Man-machine co-driving automatic obstacle avoidance method for campus unmanned logistics vehicle

Abstract

The invention relates to a man-machine co-driving automatic obstacle avoidance method for a campus unmanned logistics vehicle. The method comprises the steps that real-time environment data of the surrounding environment of the unmanned logistics vehicle is acquired, and a grid map with the unmanned logistics vehicle as the center is constructed; sending the image data to the background manual operation platform, and then receiving a target point; on the basis of the target point, path point sampling is carried out according to the multiple preset radius level parameters, and a candidate path set is generated through polynomial curve fitting; and calculating the cost of each candidate path in the candidate path set according to the grid map, selecting the path with the minimum cost as the optimal path, and generating a control instruction. According to the method, an artificial target point and a dynamic path planning technology are introduced, so that not only is the dependence on a static map eliminated, but also the dynamic environment response capability is improved through multilayer sampling and multi-target evaluation, a safe path can be quickly generated in a complex scene, and the obstacle avoidance capability and the driving efficiency of the unmanned logistics vehicle are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned vehicle control, and in particular relates to an automatic obstacle avoidance method for a human-machine co-driving unmanned logistics vehicle on campus. Background Art

[0002] Unstructured campus roads present a complex dynamic environment, with frequent presence of dynamic elements such as pedestrians and temporary obstacles. Traditional campus-based unmanned logistics vehicles typically rely on pre-collected static maps to calculate point-to-point driving routes through planning algorithms such as sampling, search, or optimization. However, due to the uncertainty of the dynamic environment, the paths calculated by these algorithms are difficult to guarantee safe driving. Furthermore, static maps require real-time updates as the campus layout adjusts and obstacles change, resulting in high maintenance costs. For example, when pedestrians cross the road or goods are temporarily piled up within the campus, static maps cannot reflect these changes in real time, resulting in the risk of collision when the unmanned logistics vehicle follows the pre-planned route. Furthermore, the cost of re-collecting and updating maps is high. Furthermore, while purely autonomous dynamic obstacle avoidance solutions attempt to achieve obstacle avoidance using real-time sensor data, they face challenges such as high computational complexity, increased hardware costs, and insufficient ability to escape in complex scenarios, making it difficult to achieve a balance between safety and operational efficiency in unstructured campuses. Summary of the Invention

[0003] Based on this, it is necessary to provide an automatic obstacle avoidance method for human-machine co-driving of campus unmanned logistics vehicles to address the above technical problems, aiming to get rid of the dependence on static maps, improve the vehicle's responsiveness and escape ability to dynamic environments, reduce map maintenance costs and improve the efficiency of campus logistics operations.

[0004] In a first aspect, the present application provides a method for automatic obstacle avoidance for a human-machine co-driving unmanned logistics vehicle on campus, comprising:

[0005] Acquire real-time environmental data of the environment surrounding the unmanned logistics vehicle and build a raster map centered on the unmanned logistics vehicle based on the real-time environmental data. The real-time environmental data includes image data and point cloud data.

[0006] After sending the image data to the backend manual operation platform, the target point is received, which is the next destination of the unmanned logistics vehicle;

[0007] Based on the target point, path points are sampled according to multiple preset radius level parameters, and a candidate path set is generated through polynomial curve fitting;

[0008] According to the grid map, the cost of each candidate path in the candidate path set is calculated, the path with the minimum cost is selected as the optimal path, and control instructions are generated. The control instructions are used to control the unmanned vehicle to drive along the optimal path. The cost includes navigation cost, collision cost and consistency cost.

[0009] In one embodiment, after sending the image data to the backend manual operation platform, receiving the target point includes:

[0010] After sending the image data to the backend manual operation platform, the manually specified driving direction angle is obtained;

[0011] According to the manually specified driving direction angle and the current position of the unmanned logistics vehicle, combined with the preset driving radius, the spatial point coordinates are calculated through geometric mapping to obtain the target point.

[0012] In one embodiment, based on the target point, path point sampling is performed according to multiple preset radius level parameters, and a candidate path set is generated by polynomial curve fitting, including:

[0013] Determine each preset radius level parameter according to the target point, and perform radius level division processing according to each preset radius level parameter to obtain a first radius level, a second radius level, and a third radius level;

[0014] Based on the current position of the unmanned logistics vehicle and the preset angle range, for the first radius level, the second radius level, and the third radius level, a first sampling point set, a second sampling point set, and a third sampling point set are generated respectively through uniform angle sampling operations;

[0015] Performing a Cartesian product combination operation on the first sampling point set, the second sampling point set, and the third sampling point set to obtain multiple path endpoint combinations;

[0016] For each path endpoint combination, the current position of the unmanned logistics vehicle is used as the starting point, and a single path is generated through cubic polynomial curve fitting. The single paths are combined to obtain a set of candidate paths.

