Path planning method, traffic path planning method and path planning model training method

Abstract

Provided in the embodiments of the present disclosure are a path planning method, a traffic path planning method and a path planning model training method. The path planning method comprises: acquiring path planning requirement information, wherein the path planning requirement information comprises a path starting point and a path ending point; inputting the path starting point and the path ending point into a path planning model, so as to obtain a plurality of reference paths between the path starting point and the path ending point, and first decision information corresponding to the plurality of reference paths, respectively, wherein the path planning model is obtained by means of performing training on the basis of a path prediction loss and a constraint condition loss; and on the basis of the plurality of reference paths and the first decision information, determining a target planned path corresponding to the path planning requirement information. A target planned path is automatically generated by using a path planning model, the path planning link is simple, and thus the stability and service performance of path planning are improved; in addition, since there is no need to manually design different map road network weights during path planning, the efficiency and scalability of path planning are improved.

Description

[0001]Path Planning, Traffic Path Planning, and Path Planning Model Training Method. This disclosure claims priority to Chinese patent application number 202410040375, filed with the China Patent Office on January 9, 2024. The entire contents of this application are incorporated herein by reference. Technical Field: The embodiments of this disclosure relate to the field of computer technology, and more particularly to path planning, traffic path planning, and path planning model training methods. Background: With the development of internet technology, path planning problems are increasingly relying on the internet. Path planning is a key technology in many fields, including transportation, logistics, and robotics. The research and application of related path planning methods have long been a hot topic. Currently, roads are typically manually weighted based on real-time road conditions, and path planning is performed based on the weight of each road. However, the map network involved in actual road planning is complex and diverse. The road weighting process requires the design of different weighting strategies based on the actual map network, resulting in low path planning efficiency. Therefore, an efficient path planning solution is urgently needed. SUMMARY OF THE INVENTION In view of this, embodiments of the present disclosure provide a path planning method. One or more embodiments of the present disclosure also relate to a traffic path planning method, a path planning model training method, a path planning device, a traffic path planning device, a path planning model training device, a computing device, a computer-readable storage medium, and a computer program to address technical deficiencies in the prior art. According to a first aspect of an embodiment of the present disclosure, a path planning method is provided, comprising: obtaining path planning requirement information, wherein the path planning requirement information includes a path start point and a path end point; inputting the path start point and the path end point into a path planning model, obtaining multiple reference paths between the path start point and the path end point, and first decision information corresponding to each of the multiple reference paths, wherein the path planning model is trained based on path prediction loss and constraint condition loss; and determining a target planning path corresponding to the path planning requirement information based on the multiple reference paths and the first decision information. According to a second aspect of an embodiment of the present disclosure, a traffic route planning method is provided, comprising: obtaining traffic route planning requirement information, wherein the traffic route planning requirement information includes a route start point and a route end point; inputting the route start point and the route end point into a route planning model to obtain multiple reference routes between the route start point and the route end point and first decision information corresponding to each of the multiple reference routes, wherein the route planning model is trained based on path prediction loss and constraint condition loss; and determining a target traffic planning route corresponding to the traffic route planning requirement information based on the multiple reference routes and the first decision information.According to a third aspect of an embodiment of the present disclosure, a path planning model training method is provided, which is applied to a cloud-side device, including: obtaining multiple historical paths; inputting the multiple historical paths into an initial path planning model to obtain predicted recovery paths corresponding to the multiple historical paths; calculating path prediction loss based on the multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths; calculating constraint loss based on path planning constraints and the predicted recovery paths corresponding to the multiple historical paths; and adjusting model parameters of the initial path planning model based on the path prediction loss and the constraint loss to obtain a trained path planning model. According to a fourth aspect of an embodiment of the present disclosure, a path planning device is provided, comprising: a first acquisition module configured to acquire path planning requirement information, wherein the path planning requirement information includes a path start point and a path end point; a first input module configured to input the path start point and the path end point into a path planning model to obtain multiple reference paths between the path start point and the path end point and first decision information corresponding to the multiple reference paths, wherein the path planning model is trained based on path prediction loss and constraint condition loss; and a first determination module configured to determine a target planning path corresponding to the path planning requirement information based on the multiple reference paths and the first decision information. According to a fifth aspect of an embodiment of the present disclosure, a traffic path planning apparatus is provided, comprising: a second acquisition module configured to acquire traffic path planning requirement information, wherein the traffic path planning requirement information includes a path start point and a path end point; a second input module configured to input the path start point and the path end point into a path planning model to obtain multiple reference paths between the path start point and the path end point and first decision information corresponding to the multiple reference paths, wherein the path planning model is trained based on path prediction loss and constraint condition loss; and a second determination module configured to determine a target traffic planning path corresponding to the traffic path planning requirement information based on the multiple reference paths and the first decision information. According to a sixth aspect of an embodiment of the present disclosure, a path planning model training device is provided, which is applied to a cloud-side device, and includes: a third acquisition module, configured to acquire multiple historical paths; a third input module, configured to input the multiple historical paths into the initial path planning model, and obtain predicted recovery paths corresponding to the multiple historical paths; a first calculation module, configured to calculate the path prediction loss based on the multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths; a second calculation module, configured to calculate the constraint condition loss based on the path planning constraints and the predicted recovery paths corresponding to the multiple historical paths; a first adjustment module, configured to adjust the model parameters of the initial path planning model based on the path prediction loss and the constraint condition loss, and obtain a trained path planning model.According to a seventh aspect of an embodiment of the present disclosure, a computing device is provided, comprising: a memory and a processor; the memory is configured to store computer-executable instructions, and the processor is configured to execute the computer-executable instructions, wherein when the computer-executable instructions are executed by the processor, the steps of the method provided in the first, second, or third aspect are implemented. According to an eighth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, the medium storing computer-executable instructions, wherein when the instructions are executed by the processor, the steps of the method provided in the first, second, or third aspect are implemented. According to a ninth aspect of an embodiment of the present disclosure, a computer program is provided, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the method provided in the first, second, or third aspect. One embodiment of the present disclosure provides a path planning method, comprising: obtaining path planning requirement information, wherein the path planning requirement information includes a path start point and a path end point; inputting the path start point and the path end point into a path planning model to obtain multiple reference paths between the path start point and the path end point and first decision information corresponding to each of the multiple reference paths, wherein the path planning model is trained based on path prediction loss and constraint loss; and determining a target planning path corresponding to the path planning requirement information based on the multiple reference paths and the first decision information. Because constraint loss is considered during the path planning model training process, prior knowledge is incorporated into the training process, improving the accuracy of the path planning model. Therefore, by automatically generating the target planning path using the path planning model, the path planning chain is simplified, ensuring the stability and service performance of the path planning. Furthermore, since the path planning process eliminates the need to manually design different map road network weights, the efficiency and scalability of path planning are improved.BRIEF DESCRIPTION OF THE DRAWINGS FIG1 is a flow chart of a traditional path planning method; FIG2 is an architecture diagram of a path planning system provided by an embodiment of the present disclosure; FIG3 is an architecture diagram of another path planning system provided by an embodiment of the present disclosure; FIG4 is a flow chart of a path planning method provided by an embodiment of the present disclosure; FIG5 is a flow chart of training a path planning model in a path planning method provided by an embodiment of the present disclosure; FIG6 is a flow chart of training a conditional generation model in a path planning method provided by an embodiment of the present disclosure; FIG7 is a flow chart of a traffic path planning method provided by an embodiment of the present disclosure; FIG8 is a flow chart of a path planning model training method provided by an embodiment of the present disclosure; FIG9 is a flow chart of a processing training process of a path planning method provided by an embodiment of the present disclosure; FIG10 is a flow chart of a processing training process of another path planning method provided by an embodiment of the present disclosure; FIG11 is a structural schematic diagram of a path planning device provided by an embodiment of the present disclosure; FIG12 is a structural schematic diagram of a traffic path planning device provided by an embodiment of the present disclosure; FIG13 is a structural schematic diagram of a path planning model training device provided by an embodiment of the present disclosure; FIG14 is a structural block diagram of a computing device provided by an embodiment of the present disclosure. The following description sets forth numerous specific details to facilitate a thorough understanding of the present disclosure. However, the present disclosure can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without departing from the scope of the present disclosure. Therefore, the present disclosure is not limited to the specific implementations disclosed below. The terminology used in one or more embodiments of the present disclosure is intended solely for the purpose of describing specific embodiments and is not intended to limit the present disclosure. As used in one or more embodiments of the present disclosure and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of the present disclosure refers to and