A continuous learning motion planning method, device, equipment, medium and product

Abstract

The present application discloses a motion planning method, device, equipment, medium and product for continuous learning, which relates to the field of intelligent driving. The method includes: projecting the point cloud data of the current off-road environment onto a grid map to generate a terrain height map and a road roughness map; constructing a guide map based on speed information, and inputting the terrain height map, road roughness map and guide map into the motion planning model after the k-1th update to generate a planned trajectory map of the current off-road environment; obtaining the real trajectory of the unmanned vehicle after driving according to the planned trajectory map, and adding the real trajectory to the sample data set; according to the sample data set of the current off-road environment and the preset historical sample data set, using the mean square error as the loss function and updating the motion planning model for the kth time through the back propagation algorithm. The present application improves the motion planning capability and driving safety of the unmanned vehicle in complex off-road scenarios through continuous learning and iterative optimization.

Description

Technical Field

[0001] The present application relates to the field of intelligent driving, and particularly to a motion planning method, device, equipment, medium and product with continuous learning. Background Art

[0002] Intelligent driving is the product of the deep integration of the automotive industry and information technology. It uses advanced technologies such as sensors, computer vision, and artificial intelligence to enable vehicles to have the ability of autonomous driving. Intelligent driving can not only improve the safety and efficiency of road driving, but also effectively reduce traffic congestion and environmental pollution. It is an important direction for the future development of automobiles.

[0003] In the field of intelligent driving, motion planning, as one of the core modules, is responsible for calculating the optimal or approximately optimal driving trajectory according to the current environmental information and vehicle state to ensure that the vehicle can reach the destination safely and efficiently. Especially in off-road environments, due to the complex terrain, variable obstacles, and continuous changes in environmental factors, the requirements for motion planning algorithms are more stringent.

[0004] The motion planning methods of related technologies have many deficiencies in dealing with off-road scenarios. Traditional rule-based or parameter-based methods often have difficulty adapting to complex and variable terrain features and environmental conditions, resulting in sub-optimal planned paths and even potential safety problems. Although learning-based methods perform well in some scenarios, their generalization ability and robustness are often limited in cases of scarce data or complex environmental features, affecting the overall performance and reliability of the system. Summary of the Invention

[0005] The purpose of the present application is to provide a motion planning method, device, equipment, medium and product with continuous learning, which can handle complex off-road terrain features and generate adaptive planned paths.

[0006] To achieve the above purpose, the present application provides the following solutions.

[0007] In a first aspect, the present application provides a continuous learning motion planning method, including: projecting the point cloud data of the current off-road environment onto the grid map of the current off-road environment to obtain the terrain height map of the current off-road environment; calculating the road surface roughness of the current off-road environment based on the pixel values of the terrain height map of the current off-road environment to obtain the road surface roughness map of the current off-road environment; constructing a guidance map from the starting point to the target point according to the speed information of the unmanned vehicle in the current off-road environment, where the target point is the end point of the global trajectory; inputting the terrain height map, road surface roughness map, and guidance map of the current off-road environment into the motion planning model updated for the (k - 1)th time to obtain the planned trajectory map of the current off-road environment, where k is a natural number greater than or equal to 2, and the previous (k - 1) updates correspond to (k - 1) off-road environments; obtaining the real trajectory after the unmanned vehicle travels according to the planned trajectory map of the current off-road environment, and adding the real trajectory to the sample data set; the sample data set includes the terrain height map, road surface roughness map, guidance map, and the corresponding real trajectory of the current off-road environment; performing the kth update on the motion planning model according to the sample data set of the current off-road environment and the preset historical sample data set, using the mean square error as the loss function and through the backpropagation algorithm, where the preset historical sample data set includes the sample data set of the previous off-road environment different from the current off-road environment.

[0008] Optionally, the terrain height calculation formula of the terrain height map is: 。

[0009] Where M h represents the terrain height of the grid cell, x 、 y respectively represent the abscissa and ordinate of the grid cell, G xy represents the grid cell on the grid map, represents the average height value of all the point cloud data in the grid cell, n xy represents the total number of point cloud data within a grid cell, z i represents the i th height value of the point cloud data, i = 1, 2, 3,..., n xy 。

[0010] Optionally, the road surface roughness calculation formula of the road surface roughness map is: 。

[0011] Where M r represents the roughness of the grid cell, x and yrespectively represent the abscissa and ordinate of the grid cell, represents the mean square error of the height values of all the point cloud data in the grid cell, G xy represents the grid cell on the grid map, represents the average height value of all the point cloud data in the grid cell, n xy represents the total number of point cloud data within a grid cell, z i represents the i th height value of the point cloud data, i = 1, 2, 3, ..., n xy .