[0017] In one embodiment, calculating the cost of each candidate path in the candidate path set according to the grid map includes:

[0018] Based on the coordinates of the end point and the target point of each candidate path, the Euclidean distance is calculated to obtain the original navigation distance value, and the original navigation distance value is normalized to obtain the navigation cost of each candidate path;

[0019] For each candidate path, a uniform sampling point generation process is performed according to the path curve equation to obtain multiple path position points;

[0020] Based on each path location point, the grid value verification process is performed through the grid map to obtain the number of collision points, and the ratio of the number of collision points to the path location points is calculated to obtain the collision cost;

[0021] Get the endpoint coordinates of the previous frame's optimal path. Calculate the original offset distance using Euclidean distance based on the endpoint coordinates of each candidate path and the endpoint coordinates of the previous frame's optimal path. Normalize the original offset distance to obtain the consistency cost of each candidate path.

[0022] According to the preset weight coefficient, the navigation cost, collision cost and consistency cost of each candidate path are weighted and fused to obtain the corresponding cost.

[0023] In one embodiment, a grid map centered on an unmanned logistics vehicle is constructed based on real-time data, including:

[0024] Perform grid map initialization processing according to the preset grid resolution and map radius parameters to obtain an initialized grid map centered on the unmanned logistics vehicle, where the initial value of each grid in the initialized grid map is 0;

[0025] For each point cloud coordinate in the point cloud data, relative coordinate conversion is performed based on the point cloud coordinates and the coordinates of the unmanned logistics vehicle to obtain the Cartesian coordinates relative to the unmanned logistics vehicle;

[0026] Based on the Cartesian coordinates and the preset grid resolution, the grid index calculation is performed to obtain the row and column indexes of the corresponding grid. Based on the initialized grid map, the grid value is updated according to the row and column indexes, and the corresponding grid value is set to 1 to obtain the grid map.

[0027] In one embodiment, based on each path location point, a grid value verification process is performed through a grid map to obtain the number of collision points, including:

[0028] If the grid value of each path position point in the grid map is 1, the corresponding path position point is marked as a collision point;

[0029] If the grid value of each path location point in the grid map is 0, the corresponding path location point is marked as a safe point;

[0030] Combine the collision points to get the number of collision points.

[0031] In one embodiment, the target point is obtained by one of the following methods:

[0032] Receive the absolute coordinate instruction sent by the background manual operation platform, generate the coordinates of the target point according to the spatial position in the absolute coordinate instruction, and obtain the target point;

[0033] Alternatively, the target point can be obtained by manually selecting the electronic map of the park displayed on the back-end manual operation platform.

[0034] Secondly, this application also provides a campus unmanned logistics vehicle human-machine co-driving automatic obstacle avoidance system, including:

[0035] The environmental perception and map construction module is used to obtain real-time environmental data of the environment around the unmanned logistics vehicle and build a raster map centered on the unmanned logistics vehicle based on the real-time environmental data. The real-time environmental data includes image data and point cloud data;

[0036] The human-computer interaction target point generation module is used to send image data to the background manual operation platform and then receive the target point, which is the next destination of the unmanned logistics vehicle;

[0037] A multi-layer path sampling and fitting module is used to sample path points based on the target point according to multiple preset radius level parameters, and generate a set of candidate paths through polynomial curve fitting;

[0038] The path evaluation and decision control module is used to calculate the cost of each candidate path in the candidate path set based on the grid map, select the path with the minimum cost as the optimal path, and generate control instructions. The control instructions are used to control the unmanned vehicle to drive along the optimal path. The cost includes navigation cost, collision cost and consistency cost.

[0039] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the first aspect when executing the computer program.

[0040] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps in the first aspect when the computer program is processed.

[0041] The aforementioned automatic obstacle avoidance method for a campus unmanned logistics vehicle, operated by a human and machine, acquires real-time environmental data of the vehicle's surroundings and constructs a grid map. This method accurately reflects the distribution of static and dynamic obstacles around the vehicle in real time, providing precise environmental data for subsequent path planning and obstacle avoidance decisions. This allows the vehicle to plan its actions based on the real environment. Secondly, image data is sent to a backend manual operation platform and a target point is received. The target point is determined using human observation of the real-time imagery. This effectively eliminates reliance on pre-collected and maintained static maps, reducing maintenance costs. Furthermore, the vehicle can adjust its target direction based on human judgment in a dynamically changing environment, improving its responsiveness and ability to escape obstacles. Furthermore, based on the target point and preset radius level parameters, a polynomial sampling method is used to perform multi-level sampling within different radius and angle ranges to generate a set of candidate paths. This method covers a variety of possible driving paths around the vehicle, increasing the diversity of path selection.