encompasses any and all possible combinations of one or more of the associated listed items. It should be understood that while the terms "first," "second," and so on may be used to describe various information in one or more embodiments of the present disclosure, such information should not be limited to these terms. These terms are used solely to distinguish information of the same type from one another. For example, "first" could be referred to as "second," and similarly, "second" could be referred to as "first" without departing from the scope of one or more embodiments of the present disclosure. The word "if" as used herein may be interpreted as "when" or "when" or "in response to determining," depending on the context.Furthermore, 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 used for analysis, storage, and display) involved in one or more embodiments of this disclosure are all authorized by the user or fully authorized by all parties. The collection, use, and processing of the relevant data must comply with the relevant laws, regulations, and standards of the relevant region, and corresponding operation portals are provided for users to choose to authorize or refuse. First, the terms used in one or more embodiments of this disclosure are explained. Generative Artificial Intelligence: Generative Artificial Intelligence (AIGC) refers to AI technologies and methods based on generative adversarial networks and large pre-trained models. It generates relevant content with appropriate generalization capabilities by learning and recognizing existing data. Diffusion Model: The Diffusion Model (DM) is an advanced generative model that focuses on generating high-quality samples and shows great potential in complex generative tasks. Diffusion models are essentially based on probability distributions and use a series of iterative steps to gradually construct samples. The core mechanism of the diffusion model draws on Markov Chain Monte Carlo (MCMC) sampling. At each iteration, the model predicts the next state based on the current state. Through continuous local transformations, the diffusion model gradually "diffuses" a simple initial state into a complex sample that conforms to the target probability distribution. More specifically, the diffusion model defines a transition probability function that guides the evolution from the current state to the next state. During successive iterations, the diffusion model proposes new candidate samples based on the current state and continuously adjusts these samples to bring them closer to the desired target probability distribution. Diffusion Generation: Diffusion Generation is a typical AI generation technology. Based on the diffusion model, it continuously adds noise to a piece of content data to generate a set of data. This data is then used to train a deep neural network, enabling the network to generate new content. Travel time prediction: Travel time prediction (ETA) is an important technical application in the field of transportation. It estimates the estimated time required for a vehicle to reach its destination from its current location through complex algorithm models such as real-time or historical data, road conditions, and weather conditions.Variational Autoencoder: A variational autoencoder (VAE) is a deep learning model primarily used for probabilistic generation and unsupervised learning. Combining the concepts of autoencoders with variational inference, VAEs can model data distributions and generate new data samples similar to the training data. Flow Models: Flow-based models are a category of generative deep learning methods primarily used for probability density modeling and efficient sampling of random variables. In machine learning, flow models transform simple probability distributions into complex underlying data distributions through a series of reversible transformations. Specifically, flow models gradually transform an underlying distribution (such as a standard normal distribution or uniform distribution) into a complex distribution that matches the observations in a dataset by designing a set of continuous, differentiable, and easily sampleable variable-to-variable mappings (typically implemented using neural networks). Refer to Figure 1, which shows a flowchart of a traditional route planning method. As shown in Figure 1, route planning in traditional map services, based on the map road network, first requires real-time collection of historical routes of various vehicles from multiple sources. Then, based on these historical routes, real-time traffic conditions are calculated. Specifically, real-time vehicle speed and traffic volume information is calculated for each road segment. Road segments are weighted based on real-time traffic conditions, with the weight typically representing the travel time for each segment. Finally, based on the weighted network obtained by this assignment, a shortest path planning algorithm is used to plan routes based on the start and end points and the weighted network, resulting in a planned route. Furthermore, the route planning process can also reference the start and end travel times obtained through ETA predictions to determine the final planned route. However, traditional path planning methods require pre-calculation of real-time weights for the entire map network. Road weighting requires designing different weighting strategies based on the actual map network, resulting in low path planning efficiency and poor scalability. To address these issues, embodiments of the present disclosure propose a machine learning-based path planning solution. This solution eliminates the need for pre-calculation of real-time weights for the entire map network. Instead, it trains a machine learning model, namely a path planning model, using a large number of historical paths. This model then implements fully automated path planning. Specifically, path planning requirement information is obtained, including a path start point and a path end point. The path start point and path end point are input into the path planning model, and multiple reference paths between the path start point and the path end point, along with first decision information corresponding to each of the reference paths, are obtained. The path planning model is trained based on path prediction loss and constraint condition loss. Based on the multiple reference paths and the first decision information, a target planning path corresponding to the path planning requirement information is determined.It is worth noting that, since constraint loss is considered during the path planning model training process, prior knowledge is incorporated into the training process, thereby improving the accuracy of the path planning model. Therefore, by automatically generating the target planning path using the path planning model, the path planning link is simple, ensuring the stability and service performance of the path planning. In addition, since different map road network weights do not need to be manually designed during the path planning process, the efficiency and scalability of the path planning are improved. In the present disclosure, a path planning method is provided. The present disclosure also relates to a traffic path planning method, a path planning model training method, a path planning device, a traffic path planning device, a path planning model training device, a computing device, and a computer-readable storage medium, each of which is described in detail in the following embodiments. 2 shows an architecture diagram of a path planning system provided by an embodiment of the present disclosure. The path planning system may include a client 100 and a server 200. The client 100 is configured to send path planning requirement information to the server 200, wherein the path planning requirement information includes a path start point and a path end point. The server 200 is configured to input the path start point and the path end point into a path planning model, obtain multiple reference paths between the path start point and the path end point, and first decision information corresponding to the multiple reference paths, wherein the path planning model is trained based on path prediction loss and constraint condition loss. The server 200 is configured to determine a target planning path corresponding to the path planning requirement information based on the multiple reference paths and the first decision information, and send the target planning path to the client 100. The client 100 is also configured to receive the target planning path sent by the server 200. By applying the solution of the embodiments of the present disclosure, constraint loss is considered during the path planning model training process, thereby incorporating prior knowledge into the training process and improving the accuracy of the path planning model. Consequently, by automatically generating the target planning path using the path planning model, the path planning process is simplified, ensuring path planning stability and service performance. Furthermore, since the path planning process eliminates the need to manually design different map and road network weights, the efficiency and scalability of path planning are improved. Referring to Figure 3, it shows an architecture diagram of another path planning system provided by one embodiment of the present disclosure. The path planning system may include multiple clients 100 and a server 200. The clients 100 may comprise end-side devices, and the server 200 may comprise cloud-side devices.Multiple clients 100 can establish communication connections through a server 200. In a route planning scenario, the server 200 is used to provide route planning services between multiple clients 100. Multiple clients 100 can act as senders or receivers, communicating through the server 200. Users can interact with the server 200 through the client 100 to receive data from other clients 100 or send data to other clients 100. In a route planning scenario, a user can publish a data stream to the server 200 through the client 100. The server 200 generates a target planned path based on the data stream and pushes the target planned path to other clients with whom communication has been established. The connection between the client 100 and the server 200 is established through a network. The network provides the medium for the communication link between the client 100 and the server 200. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. The data transmitted by the client 100 may need to be encoded, transcoded, compressed, or otherwise processed before being published to the server 200. The client 100 may be a browser, an APP (Application Program), a web application such as an H5 (HyperText Markup Languages, version 5) application, a light application (also known as a mini-program, a lightweight application), or a cloud application. The client 100 may be developed based on a software development kit (SDK) for a corresponding service provided by the server 200, such as a real-time communication (RTC) SDK. The client 100 may be deployed in an electronic device and may rely on the device or certain APPs in the device to operate. For example, the electronic device may have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, a tablet computer, or a personal computer. Electronic devices may also typically be configured with various other types of applications, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social networking platform software, etc. The server 200 may include servers that provide various services, such as servers that provide communication services to multiple clients, servers that support background training for models used on clients, and servers that process data sent by clients.It should be noted that the server 200 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server in a distributed system, or a server integrated with blockchain. The server can also be a cloud server for basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. It is worth noting that the path planning method provided in the embodiments of the present disclosure is generally executed by the server. However, in other embodiments of the present disclosure, the client can also have similar functions to the server and thus execute the path planning method provided in the embodiments of the present disclosure. In other embodiments, the path planning method provided in the embodiments of the present disclosure can also be jointly executed by the client and the server. Referring to FIG. 4 , FIG. 4 shows a flow chart of a path planning method provided in one embodiment of the present disclosure, which specifically includes the following steps: Step 402: Obtain path planning requirement information, where the path planning requirement information includes a path starting point and a path end point. In one or more embodiments of the present disclosure, during route planning, route planning requirements information can be obtained, thereby generating an accurate target planned route based on the route planning