[0012] Optionally, the expression of the guidance map is specifically: . Among them, M s represents the pixel value of the coordinate point in the guidance map, E represents the starting point, x ego and y ego respectively represent the abscissa and ordinate of the starting point, G represents the target point, x target and y target respectively represent the abscissa and ordinate of the target point, x j and y j respectively represent the abscissa and ordinate of each coordinate point in the guidance map, j = 1, 2, 3, ... J , J represents the total number of coordinate points in the guidance map, v represents the current speed of the unmanned vehicle.

[0013] Optionally, the constraint condition for using the mean square error as the loss function is: .

[0014] Among them, M represents the preset historical sample data set, represents the motion planning model updated t -1 times based on the sample data set of the current off-road environment, represents the motion planning model updated the tth time based on the sample data set of the current off-road environment, l ( ) represents the loss function of the motion planning model.

[0015] Optionally, the gradient update formula in the backpropagation algorithm is as follows: .

[0016] Among them, is the final updated gradient value of the motion planning model for the current off-road environment, is the updated gradient value of the motion planning model for the current off-road environment, is the average gradient of the motion planning models that have been updated for all off-road environments, is transpose, is transpose.

[0017] In a second aspect, the present application provides a motion planning device for continuous learning, including: a terrain height map acquisition module, a road surface roughness map acquisition module, a guidance map acquisition module, a planned trajectory map acquisition module, an actual trajectory acquisition module, and a model update module.

[0018] The terrain height map acquisition module is configured to project the point cloud data of the current off-road environment onto the grid map of the current off-road environment to obtain the terrain height map of the current off-road environment.

[0019] The road surface roughness map acquisition module is configured to calculate the road surface roughness of the current off-road environment according to the pixel values of the terrain height map of the current off-road environment to obtain the road surface roughness map of the current off-road environment.

[0020] The guidance map acquisition module is configured to construct a guidance map from the starting point to the target point according to the speed information of the unmanned vehicle in the current off-road environment; wherein, the target point is the end point of the global trajectory.

[0021] The planned trajectory map acquisition module is configured to input the terrain height map, the road surface roughness map, and the guidance map of the current off-road environment into the motion planning model updated for the (k - 1)th time to obtain the planned trajectory map of the current off-road environment; wherein, k is a natural number greater than or equal to 2, and the previous (k - 1) updates correspond to (k - 1) off-road environments.

[0022] The actual trajectory acquisition module is configured to obtain the actual trajectory after the unmanned vehicle travels according to the planned trajectory map of the current off-road environment, and add the actual trajectory to the sample data set; the sample data set includes the terrain height map, the road surface roughness map, the guidance map, and the corresponding actual trajectory of the current off-road environment.

[0023] The model update module is configured to perform the kth update on the motion planning model according to the sample data set of the current off-road environment and the preset historical sample data set, using the mean square error as the loss function and through the backpropagation algorithm; wherein, the preset historical sample data set includes the sample data set of the previous off-road environment different from the current off-road environment.

[0024] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the continuous learning motion planning method described in any one of the above.

[0025] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the continuous learning motion planning method described in any one of the above.

[0026] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the continuous learning motion planning method described in any one of the above.