[0042] Finally, the cost of the candidate paths is calculated based on the grid map and the path with the lowest cost is selected. The navigation cost, collision cost and consistency cost are comprehensively considered, and the final cost of each path is obtained through weighted calculation. The optimal path that is safe, meets the navigation goals and has smooth driving can be screened out from the candidate paths, thereby generating control instructions to control the unmanned vehicle to drive along the optimal path, further improving the driving safety and operational efficiency of the unmanned logistics vehicle.

[0043] Compared with traditional unmanned logistics vehicle obstacle avoidance methods, this method not only gets rid of the dependence on static maps and saves map maintenance costs by introducing artificial target points and dynamic path planning technology, but also improves the dynamic environment response capability through multi-layer sampling and multi-target evaluation. It can quickly generate safe paths in complex scenarios such as pedestrians crossing and temporary obstacles, improves the obstacle avoidance capability and driving efficiency of unmanned logistics vehicles, and provides a reliable solution for the safe and efficient operation of unmanned logistics vehicles in the park. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 A flow chart of an automatic obstacle avoidance method for a campus unmanned logistics vehicle driven by a human and a machine, provided by an exemplary embodiment of the present invention;

[0046] Figure 2 A schematic flow chart of a method for generating a candidate path set provided by an exemplary embodiment of the present invention;

[0047] Figure 3 A schematic structural diagram of an automatic obstacle avoidance system for human-machine co-driving of a campus unmanned logistics vehicle is provided as an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0049] In one embodiment, Figure 1As shown, a method for automatic obstacle avoidance for a campus unmanned logistics vehicle with human-machine co-driving is provided. This embodiment uses the method applied to a terminal as an example. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0050] S101: Acquire real-time environmental data of the environment surrounding the unmanned logistics vehicle, and construct a raster map centered on the unmanned logistics vehicle based on the real-time environmental data, where the real-time environmental data includes image data and point cloud data.

[0051] Specifically, the unmanned logistics vehicle can obtain real-time data on the surrounding environment through the sensor module it carries. Schematically, the sensor module includes a camera and a lidar, wherein the camera is used to collect image data and the lidar is used to obtain point cloud data. The image data can provide visual information of the environment, while the point cloud data can accurately perceive the distance and shape of surrounding objects. Moreover, based on the real-time environmental data, a grid map can be constructed. The grid map is a three-dimensional space centered on the unmanned logistics vehicle divided into multiple grid units, and the state of each grid unit can be determined by analyzing the point cloud data. Schematically, the grid unit corresponding to the obstacle position detected by the lidar can be marked as occupied, while the grid unit where no obstacle is detected can be marked as idle. By constructing a grid map, the dynamic changes of the environment around the unmanned logistics vehicle can be reflected in real time, providing basic data support for subsequent path planning.

[0052] S102: After sending the image data to the backend manual operation platform, the target point is received. The target point is the next target destination of the unmanned logistics vehicle.

[0053] Specifically, after acquiring image data, the unmanned logistics vehicle can transmit this image data in real time to a backend manual operation platform via network protocols such as MQTT. Backend operators can use the image data to visually observe the environmental conditions surrounding the unmanned logistics vehicle and manually specify a target point based on actual needs and environmental conditions. This target point is the location the unmanned logistics vehicle needs to reach next, providing clear directional guidance for the unmanned logistics vehicle's travel. Compared to traditional path planning methods based on static maps, manually specified target points can adapt to environmental changes in real time, avoiding path failures caused by untimely map updates. Furthermore, manual intervention can better handle complex scenarios, such as pedestrians crossing the road or temporary obstacles, thereby improving the driving safety and flexibility of unmanned logistics vehicles.

[0054] S103: Based on the target point, path point sampling is performed according to a plurality of preset radius level parameters, and a candidate path set is generated by polynomial curve fitting.

[0055] Specifically, after receiving the target point, a multi-level radius sampling strategy can be used to perform path planning based on the target point. The sampling angle range of each level can be set, and the number of sampling points for each level can be set according to actual needs, for example, 5 points are sampled for each level. Through this multi-level sampling method, a rich set of path points can be generated, providing diverse options for subsequent path generation. After sampling is completed, the path points can be connected through a polynomial curve fitting method to generate a set of candidate paths. This polynomial curve fitting can ensure the smoothness of the path and avoid the path being too stiff, which may cause unstable vehicle driving.