requirements information. Specifically, the route planning requirements information is used to characterize the user's route planning intent. The route planning requirements information can be route planning requirements information for different scenarios, such as virtual game scenarios or intelligent transportation scenarios. The route planning requirements information includes, but is not limited to, the route starting point, the route ending point, and the route planning conditions. The route planning conditions include the user's personalized planning information, such as travel time, toll information, terrain conditions, and so on. The route starting point refers to the starting location of the route. The route ending point refers to the target location or destination of the route. The route starting point and the route ending point can be locations on a map, such as a certain building, or geographic coordinates, depending on the specific circumstances, and the present disclosure does not impose any restrictions on this. The route planning method can be applied to and used in various application scenarios that utilize map services, such as electronic maps, navigation, intelligent transportation, travel services, smart cities, and other applications. In actual applications, there are many ways to obtain route planning requirements information, and the selection can be based on the specific circumstances, and the present disclosure does not impose any restrictions on this. In a possible implementation of the present disclosure, path planning requirement information sent by a user may be received.In another possible implementation of the present disclosure, path planning requirement information can be read from other data acquisition devices or databases. Step 404: Input the path start point and path end point into the path planning model to obtain multiple reference paths between the path start point and the path end point, and first decision information corresponding to each of the multiple reference paths. The path planning model is trained based on path prediction loss and constraint loss. In one or more embodiments of the present disclosure, after obtaining the path planning requirement information, the path start point and path end point can be further input into the path planning model to obtain multiple reference paths between the path start point and the path end point, and first decision information corresponding to each of the multiple reference paths. The path planning model is trained based on path prediction loss and constraint loss. Specifically, the path planning model can be various generative models trained based on massive historical paths, such as variational autoencoders, diffusion models, flow models, and so on. Since path generation conditions are not incorporated into the path planning model training process, the path planning model can be understood as an unconditional generative model. During the path planning model training process, the model parameters of the initial path planning model can be adjusted based on the path prediction loss and constraint loss. The path prediction loss describes the deviation between the predicted recovery path generated by the initial path planning model and the historical path. The constraint loss describes the deviation between the predicted recovery path generated by the initial path planning model and the path planning constraints. A reference path can be understood as a continuous map trajectory between the path start point and the path end point. The first decision information represents the likelihood of selecting the corresponding reference path and can be a decision probability. It should be noted that the path start point and the path end point are input into the path planning model, which can perform restoration and prediction on the noisy image of the map road network to obtain multiple reference paths between the path start point and the path end point and the first decision information corresponding to each of the multiple reference paths. Step 406: Determine a target planned path corresponding to the path planning requirement information based on the multiple reference paths and the first decision information. In one or more embodiments of the present disclosure, after obtaining path planning requirement information, inputting the path start point and the path end point into the path planning model, and obtaining multiple reference paths between the path start point and the path end point and the first decision information corresponding to each of the multiple reference paths, the target planned path corresponding to the path planning requirement information can be further determined based on the multiple reference paths and the first decision information. Specifically, the target planned path refers to a planned path that meets the path planning requirement information. The target planned path may be a road segment excluding the path endpoint, or a complete path including the path endpoint. The selection is made based on actual conditions, and the embodiments of the present disclosure do not impose any limitation on this.In practical applications, there are various methods for determining the target planned path corresponding to the path planning requirement information based on multiple reference paths and the first decision information. The method of selection depends on actual circumstances and is not limited in the present embodiment. In one possible implementation of the present disclosure, the multiple reference paths can be ranked based on the first decision information corresponding to each reference path, and the reference path with the largest first decision information and a higher ranking can be determined as the target planned path. In another possible implementation of the present disclosure, a preset decision condition can be obtained, the first decision information corresponding to each of the multiple reference paths can be compared with the preset decision condition, and the target planned path can be randomly selected from the reference paths that meet the preset decision condition. The preset decision condition, such as a decision probability greater than 0.8, can be set based on actual circumstances. Applying the solution of the embodiments of the present disclosure, since constraint loss is considered during the path planning model training process, prior knowledge is incorporated into the training process, improving the accuracy of the path planning model. Consequently, by automatically generating the target planned path using the path planning model, the path planning process is simplified, ensuring path planning stability and service performance. Furthermore, since the path planning process eliminates the need to manually design different map road network weights, the efficiency and scalability of path planning are improved. In an optional embodiment of the present disclosure, after determining the target planned path corresponding to the path planning requirement information based on multiple reference paths and the first decision information, the following steps may be included: receiving path adjustment information sent by the user based on the target planned path, and adjusting the model parameters of the path planning model based on the path adjustment information. It should be noted that after determining the target planned path, the target planned path may be sent to the user. After receiving the target planned path, the user may determine whether the target planned path meets the actual requirements. If the target planned path does not meet the actual requirements, path adjustment information may be sent based on the target planned path. The path adjustment information includes, but is not limited to, the target planned path updated by the user based on actual needs, the updated path starting point, and the updated path ending point. The specific selection depends on actual circumstances and is not limited in this embodiment of the present disclosure. Furthermore, after receiving the video adjustment information sent by the user based on the target video, model parameters of the path planning model can be adjusted based on the path adjustment information, or model parameters of the conditional generation model can be adjusted based on the path adjustment information. The method for adjusting the model parameters of the path planning model or the conditional generation model can refer to the training method of the path planning model or the training method of the conditional generation model, and will not be further described in this embodiment of the present disclosure.Applying the solution of the embodiments of the present disclosure, a method of receiving path adjustment information sent by a user based on a target planned path and adjusting the model parameters of a path planning model based on the path adjustment information enables data interaction with the user and improves the user experience. In an optional embodiment of the present disclosure, since the user's path planning requirements during path planning may not be limited to the path starting point and path ending point, path planning can be performed based on additional path planning conditions. Specifically, the path planning requirement information includes the path planning conditions. After inputting the path starting point and path ending point into the path planning model and obtaining multiple reference paths between the path starting point and path ending point and first decision information corresponding to each of the multiple reference paths, the method may further include the following steps: inputting the path starting point, path ending point, and path planning conditions into a condition generation model to obtain multiple candidate paths and second decision information corresponding to each of the multiple candidate paths; and determining the target planned path corresponding to the path planning requirement information based on the multiple reference paths and the first decision information. This may include the following steps: determining the target planned path based on the multiple reference paths, the multiple candidate paths, the first decision information, and the second decision information. Specifically, the path planning conditions refer to a series of conditional factors that need to be considered during the path planning process. Path planning conditions include the user's personalized planning information, such as travel time, toll information, and terrain conditions. A candidate path refers to a path obtained under the path planning conditions and the constraints of the path starting point and path ending point. A candidate path can be a road segment excluding the path ending point or a complete path including the path ending point. The selection is based on actual circumstances and is not limited in this disclosure. The second decision information represents the likelihood of the corresponding candidate path being selected and can be a decision probability. The conditional generation model can be a conventional prediction model, i.e., a decision model that predicts a path given path planning conditions. The conditional generation model can be trained using an initial conditional generation model based on multiple historical paths and the historical path planning conditions corresponding to each of the multiple historical paths. The initial conditional generation model can be a machine learning model or a deep neural network model. Machine learning models include gradient boosting decision tree (GBDT) models and factorization machines (FM) models. In actual applications, there are multiple ways to determine the target planning path based on multiple reference paths, multiple candidate paths, the first decision information, and the second decision information. The specific method is selected according to the actual situation, and the embodiments of the present disclosure do not impose any limitation on this.In one possible implementation of the present disclosure, a first candidate path and a first reference path belonging to the same map track can be determined. Third decision information corresponding to the first candidate path can be determined based on first decision information corresponding to the first candidate path and second decision information corresponding to the first reference path. A target planned path can be selected from the multiple candidate paths based on the third decision information corresponding to each of the multiple candidate paths. In another possible implementation of the present disclosure, a first candidate path and a first reference path belonging to the same map track can be determined. Fourth decision information corresponding to the first reference path can be determined based on first decision information corresponding to the first candidate path and second decision information corresponding to the first reference path. A target planned path can be selected from the multiple reference paths based on the fourth decision information