[0027] According to the specific embodiments provided by the present application, the present application has the following technical effects: The present application provides a continuous learning motion planning method, device, equipment, medium and product. By projecting the point cloud data of the current off-road environment onto a grid map, a terrain height map is obtained, which solves the problems of complex terrain in the off-road environment, difficult intuitive understanding and quantitative analysis, realizes the accurate digital expression of terrain information, and provides a solid and reliable data basis for subsequent motion planning; By combining the pixel values of the terrain height map, the road surface roughness of the current off-road environment is calculated, and a road surface roughness map is drawn. At the same time, according to the real-time speed information of the unmanned vehicle, a guidance map from the starting point to the target point is constructed, which solves the challenges of changing road conditions and difficult accurate judgment of driving directions in the off-road environment, realizes the comprehensive evaluation of road conditions and the intelligent optimization of driving paths, and significantly improves the driving safety and driving efficiency of the unmanned vehicle; By inputting key information such as the terrain height map, the road surface roughness map and the guidance map into a motion planning model updated through multiple iterations, a planned trajectory map under the current off-road environment is obtained, and based on the real trajectory after the actual driving of the unmanned vehicle, the model is continuously corrected and optimized, realizing the continuous learning and self-improvement of the motion planning model, solving the problem that traditional motion planning methods are difficult to adapt to the changes of complex off-road environments, and improving the adaptive ability and planning accuracy of the unmanned vehicle in different off-road scenarios. Description of the Drawings

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0029] Figure 1 This is an application environment diagram of a continuous learning motion planning method in an embodiment of the present application.

[0030] Figure 2 This is a schematic flowchart of a continuous learning motion planning method provided in an embodiment of the present application.

[0031] Figure 3 This is a schematic flowchart of a continuous learning motion planning method provided in another embodiment of the present application.

[0032] Figure 4 This is a schematic diagram of functional modules of a continuous learning motion planning device provided in an embodiment of the present application.

[0033] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Detailed implementation manners

[0034] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0035] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the drawings and specific implementation manners.

[0036] In an exemplary embodiment, as Figure 2 shown, a continuous learning motion planning method is provided. This method is executed by a computer device, and specifically, it can be executed independently by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, taking this method applied to Figure 1 the server as an example for illustration, it includes the following steps 101 to step 106.

[0037] Step 101: Project the point cloud data of the current off-road environment onto the grid map of the current off-road environment to obtain the terrain height map of the current off-road environment.

[0038] Step 102: Calculate the road surface roughness of the current off-road environment according to the pixel values of the terrain height map of the current off-road environment to obtain the road surface roughness map of the current off-road environment.

[0039] Step 103: Construct a guidance map from the starting point to the target point according to the speed information of the unmanned vehicle in the current off-road environment; where the target point is the end point of the global trajectory.

[0040] Step 104: Input the terrain height map, road surface roughness map, and guidance map of the current off-road environment into the motion planning model updated for the (k - 1)th time to obtain the planned trajectory map of the current off-road environment; where k is a natural number greater than or equal to 2, and the previous k - 1 updates correspond to k - 1 off-road environments, and any two off-road environments can be the same or different.

[0041] Step 105: Obtain the true trajectory after the unmanned vehicle travels according to the planned trajectory map of the current off-road environment, and add the true trajectory to the sample data set; the sample data set includes the terrain height map, road surface roughness map, guidance map, and the corresponding true trajectory of the current off-road environment.

[0042] Step 106: According to the sample data set of the current off-road environment and the preset historical sample data set, use the mean square error as the loss function and update the motion planning model for the kth time through the backpropagation algorithm; where the preset historical sample data set includes the sample data set of the previous off-road environment different from the current off-road environment.

[0043] Implementing the above steps 101 to 106 can, at each model update, ensure that while learning new scenario data, the average loss of the old scenario is not increased by controlling the constraint conditions, enabling the model to effectively retain and transfer the knowledge it has mastered while adapting to new scenarios.

[0044] In another exemplary embodiment of the present application, as Figure 3 shown, another motion planning method for continuous learning is provided. Taking a certain off-road scenario as an example, the specific implementation steps are as follows.

[0045] Step S1: Environment map and data processing. The collected current off-road environment data is processed into a single-channel environment map and a guidance map. The environment map and the guidance map are used as model inputs, and the trajectory ground truth is used as the label. Various features in the environment are represented in the form of a single-channel map, and the features are distinguished by the pixel values of the map. The represented features such as terrain height, roughness, slope, etc. can be determined according to the main features of the environment and the planning requirements.

[0046] Step S11: Terrain height map . At time , the point cloud data collected by the unmanned vehicle is projected onto the grid map , the x and y coordinates of each point cloud correspond to the grid cell , and the mean value of the z values of the point clouds projected into the grid cell is taken to generate the terrain height map.

[0047] The calculation formula for the terrain height is: .