[0056] S104: Calculate the cost of each candidate path in the candidate path set based on the grid map, select the path with the minimum cost as the optimal path, and generate control instructions. The control instructions are used to control the unmanned vehicle to drive along the optimal path. The cost includes navigation cost, collision cost, and consistency cost.

[0057] Specifically, the navigation cost is used to measure the distance between the endpoint of a candidate path and the target point. The smaller the distance, the lower the navigation cost. The collision cost can be used to assess whether a candidate path has a collision risk. By introducing the collision cost, the unmanned logistics vehicle can effectively avoid collisions with obstacles during driving. The consistency cost is used to measure the consistency between the candidate path and the optimal path selected in the previous frame. The introduction of the consistency cost can ensure the smoothness and consistency of the unmanned logistics vehicle's driving path, avoiding unstable vehicle driving caused by frequent path changes. The total cost is calculated by calculating the weighted sum of the navigation cost, collision cost, and consistency cost for each candidate path. By comparing the cost costs of each candidate path, the path with the lowest cost can be selected as the optimal path. Finally, the unmanned logistics vehicle can generate control instructions based on the optimal path to control the unmanned logistics vehicle to drive along the optimal path, thereby achieving safe and efficient obstacle avoidance driving.

[0058] In one embodiment, after sending the image data to the backend manual operation platform, receiving the target point includes:

[0059] After sending the image data to the backend manual operation platform, the manually specified driving direction angle is obtained;

[0060] According to the manually specified driving direction angle and the current position of the unmanned logistics vehicle, combined with the preset driving radius, the spatial point coordinates are calculated through geometric mapping to obtain the target point.

[0061] Specifically, the backstage operator can understand the real-time environmental conditions around the unmanned logistics vehicle by observing the image data, including the location and distribution of dynamic elements such as pedestrians and temporary obstacles. Based on this information, the operator can input a text description through the backstage interface. Through the text conversion module, semantic mapping can be performed by extracting keywords such as 40° to the northeast to obtain the manually specified driving direction angle. The driving direction angle refers to the angle between the current driving direction of the unmanned logistics vehicle and the target direction. It is usually based on the front of the unmanned logistics vehicle, and the clockwise direction is positive. In addition, the operator can also obtain the manually specified driving direction angle through voice input, by converting the voice into text, and then through the text conversion module. Schematically, the operator can also directly input the manually specified driving direction angle through the preset window of the backstage manual operation platform, which is not limited here.

[0062] After obtaining the manually specified driving direction angle, the unmanned logistics vehicle can calculate the spatial coordinates of the target point based on this direction angle and its current position, combined with a preset driving radius. The preset driving radius is pre-set based on the unmanned logistics vehicle's driving speed, braking distance, and the safety requirements of the park environment. For example, it can be set to 10 meters. First, the unmanned logistics vehicle's own positioning system can be used to obtain its real-time position coordinates within the park. Then, based on the manually specified driving direction angle and the unmanned logistics vehicle's current position, combined with the preset driving radius, the spatial coordinates of the target point can be calculated through geometric mapping to obtain the target point.

[0063] In one embodiment, the target point can also be obtained by one of the following methods:

[0064] Receive the absolute coordinate instruction sent by the background manual operation platform, generate the coordinates of the target point according to the spatial position in the absolute coordinate instruction, and obtain the target point;

[0065] Alternatively, the target point can be obtained by manually selecting the electronic map of the park displayed on the back-end manual operation platform.

[0066] Specifically, after observing the image data, the backend operator can directly specify the absolute coordinates of the unmanned logistics vehicle's next target destination by entering the absolute coordinate values ​​through the interactive interface of the backend operation platform. The backend manual operation platform can then send a command containing the absolute coordinates of the target point to the unmanned logistics vehicle. After receiving the command, the unmanned logistics vehicle can directly use the coordinates in the command as the coordinates of the target point without performing additional calculations. This is suitable for fixed loading and unloading points or specific task areas within the park. Alternatively, the operator can also use the mouse or other input device to select the unmanned logistics vehicle's next target destination on the electronic map of the park displayed by the backend manual operation platform. The operation platform can then record the location clicked by the backend operator, convert it into the coordinates of the target point, and send it to the unmanned logistics vehicle. In this way, the backend operator can flexibly select the target location of the unmanned logistics vehicle through an intuitive graphical interface based on the actual layout of the park and real-time environmental conditions. This is suitable for scenarios where the dynamic environment changes significantly or the driving route needs to be frequently adjusted.

