Autonomous-driving training sample generation method, and device, computer program product and storage medium

Abstract

Provided in the embodiments of the present disclosure are an autonomous-driving training sample generation method, and a device, a computer program product and a storage medium. Provided is a description dimension for describing a driving environment, and a required target driving environment may be described by means of constructing description data in the description dimension; and the constructed description data can then be input into a preset generation model, such that the generation capability of the generation model is used to generate driving sensing data corresponding to the target driving environment, and the driving sensing data is used as an autonomous-driving training sample. Thus, by means of continuously changing description data in each description dimension, the description data can be combined to describe rich driving environments, especially various uncommon driving environments in daily driving, and driving sensing data can be automatically generated for the described various driving environments, such that richer training samples in more comprehensive scenarios can be provided for autonomous-driving training, thereby improving the decision-making performance of an autonomous-driving model.

Description

[0001]This disclosure claims priority to Chinese patent application number 202410367138.9, filed with the China Patent Office on March 27, 2024, entitled "A Method, Apparatus, Computer Program Product, and Storage Medium for Generating Autonomous Driving Training Samples," the entire contents of which are incorporated herein by reference. TECHNICAL FIELD This disclosure relates to the field of autonomous driving technology, and more particularly to a method, apparatus, computer program product, and storage medium for generating autonomous driving training samples. Background: Autonomous driving technology relies on the collaborative efforts of artificial intelligence, visual computing, radar, and global positioning systems to automatically and safely control a vehicle without human control. Autonomous driving relies on autonomous driving models to make control decisions. Therefore, the decision-making performance of autonomous driving models is crucial to autonomous driving technology. Currently, training samples used in autonomous driving model training primarily come from driving perception data collected from real vehicles in real driving environments. This source is prohibitively expensive, time-consuming, and provides a very limited number of training samples. It can also come from driving perception data generated by driving simulations within autonomous driving simulation engines. This requires extensive labor, such as artwork and animation, to build the driving environment before the simulation can be performed, resulting in a similarly high cost. Consequently, due to this high cost, autonomous driving training samples are insufficient, leaving the decision-making performance of autonomous driving models in need of improvement. SUMMARY Various aspects of the present disclosure provide a method, device, computer program product, and storage medium for generating autonomous driving training samples, thereby providing a richer set of autonomous driving training samples. An embodiment of the present disclosure provides a method for generating autonomous driving training samples, comprising: constructing description data for a desired target driving environment within a preset description dimension; inputting the constructed description data into a preset generation model, wherein the generation model contains knowledge for generating driving perception data for the described driving environment based on the description data; and utilizing the generation model to generate driving perception data for the target driving environment as autonomous driving training samples. Furthermore, constructing description data under a preset description dimension includes: for any description dimension, selecting existing reference data under the description dimension and modifying it to generate description data under the description dimension for describing the target driving environment; or, creating new description data under the description dimension to generate description data under the description dimension for describing the target driving environment.Furthermore, the method further includes: pre-extracting description data for the description dimension from real driving perception data to serve as reference data for the description dimension; and / or using description data constructed for other driving environments and corresponding to the description dimension as reference data for the description dimension. Furthermore, the description data is structured data, and the description dimensions include an environmental prompt dimension, a vehicle trajectory dimension, a map dimension, and / or an internal and external parameter dimension of a perception device; and the modalities used in the constructed description data include text, trajectory, and / or map. Furthermore, constructing description data for the preset description dimensions includes: constructing environmental prompt words for describing the target driving environment in the environmental prompt dimension; and / or constructing trajectory data for each vehicle within the target driving environment in the vehicle trajectory dimension; and / or constructing road attribute data within the target driving environment in the map dimension; and / or setting internal and external parameters of the perception device for the host vehicle within the target driving environment in the internal and external parameter dimension of the perception device. Furthermore, the trajectory data includes vehicle spacing, obstacle attributes, obstacle location, and / or congestion data; and the road attribute data includes road direction, intersection shape, road width, road curvature, road flatness, and / or road slope. Furthermore, constructing trajectory data for each vehicle within the target driving environment within the vehicle trajectory dimension includes: adjusting the vehicle spacing between the host vehicle and adjacent vehicles within the reference data selected within the vehicle trajectory dimension; and / or adding obstacles around the host vehicle and setting obstacle attributes; and / or adjusting the congestion