[0048] WhereM h represents the terrain height of the grid cell x 、 y respectively represent the abscissa and ordinate of the grid cell G xy represents the grid cell on the grid map represents the average height value of all the point cloud data in the grid cell n xy represents the total number of point cloud data within a grid cell z i represents the i height value of the i th point cloud data, where n xy 。

[0049] Finally, the grid map is converted into a single-channel image to obtain a single-channel terrain height map, and the pixel value of the single-channel terrain height map represents the height information of each position.

[0050] Step S12, road surface roughness map According to the mean square deviation of the z values of the point clouds projected onto the grid cells, the road surface roughness at each position is calculated.

[0051] The calculation formula for the road surface roughness is: 。

[0052] where, M r represents the roughness of the grid cell x and y respectively represent the abscissa and ordinate of the grid cell represents the mean square deviation of the height values of all the point cloud data in the grid cell G xy represents the grid cell on the grid map represents the average height value of all the point cloud data in the grid cell n xy represents the total number of point cloud data within a grid cell z i represents the i height value of the i th point cloud data, where n xy 。

[0053] Similarly, the grid map is converted into a single-channel image to obtain a single-channel road surface roughness map, and the pixel value of the single-channel road surface roughness map represents the road surface roughness information of each position.

[0054] Step S13, guidance map 。The guidance map obtains the speed information of the unmanned vehicle through IMU (Inertial Measurement Unit) data and uses it as the background pixel value of the single-channel guidance map. Starting point E : The position point of the ego-vehicle on the map, represented by the pixel value 255, specifically 。Target point G : The intersection point of the global trajectory and the boundary of the grid map, represented by the pixel value 0, specifically 。Other pixels represent the current speed of the unmanned vehicle.

[0055] The expression of the guidance map is: 。

[0056] Among them, M s represents the pixel value of the coordinate point in the guidance map, E represents the starting point, x ego and y ego represent the abscissa and ordinate of the starting point respectively, G represents the target point, x target and y target represent the abscissa and ordinate of the target point respectively, x j and y j represent the abscissa and ordinate of each coordinate point in the guidance map respectively, j =1,2,3,... J , J represents the total number of coordinate points in the guidance map, v represents the current speed of the unmanned vehicle.

[0057] Step S14, Trajectory ground truth 。Obtain the true trajectory of the unmanned vehicle after driving according to the planned trajectory map of the current off-road environment as the trajectory ground truth of the model. The trajectory ground truth is generated from GPS and IMU heading angle data. First, convert the GPS coordinates to UTM (Universal Transverse Mercator Grid System) coordinates , then take the starting point of the trajectory ground truth as the coordinate origin and the heading angle as the y-axis direction to generate the true trajectory: 。

[0058] Finally, at the future discrete time of the unmanned vehicle The trajectory coordinates are used as the true trajectory .

[0059] .

[0060] Among them, is the length of the true trajectory sequence.

[0061] Step S2: Establishment and training of the motion planning model.

[0062] Step S21: Input data formatting. Convert the single-channel terrain height map, single-channel road surface roughness map, and single-channel guidance map into a three-channel RGB image (in this embodiment, there is no need to use a conversion module because it is exactly the same size as the ViT input tensor), and input it into the Vision Transformer (ViT) network.

[0063] First, convert the three-channel RGB image into flattened two-dimensional image patches: . Among them, is the size of the image patch, is the number of image patches.

[0064] Step S22: ViT network structure. The Transformer encoder is alternately composed of a multi-head self-attention mechanism and a multilayer perceptron (MLP). Each block uses layer normalization (LN) and adds a residual connection to improve the training efficiency and stability of the motion planning model. The features output by the encoder are input into a fully connected layer to generate a planned trajectory .

[0065] The output format of the planned trajectory is: .

[0066] Step S23: Compare the predicted trajectory output by the motion planning model with the true trajectory, and use the backpropagation algorithm to update the parameters to optimize the model convergence. The loss function can use the mean squared error (MSE) or other applicable loss functions to ensure that the model gradually improves the prediction accuracy during the training process.

[0067] Through the above steps, the constructed motion planning model can effectively learn the motion planning tasks in complex off-road environments and achieve good generalization ability after training.