corresponding to each of the multiple reference paths. Applying the solution of an embodiment of the present disclosure, the path start point, path end point, and path planning conditions are input into a conditional generation model to obtain multiple candidate paths and second decision information corresponding to each of the multiple candidate paths. The target planned path can be determined based on the multiple reference paths, multiple candidate paths, the first decision information, and the second decision information. This converts the path planning problem into a conditional sampling problem, thereby integrating path planning conditions into the path planning process and improving the flexibility and scalability of path planning. In an optional embodiment of the present disclosure, after determining the target planned path based on multiple reference paths, multiple candidate paths, the first decision information, and the second decision information, the following steps may also be included: if the target planned path does not include a path endpoint, updating the path planning conditions based on the target planned path to obtain updated path planning conditions, and returning to the step of inputting the path starting point, path endpoint, and path planning conditions into a condition generation model to obtain multiple candidate paths and the second decision information corresponding to each of the multiple candidate paths, until the target planned path includes a path endpoint, thereby obtaining the target planned path. It should be noted that if the target planned path does not include a path endpoint, it indicates that the path endpoint cannot be reached based on the target planned path, and further path planning is required. In this case, the path planning conditions can be updated based on the target planned path, thereby performing secondary path planning based on the target planned path. In actual applications, when updating the path planning conditions based on the target planned path, the target planned path can be directly added to the path planning conditions as a previously traveled road segment to obtain the updated path planning conditions. Alternatively, path waypoints may be added to the path update planning conditions according to the target planned path. The method for updating the path planning conditions is selected according to the actual situation, and the embodiments of the present disclosure do not impose any limitation on this.Applying the solution of an embodiment of the present disclosure, when the target planned path does not include a path endpoint, the path planning conditions are updated based on the target planned path to obtain updated path planning conditions. The process then returns to the step of inputting the path start point, path endpoint, and path planning conditions into a condition generation model to obtain multiple candidate paths and the second decision information corresponding to each of the multiple candidate paths, until the target planned path includes the path endpoint and the target planned path is obtained. This ensures the integrity of the target planned path and improves path planning flexibility. In an optional embodiment of the present disclosure, determining the target planned path based on multiple reference paths, multiple candidate paths, the first decision information, and the second decision information may include the following steps: selecting a first reference path from the multiple reference paths for a first candidate path, wherein the first candidate path is any one of the multiple candidate paths, and the first candidate path and the first reference path belong to the same map track; determining third decision information corresponding to the first candidate path based on the first decision information corresponding to the first reference path and the second decision information corresponding to the first candidate path; and selecting the target planned path from the multiple candidate paths based on the third decision information corresponding to each of the multiple candidate paths. It should be noted that when determining the third decision information corresponding to the first candidate path based on the first decision information corresponding to the first reference path and the second decision information corresponding to the first candidate path, the first decision information and the second decision information can be multiplied to obtain the third decision information. In practical applications, there are various methods for selecting a target planning path from multiple candidate paths based on the third decision information corresponding to each of the multiple candidate paths. The specific method to be selected depends on actual circumstances and is not limited in this embodiment of the present disclosure. In one possible implementation of the present disclosure, the multiple candidate paths can be ranked based on the third decision information corresponding to each of the multiple candidate paths, and the candidate path with the largest third decision information and a higher ranking can be further determined as the target planning path. In another possible implementation of the present disclosure, a preset decision condition can be obtained, the third decision information corresponding to each of the multiple candidate paths can be compared with the preset decision condition, and the target planning path can be randomly selected from the candidate paths that meet the preset decision condition. By applying the solution of the embodiments of the present disclosure, a first reference path is selected from multiple reference paths for a first candidate path; third decision information corresponding to the first candidate path is determined based on first decision information corresponding to the first reference path and second decision information corresponding to the first candidate path; and a target planning path is selected from the multiple candidate paths based on the third decision information corresponding to each of the multiple candidate paths. This improves the accuracy of the target planning path through the joint prediction of the unconditional generative model and the conditional generative model.In practical applications, there are various methods for selecting the first candidate path from multiple reference paths. The method can be selected based on practical circumstances and is not limited in the present disclosure. In one possible implementation of the present disclosure, the geographic coordinates of all reference paths and the first candidate path can be compared, and the first reference path with the same geographic coordinates as the first candidate path can be selected from the multiple reference paths. In another possible implementation of the present disclosure, the map tracks of all reference paths and the first candidate path can be compared, and the first reference path with the same map track as the first candidate path can be selected from the multiple reference paths. Specifically, selecting the first reference path from the multiple reference paths for the first candidate path can include the following steps: determining, on a preset map road network, the first map track corresponding to the first candidate path and the reference map tracks corresponding to the multiple reference paths; selecting, from the multiple reference map tracks, a target reference map track that is the same as the first map track; and determining the reference path corresponding to the target reference map track as the first reference path. Specifically, the preset map road network can be a map road network in a virtual game scene or a map road network in an actual traffic scene. A map road network refers to a series of interconnected road systems represented on a map, including but not limited to streets, highways, bridges, tunnels, ring roads, auxiliary roads, and branch roads. It can depict the actual geographic location and shape of roads and also reflect the spatial relationships between roads, such as intersections, overpasses, entrances, and exits. A map trajectory refers to a series of consecutive points recorded or depicted on a preset map road network. These points are connected in chronological order to form the path of an object's movement in geographic space. It should be noted that there are multiple methods for determining the first map trajectory corresponding to the first candidate path on the preset map road network, and the specific method selected depends on actual circumstances. This embodiment of the present disclosure does not impose any limitation on this method. In one possible implementation of the present disclosure, the geographic coordinates of all points on the first candidate path can be obtained, the positions of these points can be determined on the preset map road network, and these points can be further connected in chronological order to obtain the first map trajectory. In another possible implementation of the present disclosure, the first candidate path and the preset map road network can be input into a trajectory generation model to output the first map trajectory. In the embodiment of the present disclosure, the method for determining the reference map tracks corresponding to the reference paths is the same as the method for determining the first map track corresponding to the first candidate path, and therefore will not be further described in detail in this embodiment. In actual applications, there are various methods for selecting a target reference map track identical to the first map track from multiple reference map tracks, and the method selected depends on the actual situation. This embodiment of the present disclosure does not impose any limitation on this method.In one possible implementation of the present disclosure, a track among the reference map tracks that overlaps with the first map track can be determined as a target reference map track. In another possible implementation of the present disclosure, the first map track and multiple reference map tracks can be input into a track analysis model to obtain a target reference map track identical to the first map track. Applying the solution of an embodiment of the present disclosure, on a preset map road network, the first map track corresponding to the first candidate path and the reference map tracks corresponding to the multiple reference paths are determined; the target reference map track identical to the first map track is selected from the multiple reference map tracks; and the reference path corresponding to the target reference map track is determined as the first reference path, thereby ensuring the accuracy of the first reference path. In an optional embodiment of the present disclosure, before inputting the path start point and path end point into the path planning model and obtaining multiple reference paths between the path start point and the path end point and first decision information corresponding to each of the multiple reference paths, the following steps may be included: obtaining multiple historical paths; perturbing the multiple historical paths using the initial path planning model to obtain noise data corresponding to each of the multiple historical paths, and restoring the noise data to obtain predicted recovery paths corresponding to each of the multiple historical paths; calculating path prediction losses based on the multiple historical paths and the predicted recovery paths corresponding to each of the multiple historical paths; and adjusting model parameters of the initial path planning model based on the path prediction losses to obtain a trained path planning model. Specifically, the multiple historical paths may be the historical travel paths of any floating vehicle, such as a bus or taxi equipped with an onboard positioning device and traveling on a city's main roads. There are various methods for obtaining multiple historical paths, and the selection is based on actual circumstances. This embodiment of the present disclosure does not impose any limitation thereto. In one possible implementation of the present disclosure, the multiple historical paths may be received from a user. In another possible implementation of the present disclosure, the multiple historical paths may be read from another data acquisition device or database. Furthermore, to ensure the accuracy of model training, after obtaining multiple historical paths, these paths can be filtered, such as by filtering out untraveled paths, filtering out looping paths, and so on. It should be noted that the training process of the path planning model is unsupervised. The method of perturbing the historical paths can vary for different initial path planning models.For example, if the initial path planning model is a variational autoencoder, feature compression can be performed on historical paths to achieve random perturbation and obtain noisy data. If the initial path planning model is a diffusion model, random noise can be added to historical paths to achieve