving path of the target delivery robot according to the obstacle avoidance path to obtain an optimized driving path; The obstacle classification model includes a first sub-model and a second sub-model; the server uses the pre-acquired obstacle classification model to process the image data to determine whether there is an obstacle in front of the target delivery robot, and the obstacle type and obstacle posture information when there is an obstacle, including: The server inputs the image data into the first sub-model, extracts local features, and optimizes the extracted local features to obtain an optimized feature image; The server inputs the optimized feature image into the second sub-model, performs global feature extraction, and identifies and classifies obstacles based on the extracted global features to obtain identification and classification results; The identification and classification results include: A static obstacle, and position information of the static obstacle; or A dynamic obstacle, and position information of the dynamic obstacle; or No obstacles.

2. The method according to claim 1, characterized in that The server performs path planning based on the location of the target delivery robot, the location of the item, and the target location to obtain a driving path, including: The server performs path planning for the target delivery robot to pick up goods based on the architectural design of the current building, the location of the target delivery robot and the location of the item to obtain the first stage path; The server performs path planning for the delivery process of the target delivery robot based on the architectural design drawing of the current building, the location of the item and the target location to obtain the second stage path; The architectural design drawing includes the location and shape of each building in the building, the entrance and exit of each building, and the internal layout.

3. The method according to claim 1 or 2, characterized in that: The first sub-model includes a convolution module and a pooling module. The server inputs the image data into the first sub-model, extracts local features, and optimizes the extracted local features to obtain an optimized feature image, including: The server inputs the image data into the convolution module of the first sub-model to extract local features, thereby obtaining local features in the image data; The server inputs the local features into the pooling module of the first sub-model to perform dimension reduction and retain spatial feature processing to obtain the optimized feature image.

4. The method according to claim 1 or 2, characterized in that: The second sub-model includes an image segmentation module, a linear embedding module, a coding module and a recognition and classification module. The server inputs the optimized feature image into the second sub-model, extracts global features, and recognizes and classifies obstacles based on the extracted global features to obtain recognition and classification results, including: The server inputs the optimized feature image into the image segmentation module of the second sub-model to perform image segmentation to obtain a plurality of image blocks; The server inputs the multiple image blocks into the linear embedding module of the second sub-model for position encoding to obtain an embedding vector; The server inputs the embedding vector into the encoding module of the second sub-model for feature extraction and representation learning to obtain a high-dimensional feature representation; The server inputs the high-dimensional feature representation into the recognition and classification module of the second sub-model for recognition and classification processing to obtain the classification result.

5. The method according to claim 1 or 2, characterized in that: The server plans an obstacle avoidance path to avoid the obstacle using an obstacle avoidance path planning algorithm according to the obstacle type, the position information of the obstacle and the current position information of the target delivery robot, including: The server determines the state space representation and the action space representation of the target delivery robot based on the obstacle type and the position information of the obstacle; The server adopts an obstacle avoidance path planning algorithm based on the state space representation, action space representation and position information of the target delivery robot to obtain the obstacle avoidance path that avoids the obstacle.

6. The method according to claim 1 or 2, characterized in that: The server optimizes the driving path of the target delivery robot according to the obstacle avoidance path to obtain an optimized driving path, including: Based on the current position of the target delivery robot and the posture information of the obstacle, the server uses the obstacle avoidance path to replace the local path from the target delivery robot to the area where the obstacle is located in the current driving path, so as to obtain the optimized driving path.

7. The method according to claim 1 or 2, characterized in that: The image data includes: the target delivery robot collects images of the front area through a laser radar to obtain a three-dimensional point cloud image during the driving process, and obtains a two-dimensional image by shooting the front area through a camera.

8. A building navigation planning system, characterized in that: include: Servers, terminal devices, and multiple delivery robots; The terminal device is used to obtain information about items to be delivered, item locations, and target locations, and to generate an item delivery request based on the item information, item locations, and target locations, and to send the item delivery request to the server; The server is used to execute the combined positioning and navigation planning method between buildings according to any one of claims 1 to 7; Each delivery robot is equipped with a laser radar and a camera. The laser radar is used to collect three-dimensional point cloud images in front of the vehicle, and the camera is used to collect two-dimensional images in front of the vehicle.

Images