[0068] Step S3: Multi-environment adaptation based on continuous learning. To address the "catastrophic forgetting" problem, a continuous learning algorithm is introduced to ensure that the knowledge already mastered is not affected when learning new scenario data. Specifically, when updating the model each time, constraint conditions are set to ensure that the average loss of the old scenarios does not increase. This constraint can be expressed by the following formula.

[0069] .

[0070] Among them, M represents the preset historical sample data set, represents the motion planning model that has been updated t -1 times based on the sample data set of the current off-road environment, represents the motion planning model that has been updated t times based on the sample data set of the current off-road environment, l ( ) represents the loss function of the motion planning model.

[0071] At the gradient update level, the finally updated gradient is expressed as: .

[0072] Among them, is the final updated gradient value of the motion planning model for the current off-road environment, is the updated gradient value of the motion planning model for the current off-road environment, is the average gradient of the motion planning models that have been updated for all off-road environments, is transpose of, is transpose of. This mechanism ensures the balance between old and new knowledge and improves the adaptability and robustness of the model.

[0073] Repeat the above steps. This embodiment realizes multi-scenario adaptive motion planning in complex off-road environments through the combination of environmental data processing, deep learning model training, and continuous learning algorithms. By introducing a continuous learning algorithm, this embodiment enables the model to adapt to complex changes in different scenarios without losing the knowledge already mastered. This method classifies and processes multiple key elements in the off-road environment, effectively enhancing the comprehensive understanding of the environment and the planning accuracy. During the process of dealing with multiple scenarios, the model realizes the improvement of robustness and adaptability and can maintain high stability in the case of scarce data or new environments.

[0074] The beneficial effects of this application are as follows: (1) The end-to-end motion planning model constructed in this application can handle complex off-road terrain features and generate adaptive planned paths. The motion planning model can accurately analyze environmental data, respond in real time to dynamic changes and complex environmental conditions in off-road scenarios, and ensure that the unmanned vehicle can safely and efficiently plan its motion path under various terrains. (2) The motion planning model constructed in this application exhibits good generalization ability. The motion planning model can transfer the knowledge learned from one off-road scenario to a new off-road scenario. Even when encountering data-scarce or unseen off-road scenarios, the motion planning model can still maintain a good planning effect. This enables the motion planning method of continuous learning in multiple off-road scenarios to have wide application adaptability and can work effectively in a changing off-road environment. (3) The continuous learning algorithm proposed in this application enables the motion planning model to effectively retain existing knowledge during updates and avoid "catastrophic forgetting" when learning new off-road scenarios. This mechanism enables the motion planning model to continuously adapt to new off-road scenarios without losing its grasp of old off-road scenarios, thereby improving the robustness and long-term adaptability of the motion planning model in multiple off-road scenarios.

[0075] This application also provides an application scenario that applies the above-mentioned motion planning method of continuous learning. Specifically: The motion planning method of continuous learning provided in this embodiment can be applied in the scenario of unmanned vehicle autonomous exploration. The scenario of unmanned vehicle autonomous exploration includes an environmental perception link, a path planning link, and an execution navigation link; the unmanned vehicle enters the path planning link from the environmental perception link, generates a safe driving path by processing and analyzing the perceived off-road environmental data, and enters the execution navigation link for actual driving operations. The motion planning method of continuous learning provided in this embodiment belongs to the path planning link in unmanned vehicle autonomous exploration. Specifically, during the process of unmanned vehicle autonomous exploration, the motion planning model can be continuously updated and optimized based on the real-time collected environmental data to adapt to the changing off-road environment and ensure that the unmanned vehicle can safely and efficiently reach the destination.

[0076] Based on the same inventive concept, the embodiments of this application also provide a motion planning device for continuous learning in multiple off-road scenarios for implementing the above-mentioned motion planning method of continuous learning. The implementation solutions provided by this device to solve problems are similar to those described in the above method. Therefore, the specific limitations in one or more of the following embodiments of the motion planning device for continuous learning can refer to the limitations on the motion planning method of continuous learning in the above text and will not be elaborated here.

[0077] In an exemplary embodiment, as Figure 4As shown in the figure, a motion planning device for continuous learning is provided, including: a terrain height map acquisition module 201, a road surface roughness map acquisition module 202, a guidance map acquisition module 203, a planned trajectory map acquisition module 204, an actual trajectory acquisition module 205, and a model update module 206.