random perturbation and obtain noisy data. If the initial path planning model is a flow model, feature compression or noise addition can be performed on historical paths to obtain noisy data, but the perturbation is performed using a given compression or noise. In practical applications, various functions are used to calculate path prediction loss based on multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths, such as the cross-beam loss function, the L1-norm loss function, the L2-norm loss function, and the KL divergence (Kull-back-Leibler divergence). The specific choice depends on actual circumstances and is not limited in the present embodiment. Furthermore, based on the path prediction loss, the model parameters of the initial path planning model are adjusted until a first preset stopping condition is met, thereby obtaining a trained path planning model. In one possible implementation of the present disclosure, the first preset stopping condition includes a path prediction loss less than or equal to a first preset threshold. After calculating the path prediction loss based on multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths, the path prediction loss is compared with the first preset threshold. Specifically, if the path prediction loss is greater than the first preset threshold, this indicates a significant difference between the historical path and the predicted recovery path, indicating that the initial path planning model has poor path prediction capabilities. In this case, model parameters of the initial path planning model can be adjusted, and the initial path planning model can be trained continuously until the path prediction loss is less than or equal to the first preset threshold, indicating a small difference between the historical path and the predicted recovery path, and the first preset stopping condition is met, thereby obtaining a trained path planning model. In another possible implementation of the present disclosure, in addition to comparing the path prediction loss with the first preset threshold, the number of iterations can also be used to determine whether the current initial path planning model has been trained. Specifically, if the path prediction loss is greater than a first preset threshold, the model parameters of the initial path planning model are adjusted, and the initial path planning model is continuously trained until a first preset number of iterations is reached. The iterations are then stopped to obtain a trained path planning model. The first preset threshold and the first preset number of iterations are selected based on actual circumstances and are not limited in this embodiment. By applying the solution of this embodiment, the model parameters of the initial path planning model are adjusted based on the path prediction loss to obtain a trained path planning model. By continuously adjusting the model parameters of the initial path planning model, the resulting path planning model can be made more accurate.Referring to FIG. 5 , FIG. 5 illustrates a flowchart for training a path planning model in a path planning method provided by an embodiment of the present disclosure. As shown in FIG. 5 , multiple historical paths are obtained and input into an initial path planning model. Feature extraction is performed on the historical paths in the initial path planning model to obtain historical path features corresponding to the multiple historical paths. Perturbation recovery is performed on the historical path features to obtain a predicted recovery path. Path prediction loss is calculated based on the multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths. Based on the path prediction loss, model parameters of the initial path planning model are adjusted to obtain a trained path planning model. In an optional embodiment of the present disclosure, the initial path planning model can be trained based on prior knowledge to ensure the accuracy of the path planning model. Specifically, after perturbing multiple historical paths using the initial path planning model to obtain noise data corresponding to each of the multiple historical paths, restoring the noise data, and obtaining predicted recovery paths corresponding to each of the multiple historical paths, the following steps may be included: obtaining path planning constraints, where the path planning constraints include at least one of a path continuity condition, a path passability condition, and a path compliance condition; calculating a constraint loss based on the path planning constraints and the predicted recovery path; and adjusting model parameters of the initial path planning model based on the path prediction loss to obtain a trained path planning model. This includes: adjusting model parameters of the initial path planning model based on the path prediction loss and the constraint loss to obtain a trained path planning model. Specifically, the path planning constraints may be understood as prior knowledge in the path planning scenario. Path continuity conditions are used to constrain path continuity, path passability conditions are used to constrain path feasibility, and path compliance conditions are used to constrain path compliance, such as whether the path complies with traffic regulations or whether it detours. In practical applications, there are various ways to obtain path planning constraints, and the specific method used depends on the actual situation. This disclosure does not impose any restrictions on this method. In one possible implementation of this disclosure, path planning constraints can be received from a user. In another possible implementation of this disclosure, path planning constraints can be obtained by reading them from other data acquisition devices or databases. It should be noted that the implementation of "adjusting the model parameters of the initial path planning model based on the path prediction loss and the constraint condition loss to obtain a trained path planning model" can be referred to the implementation of "adjusting the model parameters of the initial path planning model based on the path prediction loss to obtain a trained path planning model" described above, and this disclosure will not be further elaborated on in this embodiment.Furthermore, when calculating constraint losses based on the path planning constraints and the predicted recovery path, a series of penalty functions related to the path planning constraints can be defined, and the constraint losses calculated based on the penalty functions. For example, if the path is too close to an obstacle, a distance constraint loss function can be defined that increases as the distance decreases. Applying the solution of the embodiments of the present disclosure, path planning constraints are obtained; constraint losses are calculated based on the path planning constraints and the predicted recovery path; and model parameters of the initial path planning model are adjusted based on the path prediction loss and constraint losses to obtain a trained path planning model. Because constraint losses are considered during the path planning model training process, prior knowledge is incorporated into the training process, improving the accuracy of the path planning model. In an optional embodiment of the present disclosure, before inputting the path start point, path end point, and path planning conditions into the condition generation model to obtain multiple candidate paths and the second decision information corresponding to each of the candidate paths, the following steps may be further included: obtaining multiple historical paths and the historical path planning conditions corresponding to each of the multiple historical paths; inputting the multiple historical paths and the historical path planning conditions corresponding to each of the multiple historical paths into the initial condition generation model to obtain predicted paths corresponding to each of the multiple historical paths; calculating condition generation losses based on the historical paths and the predicted paths; and adjusting model parameters of the initial condition generation model based on the condition generation losses to obtain a trained condition generation model. Specifically, after obtaining the multiple historical paths and the historical path planning conditions corresponding to each of the multiple historical paths, the obtained historical paths and the historical path planning conditions corresponding to each of the multiple historical paths may be filtered to ensure the accuracy of the training of the initial condition generation model. It should be noted that the method of “obtaining multiple historical paths and historical path planning conditions corresponding to the multiple historical paths” can refer to the implementation method of “obtaining multiple historical paths” described above; the method of “inputting multiple historical paths and historical path planning conditions corresponding to the multiple historical paths into the initial condition generation model to obtain predicted paths corresponding to the multiple historical paths” can refer to the implementation method of “inputting the path starting point, path end point and path planning conditions into the condition generation model to obtain multiple candidate paths and the second decision information corresponding to the multiple candidate paths” described above; the method of “adjusting the model parameters of the initial condition generation model according to the condition generation loss to obtain the trained condition generation model” can refer to the implementation method of “adjusting the model parameters of the initial path planning model according to the path prediction loss to obtain the trained path planning model” described above, and the embodiments of the present disclosure will not elaborate on this.In practical applications, various functions are used to calculate the conditional generative loss based on historical and predicted paths, such as the cross-beam loss function, the L1-norm loss function, the L2-norm loss function, and the KL divergence. The selection is based on practical circumstances and is not limited in this embodiment of the present disclosure. By applying the solution of the embodiment of the present disclosure, the model parameters of the initial conditional generative model are adjusted based on the conditional generative loss to obtain a trained conditional generative model. By continuously adjusting the model parameters of the initial conditional generative model, the resulting conditional generative model can be made more accurate. 6 shows a training flowchart of a conditional generation model in a path planning method provided by an embodiment of the present disclosure. As shown in FIG6 , multiple historical paths and historical path planning conditions (such as occurrence times and other personalized planning information) corresponding to the multiple historical paths are obtained; the multiple historical paths and the historical path planning conditions corresponding to the multiple historical paths are input into an initial conditional generation model. In the initial conditional generation model, feature extraction is performed on the historical paths, occurrence times, and other personalized planning information to obtain historical path features, occurrence time features, and other personalized planning features; feature fusion is performed on the historical path features, occurrence time features, and other personalized planning features to obtain fused features; based on the fused features, a predicted path corresponding to the historical path is generated; a conditional generation loss is calculated based on the historical paths and the predicted path; and based on the conditional generation loss, model parameters of the initial conditional generation model are adjusted to obtain a trained conditional generation model. In an optional embodiment of the present disclosure, the initial condition generation model includes a feature extraction unit, a feature fusion unit, and a prediction unit. Inputting multiple historical paths and the historical path planning conditions corresponding to each of the multiple historical paths into the initial condition generation model to obtain predicted paths corresponding to each of the multiple historical paths may include the following steps: For a first historical path, the feature extraction unit extracts features from the first historical path and the first historical path planning condition to obtain a first historical path feature and a first historical path planning condition feature, where the first historical path is any one of the multiple historical paths, and the first historical path planning condition is the historical path planning condition corresponding to the first historical path; the feature fusion unit fuses the first historical path feature and the first historical path planning condition feature to obtain a first fused feature; and the prediction unit generates a first predicted path corresponding to