level. Furthermore, constructing road attribute data for the target driving environment in the map dimension includes: adjusting the width, curvature, flatness, and / or slope of the roads involved in the trajectory data in the reference data selected in the map dimension; and / or modifying the intersection shape and triggering the simultaneous modification of the trajectory route in the trajectory data corresponding to the target driving environment to adapt to the modified intersection shape; and / or modifying the road direction and triggering the simultaneous modification of the trajectory route in the trajectory data corresponding to the target driving environment to adapt to the modified road direction. Furthermore, the method further includes: modifying the descriptive data in some description dimensions created for the target driving environment, while keeping the descriptive data in the remaining description dimensions unchanged, to generate a set of descriptive data capable of describing the new driving environment; and inputting the generated descriptive data into the generative model to output driving perception data corresponding to the new driving environment using the generative model.Furthermore, the generation module employs a constrained diffusion model, and inputting the constructed description data into a preset generation model includes: vectorizing the description data under different description dimensions to generate representation vectors; and inputting the representation vectors into the constrained diffusion model to serve as constraints for the constrained diffusion model. Embodiments of the present disclosure also provide a computing device comprising a memory, a processor, and a communication component; the memory is configured to store one or more computer instructions; and the processor is coupled to the memory and the communication component and configured to execute the one or more computer instructions to perform the aforementioned method for generating autonomous driving training samples. Embodiments of the present disclosure also provide a computer-readable storage medium storing a computer program. When the computer program is executed by one or more processors, the computer program causes the one or more processors to perform the aforementioned method for generating autonomous driving training samples. Embodiments of the present disclosure also provide a computer program product, comprising the computer program. When the computer program is executed by one or more processors, the computer program causes the one or more processors to perform the aforementioned method for generating autonomous driving training samples. In the embodiments of the present disclosure, description dimensions for describing driving environments are proposed. Based on these dimensions, description data can be constructed for a desired target driving environment within the description dimensions to describe the target driving environment. This constructed description data can then be input into a pre-set generative model, leveraging the generative model's generative capabilities to generate driving perception data corresponding to the target driving environment, which serves as autonomous driving training samples. Consequently, in this embodiment, without transforming the description data within each description dimension, a rich variety of driving environments can be combined and described. In particular, various driving scenarios not commonly encountered in daily driving can be described, and driving perception data can be automatically generated for each described driving scenario. Compared to traditional approaches that require the construction of real-world test scenarios and simulated driving environments, this approach effectively reduces the cost of acquiring driving perception data and provides richer, more comprehensive training samples for autonomous driving training, thereby improving the decision-making performance of autonomous driving models. BRIEF DESCRIPTION OF THE DRAWINGS The drawings described herein are provided to provide a further understanding of the present disclosure and constitute a part of this disclosure. The illustrative embodiments of this disclosure and their descriptions are provided for illustrative purposes only and are not intended to unduly limit this disclosure.In the accompanying drawings: Figure 1 is a flow chart of a method for generating autonomous driving training samples provided by an exemplary embodiment of the present disclosure; Figure 2 is a logic diagram of a method for generating autonomous driving training samples provided by an exemplary embodiment of the present disclosure; Figure 3 is a logic diagram of an optional implementation scheme of a method for generating autonomous driving training samples provided by an exemplary embodiment of the present disclosure; and Figure 4 is a schematic diagram of the structure of a computing device provided by another exemplary embodiment of the present disclosure. To further clarify the objectives, technical solutions, and advantages of the present disclosure, the technical solutions of the present disclosure will be described clearly and completely below in conjunction with the specific embodiments of the present disclosure and the corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present disclosure, and are not exhaustive. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present disclosure without inventive effort are within the scope of protection of the present disclosure. Before describing in detail the technical solutions provided by the various embodiments of the present disclosure, several technical concepts involved in the present disclosure are briefly explained below. Autonomous driving technology relies on the collaborative efforts of artificial intelligence, visual computing, radar, and global positioning systems to automatically and safely control a vehicle without human control. An autonomous driving model is used to provide control decisions in autonomous driving technology. The model's input is driving perception data, and its output is the resulting control instructions. Based on the resulting control instructions, the vehicle can be automatically controlled to achieve autonomous driving. These control instructions may include braking, reversing, and steering commands. Driving perception data originates from various sensing devices installed on the vehicle, and can also refer to the resulting