[0078] The terrain height map acquisition module 201 is configured to project the point cloud data of the current off-road environment onto the grid map of the current off-road environment to obtain the terrain height map of the current off-road environment.

[0079] The road surface roughness map acquisition module 202 is configured to calculate the road surface roughness of the current off-road environment according to the pixel values of the terrain height map of the current off-road environment to obtain the road surface roughness map of the current off-road environment.

[0080] The guidance map acquisition module 203 is configured to construct a guidance map from the starting point to the target point according to the speed information of the unmanned vehicle in the current off-road environment; wherein, the target point is the end point of the global trajectory.

[0081] The planned trajectory map acquisition module 204 is configured to input the terrain height map, the road surface roughness map, and the guidance map of the current off-road environment into the motion planning model updated for the (k - 1)th time to obtain the planned trajectory map of the current off-road environment; wherein, k is a natural number greater than or equal to 2, and the previous (k - 1) updates correspond to (k - 1) off-road environments.

[0082] The actual trajectory acquisition module 205 is configured to obtain the actual trajectory after the unmanned vehicle travels according to the planned trajectory map of the current off-road environment, and add the actual trajectory to the sample data set; the sample data set includes the terrain height map, the road surface roughness map, the guidance map, and the corresponding actual trajectory of the current off-road environment.

[0083] The model update module 206 is configured to perform the kth update on the motion planning model according to the sample data set of the current off-road environment and the preset historical sample data set, using the mean square error as the loss function and through the backpropagation algorithm; wherein, the preset historical sample data set includes the sample data set of the previous off-road environment different from the current off-road environment.

[0084] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 5As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the sample dataset data of the motion planning model for continuous learning in multiple off-road scenarios. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a motion planning method for continuous learning.

[0085] Those skilled in the art can understand that Figure 5 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements the steps in the above method embodiments.

[0086] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0087] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0088] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0089] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the various embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0090] The databases involved in the various embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the various embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0091] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0092] In this article, specific examples are used to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A motion planning method for continuous learning, characterized in that: The motion planning method of continuous learning includes: Projecting the point cloud data of the current off-road environment onto the grid map of the current off-road environment to obtain a terrain height map of the current off-road environment; Calculate the road surface roughness of the current off-road environment according to the pixel values ​​of the terrain height map of the current off-road environment to obtain a road surface roughness map of the current off-road environment; According to the speed information of the unmanned vehicle in the current off-road environment, a guidance map from the starting point to the target point is constructed; wherein the target point is the end point of the global trajectory; The terrain height map, road roughness map and guide map of the current off-road environment are input into the motion planning model after the k-1th update to obtain the planned trajectory map of the current off-road environment, specifically including: the motion planning model includes a ViT network and a fully connected layer connected in sequence; the Transformer encoder in the ViT network is composed of a multi-head self-attention mechanism and a multi-layer perceptron alternately, and each block uses layer normalization and adds residual connections; the single-channel terrain height map, the single-channel road roughness map and the single-channel guide map are converted into a three-channel RGB image; the three-channel RGB image Convert to A flattened 2D image patch: , and input into the ViT network; the features output by the ViT network are input into the fully connected layer to obtain the planned trajectory map of the current off-road environment; where k is a natural number greater than or equal to 2, and the first k-1 updates correspond to k-1 off-road environments; is the size of the image block, is the number of image blocks; Obtain the actual trajectory of the unmanned vehicle after driving according to the planned trajectory map of the current off-road environment, and add the actual trajectory to the sample data set; the sample data set includes a terrain height map, a road roughness map, a guidance map and the corresponding actual trajectory of the current off-road environment; According to the sample data set of the current off-road environment and the preset historical sample data set, the motion planning model is updated for the kth time using the mean square error as the loss function and the back propagation algorithm; wherein the preset historical sample data set includes a sample data set of a previous off-road environment that is different from the current off-road environment.

2. The motion planning method of continuous learning according to claim 1, characterized in that: The terrain height calculation formula of the terrain height map is: ； in, M h represents the topographic height of the grid cell, x and y Respectively represent the horizontal and vertical coordinates of the grid cell, G xy Represents a grid cell on a grid map. Represents the average height value of all point cloud data in the grid cell. n xy Represents the total number of point cloud data in a grid cell. z i Indicates i The height value of the point cloud data, i =1,2,3,..., n xy .