the first historical path based on the first fused feature. It should be noted that the feature fusion unit can fuse the first historical path feature and the first historical path planning condition feature in a variety of ways, which can be selected based on actual circumstances and are not limited in this embodiment of the present disclosure.In one possible implementation of the present disclosure, the first historical path feature and the first historical path planning condition feature can be directly concatenated to obtain a first fused feature. In another possible implementation of the present disclosure, weights can be assigned to the first historical path feature and the first historical path planning condition feature, and then weighted fusion can be performed on the first historical path feature and the first historical path planning condition feature to obtain the first fused feature. Applying the solution of the embodiment of the present disclosure, for the first historical path, a feature extraction unit extracts features from the first historical path and the first historical path planning condition to obtain the first historical path feature and the first historical path planning condition feature. A feature fusion unit then fuses the first historical path feature and the first historical path planning condition feature to obtain the first fused feature. A prediction unit generates a first predicted path corresponding to the first historical path based on the first fused feature. This incorporates the historical path planning condition into the path prediction process, improving the flexibility and scalability of path prediction. The path planning method provided by the present disclosure will be further described below, using its application in smart transportation as an example, with reference to FIG7 . Figure 7 shows a flow chart of a traffic route planning method provided by one embodiment of the present disclosure, specifically including the following steps: Step 702: Obtain traffic route planning requirement information, where the traffic route planning requirement information includes a route start point and a route end point. Step 704: Input the route start point and the route end point into a route planning model to obtain multiple reference paths between the route start point and the route end point and first decision information corresponding to each of the multiple reference paths. The route planning model is trained based on path prediction loss and constraint condition loss. Step 706: Determine a target traffic planning path corresponding to the traffic route planning requirement information based on the multiple reference paths and the first decision information. It should be noted that the implementation of steps 702 to 706 can refer to the implementation of steps 402 to 406 above and will not be further described in this embodiment of the disclosure. By applying the solution of the embodiments of the present disclosure, constraint loss is considered during the path planning model training process, thereby incorporating prior knowledge into the training process and improving the accuracy of the path planning model. Consequently, by automatically generating the target traffic planning path using the path planning model, the path planning process is simplified, ensuring the stability and service performance of traffic route planning. Furthermore, since the traffic route planning process does not require manual design of different map road network weights, the efficiency and scalability of traffic route planning are improved. Referring to Figure 8, a flowchart of a path planning model training method provided by one embodiment of the present disclosure is shown. The path planning model training method is applied to a cloud-side device and specifically includes the following steps: Step 802: Obtain multiple historical routes.Step 804: Input multiple historical paths into the initial path planning model to obtain predicted recovery paths corresponding to the multiple historical paths. Step 806: Calculate path prediction loss based on the multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths. Step 808: Calculate constraint loss based on the path planning constraints and the predicted recovery paths corresponding to the multiple historical paths. Step 810: Adjust the model parameters of the initial path planning model based on the path prediction loss and the constraint loss to obtain a trained path planning model. It should be noted that the implementation of steps 802 to 810 can refer to the training method for the path planning model in the aforementioned path planning method and will not be further described in this embodiment. In actual applications, after obtaining the trained path planning model, the model parameters of the trained path planning model can be sent to the end-side device, allowing the user to locally construct a path planning model based on the model parameters and use the path planning model for path planning. Applying the solution of the embodiments of the present disclosure, the model parameters of the initial path planning model are adjusted based on the path prediction loss and constraint loss to obtain a trained path planning model. By considering the constraint loss during the training process of the path planning model, prior knowledge is incorporated into the training process, thereby improving the accuracy of the path planning model. Referring to Figure 9, a training flowchart of a path planning method provided by one embodiment of the present disclosure is shown. Based on a map road network, the path planning model and the condition generation model are utilized. Based on given path planning conditions, new paths are continuously generated. The path probabilities are measured, and the path with the highest probability that meets the path planning constraints is selected as the final target planned path. Specifically, the path starting point and path ending point (starting and ending points) are input into the path planning model for path planning. Multiple reference paths between the path starting point and path ending point and first decision information corresponding to each of the multiple reference paths are obtained. The path starting point, path ending point, and path planning conditions are input into the condition generation model to obtain multiple candidate paths and second decision information corresponding to each of the multiple candidate paths. The target planned path is determined based on the multiple reference paths, multiple candidate paths, the first decision information, and the second decision information. It's important to note that using only the path planning model and the conditional generation model for path generation eliminates the need to call upon numerous systems and is independent of real-time traffic condition calculations and ETA predictions. This simplifies the overall system chain, significantly improving both stability and service performance. Furthermore, path planning offers significant scalability. Path planning for diverse needs can be implemented by adding different path planning conditions to the conditional generation model, providing significant scalability.Referring to Figure 10, it illustrates a training flowchart for another path planning method according to one embodiment of the present disclosure. Based on a map road network, the method utilizes real-time traffic condition calculations and ETA predictions as more comprehensive prerequisites and conditions, building upon a path planning model and condition generation model to achieve more accurate path planning. Specifically, the path starting point and path ending point (start and end points) are input into the path planning model for path planning, obtaining multiple reference paths between the starting point and the ending point, and first decision information corresponding to each of the multiple reference paths. The path starting point, path ending point, and path planning conditions (including real-time traffic condition calculations and ETA predictions) are input into the condition generation model, obtaining multiple candidate paths and second decision information corresponding to each of the multiple candidate paths. Based on the multiple reference paths, multiple candidate paths, the first decision information, and the second decision information, a target planned path is determined. Corresponding to the aforementioned path planning method embodiment, the present disclosure also provides an embodiment of a path planning device. Figure 11 illustrates a schematic structural diagram of a path planning device according to one embodiment of the present disclosure. As shown in Figure 11 , the apparatus includes: a first acquisition module 1102 configured to acquire path planning requirement information, wherein the path planning requirement information includes a path start point and a path end point; a first input module 1104 configured to input the path start point and the path end point into a path planning model to obtain multiple reference paths between the path start point and the path end point and first decision information corresponding to each of the multiple reference paths, wherein the path planning model is trained based on path prediction loss and constraint condition loss; a first determination module 1106 configured to determine a target planned path corresponding to the path planning requirement information based on the multiple reference paths and the first decision information. Optionally, the path planning requirement information includes path planning conditions; the apparatus also includes: a fourth input module configured to input the path start point, the path end point, and the path planning conditions into a condition generation model to obtain multiple candidate paths and second decision information corresponding to each of the multiple candidate paths; and the first determination module 1106 further configured to determine the target planned path based on the multiple reference paths, the multiple candidate paths, the first decision information, and the second decision information. Optionally, the first determination module 1106 is further configured to, for the first candidate path, select a first reference path from the multiple reference paths, where the first candidate path is any one of the multiple candidate paths, and the first candidate path and the first reference path belong to the same map track; determine third decision information corresponding to the first candidate path based on the first decision information corresponding to the first reference path and the second decision information corresponding to the first candidate path; and select a target planned path from the multiple candidate paths based on the third decision information respectively corresponding to the multiple candidate paths.Optionally, the first determination module 1106 is further configured to determine, on a preset map road network, a first map track corresponding to the first candidate path and reference map tracks corresponding to the multiple reference paths; select, from the multiple reference map tracks, a target reference map track identical to the first map track; and determine the reference path corresponding to the target reference map track as the first reference path. Optionally, the apparatus further includes: an updating module configured to, if the target planned path does not include a path endpoint, update the path planning conditions based on the target planned path to obtain updated path planning conditions, and return to the step of inputting the path start point, path endpoint, and path planning conditions into the condition generation model to obtain multiple candidate paths and second decision information corresponding to the multiple candidate paths, until the target planned path includes a path endpoint, thereby obtaining the target planned path. Optionally, the apparatus further includes: a receiving module configured to receive path adjustment information sent by a user based on the target planned path, and adjust model parameters of the path planning model based on the path adjustment information. Optionally, the apparatus further includes: a path planning model training module configured to obtain multiple historical paths; perturb the multiple historical paths using an initial path planning model to obtain noise data corresponding to each of the multiple historical paths; and restore the noise data to obtain predicted recovery paths corresponding to each of the multiple historical paths; calculate path prediction losses based on the multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths; and adjust model parameters of the initial path planning model based on the path prediction losses to obtain a trained path planning model. Optionally, the apparatus further includes: a third calculation module configured to obtain path planning constraints, wherein the path planning constraints include at least one of a path continuity condition, a path passability