[0067] In one embodiment, Figure 2 As shown, based on the target point, path point sampling is performed according to multiple preset radius level parameters, and a candidate path set is generated by polynomial curve fitting, including the following steps:

[0068] S201: Determine each preset radius level parameter according to the target point, and perform radius level division processing based on each preset radius level parameter to obtain a first radius level, a second radius level, and a third radius level;

[0069] S202: Based on the current position of the unmanned logistics vehicle and the preset angle range, for the first radius level, the second radius level, and the third radius level, respectively, a first sampling point set, a second sampling point set, and a third sampling point set are generated by uniform angle sampling operations;

[0070] S203: performing a Cartesian product combination operation on the first sampling point set, the second sampling point set, and the third sampling point set to obtain a plurality of path endpoint combinations;

[0071] S204: For each combination of path endpoints, the current position of the unmanned logistics vehicle is used as the starting point, and a single path is generated by cubic polynomial curve fitting processing. The single paths are combined to obtain a set of candidate paths.

[0072] Specifically, according to the driving speed, braking distance and safety requirements of the park environment of the unmanned logistics vehicle, multiple radius level parameters are preset. For example, three radius levels can be set, namely 2 meters, 5 meters and 10 meters. According to the above-mentioned preset radius level parameters, the path point sampling range can be divided into three levels: the first radius level, the second radius level and the third radius level, that is, each level corresponds to an area with a radius of 2 meters, 5 meters and 10 meters respectively, with the current position of the unmanned logistics vehicle as the center. Through this layered method, the different driving ranges of the unmanned logistics vehicle from near to far can be covered, providing a more comprehensive set of sampling points for path planning. Subsequently, within each radius level, a path point set can be generated based on the current position of the unmanned logistics vehicle and the preset angle range. Schematically, the preset angle range can be set to plus or minus 60 degrees, indicating the path point sampling area of ​​the unmanned logistics vehicle within 60 degrees on both sides of the target direction. Based on the preset angle range, multiple angle points can be uniformly sampled within the range of plus or minus 60 degrees for each radius level. For example, 5 points can be set to be uniformly sampled, and the first sampling point set, the second sampling point set, and the third sampling point set can be obtained respectively.

[0073] Furthermore, in order to generate diverse path choices, the sampling point sets of different radius levels can be combined by Cartesian product. Cartesian product combination refers to selecting a sampling point from each set to form a path endpoint combination. For example, if each sampling point set contains 5 sampling points, the Cartesian product combination will generate 125 path endpoint combinations. Each path endpoint combination contains three sampling points from the first, second and third radius levels, respectively, providing a rich selection of path endpoints for path planning. Then, for each path endpoint combination, the current position of the unmanned logistics vehicle can be used as the starting point, and a smooth path from the current position of the unmanned logistics vehicle to the path endpoint can be generated by the cubic polynomial curve fitting method. By combining all the generated single paths, a candidate path set can be formed. The candidate path set contains multiple possible paths from the current position of the unmanned logistics vehicle to the target point, providing a rich selection for subsequent path evaluation.

[0074] In one embodiment, a grid map centered on an unmanned logistics vehicle is constructed based on real-time data, including:

[0075] Perform grid map initialization processing according to the preset grid resolution and map radius parameters to obtain an initialized grid map centered on the unmanned logistics vehicle, where the initial value of each grid in the initialized grid map is 0;

[0076] For each point cloud coordinate in the point cloud data, relative coordinate conversion is performed based on the point cloud coordinates and the coordinates of the unmanned logistics vehicle to obtain the Cartesian coordinates relative to the unmanned logistics vehicle;

[0077] Based on the Cartesian coordinates and the preset grid resolution, the grid index calculation is performed to obtain the row and column indexes of the corresponding grid. Based on the initialized grid map, the grid value is updated according to the row and column indexes, and the corresponding grid value is set to 1 to obtain the grid map.

[0078] Specifically, the grid resolution determines the accuracy of the grid map. The grid resolution can be set to 0.1 meters / grid, which means that the side length of each grid cell is 0.1 meters. The map radius defines the coverage of the grid map. You can set a fixed radius with the unmanned logistics vehicle as the center. For example, setting the map radius to 15 meters means that the grid map covers the area within 15 meters around the unmanned logistics vehicle. According to the preset grid resolution and map radius parameters, the width and height of the grid map can be calculated to construct an initialized grid map. During initialization, the values ​​of all grid cells can be set to 0 to indicate that they are all idle.