3. The motion planning method of continuous learning according to claim 1, characterized in that: The road roughness calculation formula of the road roughness map is: ； in, M r represents the roughness of the grid cell, x and y Respectively represent the horizontal and vertical coordinates of the grid cell, It represents the mean square error of the height values ​​of all point cloud data in the grid cell. G xy Represents a grid cell on a grid map. Represents the average height value of all point cloud data in the grid cell. n xy Represents the total number of point cloud data in a grid cell. z i Indicates i The height value of the point cloud data, i =1,2,3,..., n xy .

4. The motion planning method of continuous learning according to claim 1, characterized in that: The expression of the guide map is specifically: ； in, M s Represents the pixel value of the coordinate point in the guide map, E Indicates the starting point, x ego and y ego Represent the horizontal and vertical coordinates of the starting point, respectively. G represents the target point, x target and y target Respectively represent the horizontal and vertical coordinates of the target point, x j and y j Respectively represent the horizontal and vertical coordinates of each coordinate point in the guide map, j =1,2,3,... J , J Indicates the total number of coordinate points in the guide map. v Indicates the current speed of the driverless car.

5. The motion planning method of continuous learning according to claim 1, characterized in that: The constraints for using mean square error as the loss function are: ； in, M Represents the preset historical sample data set, Indicates that the sample dataset based on the current off-road environment has been updated t -1 motion planning model, Indicates that the sample dataset based on the current off-road environment has been updated t The motion planning model l ( ) represents the loss function of the motion planning model.

6. The motion planning method of continuous learning according to claim 1, characterized in that: The gradient update formula in the back propagation algorithm is: ； in, is the final updated gradient value of the motion planning model in the current off-road environment, is the gradient value updated by the motion planning model for the current off-road environment, is the average gradient of the updated motion planning model for all off-road environments, for The transpose of for The transpose of .

7. A motion planning device for continuous learning, characterized in that: The motion planning device for continuous learning comprises: A terrain height map acquisition module is used to project the point cloud data of the current off-road environment onto a grid map of the current off-road environment to obtain a terrain height map of the current off-road environment; A road surface roughness map acquisition module is used to calculate the road surface roughness of the current off-road environment according to the pixel values ​​of the terrain height map of the current off-road environment, and obtain the road surface roughness map of the current off-road environment; The guidance map acquisition module is used to construct a guidance map from the starting point to the target point according to the speed information of the unmanned vehicle in the current off-road environment; wherein the target point is the end point of the global trajectory; The planning trajectory map acquisition module is used to input the terrain height map, road roughness map and guide map of the current off-road environment into the motion planning model after the k-1th update to obtain the planning trajectory map of the current off-road environment, specifically including: the motion planning model includes a ViT network and a fully connected layer connected in sequence; the Transformer encoder in the ViT network is composed of a multi-head self-attention mechanism and a multi-layer perceptron alternately, and each block uses layer normalization and adds residual connections; the single-channel terrain height map, the single-channel road roughness map and the single-channel guide map are converted into a three-channel RGB image; the three-channel RGB image Convert to A flattened 2D image patch: , and input into the ViT network; the features output by the ViT network are input into the fully connected layer to obtain the planned trajectory map of the current off-road environment; where k is a natural number greater than or equal to 2, and the first k-1 updates correspond to k-1 off-road environments; is the size of the image block, is the number of image blocks; The real trajectory acquisition module is used to obtain the real trajectory of the unmanned vehicle after driving according to the planned trajectory map of the current off-road environment, and add the real trajectory to the sample data set; the sample data set includes the terrain height map, road roughness map, guidance map and corresponding real trajectory of the current off-road environment; A model updating module is used to update the motion planning model for the kth time based on the sample data set of the current off-road environment and the preset historical sample data set, using the mean square error as the loss function and through the back propagation algorithm; wherein the preset historical sample data set includes sample data of the previous off-road environment that is different from the current off-road environment.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the motion planning method for continuous learning according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the motion planning method of continuous learning described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the motion planning method of continuous learning described in any one of claims 1 to 6 is implemented.

Images