condition, and a path compliance condition; and calculate constraint loss based on the path planning constraints and the predicted recovery path. The path planning model training module is further configured to adjust model parameters of the initial path planning model based on the path prediction losses and the constraint loss to obtain a trained path planning model. Optionally, the apparatus further includes: a condition generation model training module configured to obtain a plurality of historical paths and the historical path planning conditions corresponding to the plurality of historical paths; input the plurality of historical paths and the historical path planning conditions corresponding to the plurality of historical paths into an initial condition generation model to obtain predicted paths corresponding to the plurality of historical paths; calculate a condition generation loss based on the historical paths and the predicted paths; and adjust model parameters of the initial condition generation model based on the condition generation loss to obtain a trained condition generation model.Optionally, the initial condition generation model includes a feature extraction unit, a feature fusion unit, and a prediction unit. Furthermore, the condition generation model training module is configured to, for a first historical path, extract features from the first historical path and the first historical path planning condition via the feature extraction unit to obtain a first historical path feature and a first historical path planning condition feature, where the first historical path is any one of a plurality of historical paths, and the first historical path planning condition is the historical path planning condition corresponding to the first historical path; fuse the first historical path feature and the first historical path planning condition feature via the feature fusion unit to obtain a first fused feature; and generate a first predicted path corresponding to the first historical path based on the first fused feature via the prediction unit. The solution of the disclosed embodiments considers constraint loss during path planning model training, thereby incorporating prior knowledge and improving the accuracy of the path planning model. Consequently, by automatically generating a target planned path using the path planning model, the path planning process is simplified, ensuring path planning stability and service performance. Furthermore, since manual design of different map network weights is not required during path planning, the efficiency and scalability of path planning are improved. The above is a schematic diagram of a path planning device according to this embodiment. It should be noted that the technical solution of this path planning device and the technical solution of the path planning method described above share the same concept. For details not described in detail in the technical solution of the path planning device, please refer to the description of the technical solution of the path planning method described above. Corresponding to the above-described traffic path planning method embodiment, the present disclosure also provides a traffic path planning device embodiment. FIG12 shows a schematic diagram of the structure of a traffic path planning device according to one embodiment of the present disclosure. As shown in FIG12 , the device includes: a second acquisition module 1202 configured to acquire traffic path planning requirement information, wherein the traffic path planning requirement information includes a path start point and a path end point; a second input module 1204 configured to input the path start point and the path end point into a path planning model to obtain multiple reference paths between the path start point and the path end point and first decision information corresponding to each of the multiple reference paths, wherein the path planning model is trained based on path prediction loss and constraint condition loss; and a second determination module 1206 configured to determine a target traffic planning path corresponding to the traffic path planning requirement information based on the multiple reference paths and the first decision information.By applying the solution of the embodiments of the present disclosure, constraint loss is considered during the path planning model training process, thereby incorporating prior knowledge into the training process and improving the accuracy of the path planning model. Consequently, by automatically generating the target traffic planning path using the path planning model, the path planning process is simplified, ensuring the stability and service performance of traffic route planning. Furthermore, since the traffic route planning process does not require manual design of different map road network weights, the efficiency and scalability of traffic route planning are improved. The above is a schematic diagram of a traffic route planning device according to this embodiment. It should be noted that the technical solution of this traffic route planning device and the technical solution of the aforementioned traffic route planning method are based on the same concept. For details not described in detail in the technical solution of the traffic route planning device, please refer to the description of the technical solution of the aforementioned traffic route planning method. Corresponding to the above-mentioned path planning model training method embodiment, the present disclosure also provides an embodiment of a path planning model training device. Figure 13 shows a schematic diagram of the structure of a path planning model training device according to one embodiment of the present disclosure. As shown in Figure 13 , the apparatus is applied to a cloud-side device and includes: a third acquisition module 1302 configured to acquire multiple historical paths; a third input module 1304 configured to input the multiple historical paths into an initial path planning model to obtain predicted recovery paths corresponding to the multiple historical paths; a first calculation module 1306 configured to calculate path prediction losses based on the multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths; a second calculation module 1308 configured to calculate constraint losses based on path planning constraints and the predicted recovery paths corresponding to the multiple historical paths; and a first adjustment module 1310 configured to adjust model parameters of the initial path planning model based on the path prediction losses and constraint losses to obtain a trained path planning model. By applying the solution of the embodiments of the present disclosure, the model parameters of the initial path planning model are adjusted based on the path prediction losses and constraint losses to obtain a trained path planning model. By considering the constraint losses during the training of the path planning model, prior knowledge is incorporated into the training process, thereby improving the accuracy of the path planning model. The above is a schematic diagram of a path planning model training device according to this embodiment. It should be noted that the technical solution of this path planning model training device and the technical solution of the path planning model training method described above share the same concept. For details not described in detail in the technical solution of the path planning model training device, please refer to the description of the technical solution of the path planning model training method described above. Figure 14 shows a block diagram of a computing device provided in one embodiment of the present disclosure.Components of the computing device 1400 include, but are not limited to, a memory 1410 and a processor 1420 . oThe processor 1420 is connected to the memory 1410 via a bus 1430. A database 1450 is used to store data. The computing device 1400 also includes an access device 1440 that enables the computing device 1400 to communicate via one or more networks 1460. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 1440 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and the like. In one embodiment of the present disclosure, the aforementioned components of the computing device 1400 and other components not shown in FIG. 14 may also be connected to each other, for example, via a bus. It should be understood that the computing device structure block diagram shown in FIG. 14 is for illustrative purposes only and does not limit the scope of the present disclosure. Those skilled in the art may add or replace other components as needed. Computing device 1400 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC, Persona 1 Computer).Computing device 1400 can also be a mobile or stationary server. Processor 1420 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the aforementioned path planning method, traffic path planning method, or path planning model training method. The above is a schematic diagram of a computing device according to this embodiment. It should be noted that the technical solution of this computing device shares the same concept as the technical solutions of the aforementioned path planning method, traffic path planning method, and path planning model training method. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solutions of the aforementioned path planning method, traffic path planning method, or path planning model training method. An embodiment of the present disclosure also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by the processor, implement the steps of the aforementioned path planning method, traffic path planning method, or path planning model training method. The above is a schematic diagram of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium is based on the same concept as the technical solution of the aforementioned path planning method, traffic path planning method, and path planning model training method. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the aforementioned path planning method, traffic path planning method, or path planning model training method. One embodiment of the present disclosure also provides a computer program. When executed on a computer, the computer executes the steps of the aforementioned path planning method, traffic path planning method, or path planning model training method. The above is an illustrative embodiment of a computer program of this embodiment. It should be noted that the technical solution of this computer program is based on the same concept as the technical solution of the aforementioned path planning method, traffic path planning method, and path planning model training method. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the aforementioned path planning method, traffic path planning method, or path planning model training method. The above describes specific embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the figures do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multi-tasking and parallel processing are also possible or may be advantageous. The computer instructions include computer program code, which may be in source code form, object code form, executable file, or some intermediate form.The computer-readable medium may include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electric carrier signal, a telecommunications signal, and a software distribution medium. It should be noted that the content of the computer-readable medium may be appropriately increased or decreased based on the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunications signals. It should be noted that for ease of description, the aforementioned method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present disclosure are not limited by the order of the actions described, as certain steps may be performed in another order or simultaneously according to the embodiments of the present disclosure. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are preferred embodiments, and the actions and modules involved are not necessarily required for the embodiments of the present disclosure. In the above embodiments, the description of each embodiment has its own emphasis. For portions not described in detail in a particular embodiment, reference should be made to the relevant descriptions of other embodiments. The preferred embodiments disclosed above are merely intended to illustrate the present disclosure. The alternative embodiments do not describe all details in detail, nor do they limit the invention to the specific implementations described. Obviously, many modifications and variations are possible based on the content of the embodiments disclosed. These embodiments are selected and described in detail in this disclosure to better explain the principles and practical applications of the embodiments of the present disclosure, thereby enabling those skilled in the art to better understand and utilize the present disclosure. The present disclosure is limited only by the claims and their full scope and equivalents.