[0079] After acquiring the point cloud data, since this data is based on the unmanned logistics vehicle's lidar-based point cloud data of its surroundings, it is typically based on the lidar's own coordinate system. Therefore, the coordinates of each point in the point cloud data can be converted to Cartesian coordinates relative to the unmanned logistics vehicle. This involves first obtaining the position coordinates of the unmanned logistics vehicle in the global coordinate system. Then, for each point in the point cloud data, its coordinates are converted to Cartesian coordinates relative to the unmanned logistics vehicle, resulting in the point cloud coordinates relative to the unmanned logistics vehicle. Finally, for each point cloud coordinate, its row and column index in the grid map is calculated using a preset grid resolution. Based on this row and column index, the corresponding grid cell in the initialized grid map is found and its value is updated to 1, indicating that the grid cell is marked as occupied by the point cloud data, indicating the presence of an obstacle. Through this grid index calculation and grid value update process, a complete grid map is ultimately obtained. This grid map, centered on the unmanned logistics vehicle, reflects the dynamic changes in the surrounding environment in real time, providing basic data support for path planning and obstacle avoidance.

[0080] In one embodiment, calculating the cost of each candidate path in the candidate path set according to the grid map includes:

[0081] Based on the coordinates of the end point and the target point of each candidate path, the Euclidean distance is calculated to obtain the original navigation distance value, and the original navigation distance value is normalized to obtain the navigation cost of each candidate path;

[0082] For each candidate path, a uniform sampling point generation process is performed according to the path curve equation to obtain multiple path position points;

[0083] Based on each path location point, the grid value verification process is performed through the grid map to obtain the number of collision points, and the ratio of the number of collision points to the path location points is calculated to obtain the collision cost;

[0084] Get the endpoint coordinates of the previous frame's optimal path. Calculate the original offset distance using Euclidean distance based on the endpoint coordinates of each candidate path and the endpoint coordinates of the previous frame's optimal path. Normalize the original offset distance to obtain the consistency cost of each candidate path.

[0085] According to the preset weight coefficient, the navigation cost, collision cost and consistency cost of each candidate path are weighted and fused to obtain the corresponding cost.

[0086] Specifically, for each candidate path, the distance between its endpoint coordinates and the target coordinates is calculated using the Euclidean distance formula to obtain a raw navigation distance value. This distance represents the spatial deviation between the path endpoint and the target point; the smaller the distance, the more consistent the path is with the navigation goal. The raw navigation distance values ​​for all candidate paths are then normalized using a softmax function and mapped to relative probability values ​​to obtain the navigation cost. Furthermore, for each candidate path, M location points are uniformly generated within the path length using a cubic polynomial curve equation. The sampling interval is dynamically adjusted based on the path curvature; greater curvature results in denser sampling points, ensuring obstacle detection accuracy. By mapping each location point coordinate to a grid map centered on the vehicle and performing grid value verification, the number of collision points is determined. The collision cost is calculated by calculating the ratio of the number of collision points to the total number of sampling points, M. The endpoint coordinates of the optimal path selected in the previous frame of path planning are also obtained. For each candidate path, the Euclidean distance is calculated based on its endpoint coordinates and the endpoint coordinates of the optimal path in the previous frame, and then normalized to obtain the consistency cost. This cost can ensure that the autonomous vehicle's driving path avoids drastic changes, improving control stability. In addition, according to the needs and priorities of path planning, the weight coefficients of navigation cost, collision cost, and consistency cost can be set. By weighted fusion, the cost of each candidate path can be obtained.

[0087] In one embodiment, based on each path location point, a grid value verification process is performed through a grid map to obtain the number of collision points, including:

[0088] If the grid value of each path position point in the grid map is 1, the corresponding path position point is marked as a collision point;

[0089] If the grid value of each path location point in the grid map is 0, the corresponding path location point is marked as a safe point;

[0090] Combine the collision points to get the number of collision points.

[0091] Specifically, for each path location point, its row and column index in the grid map can be calculated based on the grid map resolution and the current position of the unmanned logistics vehicle. This row and column index can then be used to find the corresponding grid value in the grid map. If the grid value is 1, it indicates that the path location point has collided with an obstacle and the path location point is marked as a collision point. If the grid value is 0, it indicates that the path location point is safe and the path location point is marked as a safe point. By performing grid value verification on each path location point, it is possible to quickly determine whether each path location point has collided with an obstacle. Furthermore, for each candidate path, a collision point counter can be initialized to record the number of collision points on that path. When all path locations are traversed, the collision point counter is incremented by 1 for each path location point marked as a collision point. Ultimately, the value of the collision point counter represents the number of collision points on that candidate path. A greater number of collision points indicates a greater likelihood of collision with an obstacle and a lower path safety.