Claims

Claims 1. A path planning method, comprising: Obtain path planning requirement information, where the path planning requirement information includes a path start point and a path end point; input the path start point and the path end point into a path planning model to obtain multiple reference paths between the path start point and the path end point and first decision information corresponding to each of the multiple reference paths, where the path planning model is trained based on a path prediction loss and a constraint condition loss; determine a target planning path corresponding to the path planning requirement information according to the multiple reference paths and the first decision information.

2. The method according to claim 1, wherein the path planning requirement information includes path planning conditions; after inputting the path start point and the path end point into a path planning model to obtain a plurality of reference paths between the path start point and the path end point and first decision information corresponding to each of the plurality of reference paths, the method further includes: Input the path start point, the path end point, and the path planning condition into a condition generation model to obtain multiple candidate paths and second decision information corresponding to each of the multiple candidate paths; The determining, according to the multiple reference paths and the first decision information, a target planning path corresponding to the path planning requirement information includes: determining a target planning path according to the multiple reference paths, the multiple candidate paths, the first decision information, and the second decision information.

3. The method according to claim 2, wherein determining the target planning path according to the multiple reference paths, the multiple candidate paths, the first decision information, and the second decision information comprises: For a first candidate path, screen out a first reference path from the multiple reference paths, where the first candidate path is any one of the multiple candidate paths, and the first candidate path and the first reference path belong to the same map trajectory; determine third decision information corresponding to the first candidate path according to the first decision information corresponding to the first reference path and the second decision information corresponding to the first candidate path; screen out a target planning path from the multiple candidate paths according to the third decision information corresponding to each of the multiple candidate paths.

4. The method according to claim 3, wherein screening out a first reference path from the multiple reference paths for the first candidate path comprises: On a preset map road network, determine a first map trajectory corresponding to the first candidate path and reference map trajectories corresponding to the multiple reference paths respectively; From the multiple reference map trajectories, screen out a target reference map trajectory that is the same as the first map trajectory; Determine the reference path corresponding to the target reference map trajectory as the first reference path.

5. The method according to claim 2, after determining the target planning path according to the plurality of reference paths, the plurality of candidate paths, the first decision information, and the second decision information, further comprising: In the case where the target planning path does not include the path end point, update the path planning condition according to the target planning path to obtain an updated path planning condition, and return to execute the step of inputting the path start point, the path end point, and the path planning condition into a condition generation model to obtain multiple candidate paths and second decision information corresponding to each of the multiple candidate paths, until the target planning path includes the path end point, and obtain a target planning path.

6. The method according to claim 1 or 2, after determining the target planning path corresponding to the path planning requirement information according to the multiple reference paths and the first decision information, further comprising: Receive path adjustment information sent by a user based on the target planning path, and adjust model parameters of the path planning model according to the path adjustment information.

7. For the method according to any one of claims 1 to 6, before inputting the path start point and the path end point into the path planning model to obtain multiple reference paths between the path start point and the path end point and first decision information respectively corresponding to the multiple reference paths, further comprising: Obtain multiple historical paths; Through the initial path planning model, perturb the multiple historical paths to obtain the noise data corresponding to the multiple historical paths respectively, and recover the noise data to obtain the predicted recovery paths corresponding to the multiple historical paths respectively; calculate the path prediction loss according to the multiple historical paths and the predicted recovery paths corresponding to the multiple historical paths respectively; adjust the model parameters of the initial path planning model according to the path prediction loss to obtain the trained path planning model.

8. After the initial path planning model perturbs the multiple historical paths to obtain noise data corresponding to the multiple historical paths respectively, and restores the noise data to obtain predicted restored paths corresponding to the multiple historical paths respectively, according to the method described in claim 7, it further includes: Obtain path planning constraint conditions, where the path planning constraint conditions include at least one of a path coherence condition, a path passage condition, and a path compliance condition; calculate a constraint condition loss according to the path planning constraint conditions and the predicted recovery path; the adjusting the model parameters of the initial path planning model according to the path prediction loss to obtain the trained path planning model includes: adjusting the model parameters of the initial path planning model according to the path prediction loss and the constraint condition loss to obtain the trained path planning model.

9. Before inputting the path start point, the path end point, and the path planning condition input condition into the condition generation model to obtain multiple candidate paths and second decision information respectively corresponding to the multiple candidate paths according to the method described in claim 2, further comprising: Obtain a plurality of historical paths and the historical path planning conditions corresponding to the plurality of historical paths respectively; Input the multiple historical paths and the historical path planning conditions corresponding to the multiple historical paths respectively into the initial condition generation model to obtain the predicted paths corresponding to the multiple historical paths respectively; calculate the condition generation loss according to the historical paths and the predicted paths; adjust the model parameters of the initial condition generation model according to the condition generation loss to obtain the trained condition generation model.

10. The method according to claim 9, wherein the initial condition generation model comprises a feature extraction unit, a feature fusion unit and a prediction unit; Inputting the multiple historical paths and the historical path planning conditions corresponding to the multiple historical paths into an initial condition generation model respectively to obtain the predicted paths corresponding to the multiple historical paths respectively, includes: For the first historical path, through the feature extraction unit, extract features from the first historical path and the first historical path planning condition to obtain the first historical path feature and the first historical path planning condition feature, where the first historical path is any one of the multiple historical paths, and the first historical path planning condition is the historical path planning condition corresponding to the first historical path; through the feature fusion unit, fuse the first historical path feature and the first historical path planning condition feature to obtain a first fusion feature; through the prediction unit, generate a first predicted path corresponding to the first historical path according to the first fusion feature.

11. A traffic path planning method, comprising: Obtain traffic path planning requirement information, where the traffic path planning requirement information includes a path start point and a path end point; input the path start point and the path end point into a path planning model to obtain a plurality of reference paths between the path start point and the path end point and first decision information corresponding to each of the plurality of reference paths, where the path planning model is trained based on a path prediction loss and a constraint condition loss; determine a target traffic planning path corresponding to the traffic path planning requirement information according to the plurality of reference paths and the first decision information.

12. A method for training a path planning model, applied to a cloud-side device, comprising: Obtain a plurality of historical paths; Input the plurality of historical paths into an initial path planning model to obtain predicted recovery paths corresponding to each of the plurality of historical paths; Calculate a path prediction loss according to the plurality of historical paths and the predicted recovery paths corresponding to each of the plurality of historical paths; calculate a constraint condition loss according to path planning constraint conditions and the predicted recovery paths corresponding to each of the plurality of historical paths; adjust model parameters of the initial path planning model according to the path prediction loss and the constraint condition loss to obtain a trained path planning model.

13. A computing device, comprising: A memory and a processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 10 or claim 11 or claim 12 are implemented.

14. A computer-readable storage medium storing computer-executable instructions, and when the computer-executable instructions are executed by a processor, the steps of the method according to any one of claims 1 to 10 or claim 11 or claim 12 are implemented.

15. A computer program, wherein When the computer program is executed on a computer, the computer is made to execute the steps of the method according to any one of claims 1 to 10 or claim 11 or claim 12.

Images