[0092] like Figure 3 As shown, based on the same inventive concept, the embodiment of the present application also provides a campus unmanned logistics vehicle human-machine co-driving automatic obstacle avoidance system 300 for realizing the above-mentioned campus unmanned logistics vehicle human-machine co-driving automatic obstacle avoidance method. The implementation scheme for solving the problem provided by this system is similar to the implementation scheme recorded in the above-mentioned method. Therefore, the specific limitations in one or more embodiments of the campus unmanned logistics vehicle human-machine co-driving automatic obstacle avoidance system provided below can be found in the above-mentioned limitations on the campus unmanned logistics vehicle human-machine co-driving automatic obstacle avoidance method, which will not be repeated here. The system includes:

[0093] The environmental perception and map construction module 301 is used to obtain real-time environmental data of the environment around the unmanned logistics vehicle and construct a raster map centered on the unmanned logistics vehicle based on the real-time environmental data. The real-time environmental data includes image data and point cloud data.

[0094] The human-computer interaction target point generation module 302 is used to receive the target point after sending the image data to the background manual operation platform. The target point is the next destination of the unmanned logistics vehicle.

[0095] A multi-layer path sampling and fitting module 303 is used to sample path points based on the target point according to multiple preset radius level parameters, and generate a candidate path set through polynomial curve fitting;

[0096] The path evaluation and decision control module 304 is used to calculate the cost of each candidate path in the candidate path set based on the grid map, select the path with the minimum cost as the optimal path, and generate control instructions. The control instructions are used to control the unmanned vehicle to drive along the optimal path. The cost includes navigation cost, collision cost and consistency cost.

[0097] In the above system, the environmental perception and map construction module 301 acquires real-time image data and point cloud data of the unmanned logistics vehicle's surroundings and constructs a vehicle-centered grid map based on this data. This module accurately displays the distribution of static and dynamic obstacles around the unmanned vehicle in real time, providing environmental foundational data for subsequent path planning and enabling the unmanned vehicle to dynamically adjust its driving strategy based on the real environment. The human-computer interaction target point generation module 302 transmits real-time images to the backend and receives manually set target points, transforming manual judgment of the dynamic environment into driving direction guidance for the unmanned vehicle. This effectively eliminates the reliance on static maps, reduces map maintenance costs, and allows for rapid adjustment of driving targets in complex environments through manual intervention, improving the unmanned vehicle's responsiveness and ability to escape obstacles in dynamic scenarios. The multi-layer path sampling and fitting module 303 uses a cubic polynomial sampling method based on the target points, sampling multi-level path points at different radius levels. Through polynomial curve fitting, a set of candidate paths is generated, covering a variety of feasible driving trajectories around the unmanned vehicle and increasing the diversity of path selection. The path evaluation and decision control module 304 comprehensively calculates the navigation cost, collision cost and consistency cost, and selects the path with the least cost through weighted fusion. It can screen out the optimal solution from the candidate paths that meets the navigation goal, ensures driving safety and has a smooth path, and then generates control instructions to control the unmanned vehicle to drive along the optimal path, effectively balancing obstacle avoidance safety, navigation accuracy and driving smoothness.

[0098] In an exemplary embodiment, the present invention further provides a computer device comprising a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the present application's method for automatic obstacle avoidance for human-machine co-driving of an unmanned logistics vehicle on campus. A multi-core processor is preferred to improve the system's parallel processing capabilities. Memory: Provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate large amounts of supply information and computing tasks.

[0099] In an exemplary embodiment, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for automatic obstacle avoidance of a human-machine co-driving unmanned logistics vehicle on campus according to the present application.

[0100] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A method for automatic obstacle avoidance for a human-machine co-driving unmanned logistics vehicle on campus, characterized in that: The method comprises: Acquire real-time environmental data of the environment surrounding the unmanned logistics vehicle, and construct a grid map centered on the unmanned logistics vehicle based on the real-time environmental data, wherein the real-time environmental data includes image data and point cloud data; After sending the image data to the backend manual operation platform, a target point is received, which is the next target arrival location of the unmanned logistics vehicle; Based on the target point, path point sampling is performed according to multiple preset radius level parameters, and a candidate path set is generated by polynomial curve fitting; According to the grid map, the cost of each candidate path in the candidate path set is calculated, the path with the minimum cost is selected as the optimal path, and a control instruction is generated. The control instruction is used to control the unmanned vehicle to travel along the optimal path. The cost includes navigation cost, collision cost and consistency cost.

2. The method according to claim 1, characterized in that After sending the image data to the background manual operation platform, receiving the target point includes: After sending the image data to the background manual operation platform, obtaining the manually specified driving direction angle; According to the manually specified driving direction angle and the current position of the unmanned logistics vehicle, combined with the preset driving radius, the spatial point coordinates are calculated through geometric mapping to obtain the target point.

3. The method according to claim 1, characterized in that The method of sampling path points based on the target point according to a plurality of preset radius level parameters and generating a candidate path set by polynomial curve fitting includes: Determining each of the preset radius level parameters according to the target point, and performing radius level division processing using each of the preset radius level parameters to obtain a first radius level, a second radius level, and a third radius level; Based on the current position of the unmanned logistics vehicle and the preset angle range, for the first radius level, the second radius level, and the third radius level, respectively, a first sampling point set, a second sampling point set, and a third sampling point set are generated by uniform angle sampling operations; Performing a Cartesian product combination operation on the first sampling point set, the second sampling point set, and the third sampling point set to obtain a plurality of path endpoint combinations; For each combination of the path endpoints, the current position of the unmanned logistics vehicle is used as the starting point, and a single path is generated by cubic polynomial curve fitting processing, and the single paths are combined to obtain the candidate path set.

4. The method according to claim 1, wherein Calculating the cost of each candidate path in the candidate path set according to the grid map includes: performing Euclidean distance calculation based on the coordinates of the end point of each candidate path and the coordinates of the target point to obtain an original navigation distance value, and normalizing the original navigation distance value to obtain the navigation cost of each candidate path; For each candidate path, uniform sampling point generation processing is performed according to the path curve equation to obtain multiple path position points; Based on each of the path location points, performing a grid value verification process on the grid map to obtain the number of collision points, and calculating the ratio of the number of collision points to the path location points to obtain the collision cost; Obtaining the endpoint coordinates of the optimal path of the previous frame, calculating the original offset distance using Euclidean distance based on the endpoint coordinates of each current candidate path and the endpoint coordinates of the optimal path of the previous frame, and normalizing the original offset distance to obtain the consistency cost of each candidate path; The navigation cost, the collision cost and the consistency cost of each candidate path are weighted and fused according to a preset weight coefficient to obtain the corresponding cost.

5. The method according to claim 4, characterized in that The step of constructing a grid map centered on the unmanned logistics vehicle based on the real-time data includes: Performing grid map initialization processing according to preset grid resolution and map radius parameters to obtain an initialized grid map centered on the unmanned logistics vehicle, wherein the initial value of each grid in the initialized grid map is 0; For each point cloud coordinate in the point cloud data, performing relative coordinate conversion processing according to each point cloud coordinate and the coordinate of the unmanned logistics vehicle to obtain a Cartesian coordinate relative to the unmanned logistics vehicle; Based on the Cartesian coordinates and the preset grid resolution, grid index calculation processing is performed to obtain the row and column indexes of the corresponding grid, and based on the initialized grid map, grid values ​​are updated according to the row and column indexes, and the corresponding grid values ​​are set to 1 to obtain the grid map.

6. The method according to claim 5, characterized in that The grid value verification process is performed on the grid map based on each of the path position points to obtain the number of collision points, including: If the grid value corresponding to each of the path position points in the grid map is 1, the corresponding path position point is marked as a collision point; If the grid value corresponding to each of the path position points in the grid map is 0, the corresponding path position point is marked as a safe point; The number of collision points is obtained by combining the collision points.

7. The method according to claim 1, characterized in that The target point is obtained by one of the following methods: Receiving the absolute coordinate instruction sent by the background manual operation platform, generating the coordinates of the target point according to the spatial position in the absolute coordinate instruction, and obtaining the target point; Alternatively, the target point may be obtained by manually selecting the target point based on the electronic map of the park displayed on the background manual operation platform.

8. A campus unmanned logistics vehicle human-machine co-driving automatic obstacle avoidance system, characterized by: The system comprises: An environmental perception and map construction module, which is used to obtain real-time environmental data of the environment surrounding the unmanned logistics vehicle and construct a grid map centered on the unmanned logistics vehicle based on the real-time environmental data, wherein the real-time environmental data includes image data and point cloud data; A human-computer interaction target point generation module is used to receive a target point after sending the image data to the background manual operation platform, where the target point is the next target arrival location of the unmanned logistics vehicle; A multi-layer path sampling and fitting module is used to sample path points based on the target point according to multiple preset radius level parameters, and generate a candidate path set through polynomial curve fitting; The path evaluation and decision control module is used to calculate the cost of each candidate path in the candidate path set based on the grid map, select the path with the minimum cost as the optimal path, and generate control instructions. The control instructions are used to control the unmanned vehicle to travel along the optimal path. The cost includes navigation cost, collision cost and consistency cost.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images