data obtained by processing the data collected by these sensing devices. The driving environment can be understood as the vehicle's surroundings. For a vehicle, the driving environment is complex, diverse, and constantly changing. As described in the background, currently, the training samples used in autonomous driving model training primarily come from driving perception data collected from real vehicles in real driving environments, as well as driving perception data generated through driving simulations within autonomous driving simulation engines. During their research, the inventors discovered that the driving environments encountered by real vehicles during daily driving are limited, typically safe and routine. Therefore, the driving scenarios covered by driving perception data collected from real vehicles during daily driving are also relatively limited.Driving scenarios uncommon in everyday driving require manual construction, and dangerous driving environments can even result in vehicle damage. This results in prohibitively high costs for acquiring driving perception data in these environments using real vehicles. Furthermore, autonomous driving simulation engines that wish to simulate driving in these environments require significant labor costs, such as through art and animation, to create these environments. Furthermore, the environmental details involved in these environments are highly varied and complex, making construction even more challenging. Consequently, the cost of providing driving perception data in these environments using autonomous driving simulation engines is also prohibitively high. Consequently, due to this high cost, the training samples currently available for autonomous driving models are insufficient, or the coverage of driving scenarios is incomplete. This leaves the decision-making performance of autonomous driving models in need of improvement. To address this, embodiments of the present disclosure propose a method for generating autonomous driving training samples. This method can provide autonomous driving models with a richer set of training samples to supplement their coverage of driving scenarios, thereby improving their decision-making performance. The following, in conjunction with the accompanying drawings, details the technical solutions provided by various embodiments of the present disclosure. Figure 1 is a flow chart illustrating a method for generating autonomous driving training samples, provided in accordance with an exemplary embodiment of the present disclosure. This method can be executed by a data processing device, which can be implemented as software, hardware, or a combination of software and hardware. The data processing device can be integrated into a computing device. Referring to Figure 1 , the method may include: Step 100: Constructing description data for a desired target driving environment within a preset description dimension; Step 101: Inputting the constructed description data into a preset generative model, wherein the generative model includes knowledge for generating driving perception data for the described driving environment based on the description data; Step 102: Using the generative model, generating driving perception data for the target driving environment as autonomous driving training samples. The autonomous driving training sample generation method provided in this embodiment can be used to generate a richer set of training samples for autonomous driving models. Preferably, the computing device used to execute this autonomous driving training sample generation method in this embodiment can be a standalone cloud server or a cloud service cluster, etc., to improve the efficiency of generating autonomous driving training samples in this embodiment. Alternatively, a personal computer, etc., can be employed. This embodiment does not limit the implementation form of the computing device. Figure 2 is a logical diagram of a method for generating autonomous driving training samples, provided by an exemplary embodiment of the present disclosure. Referring to Figure 2 , this embodiment introduces preset description dimensions. Different description dimensions can be understood as describing different environmental factors in the driving environment.Therefore, in this embodiment, various desired driving environments can be flexibly combined using environmental factors described under different description dimensions. Based on this, referring to FIG1 , in step 100, descriptive data can be constructed for the desired target driving environment under preset description dimensions. Descriptive data refers to data used to describe the corresponding environmental factors within the description dimensions. It should be understood that in this embodiment, the target driving environment in step 100 can be any desired driving environment. As mentioned above, in practical applications, the target driving environment can particularly include various driving environments that are uncommon in daily driving. Thus, for such uncommon driving environments, there is no need to construct a real or simulated driving environment. Instead, the descriptive data constructed under the description dimensions in step 100 in this embodiment can be used to characterize various desired uncommon driving environments. Of course, it is worth noting that in this embodiment, the target driving environment can also be a relatively common driving environment in daily driving. Based on the technical concepts provided by this embodiment, autonomous driving training samples can also be provided to supplement these comparative driving environments. This embodiment does not limit the target driving environment to whether it is common or not. In other words, the target driving environment in step 100 can be any driving environment for which autonomous driving training samples are desired to be supplemented. In this embodiment, the descriptive data constructed for the target driving environment in step 100 can include a variety of data types, including but not limited to text, trajectory, or map data types. In practical applications, appropriate data types can be selected for different description dimensions to ensure that the descriptive data in each description dimension more accurately and conveniently describes the driving environment. In this embodiment, the preset description dimensions may include but are not limited to the environmental prompt dimension, the vehicle trajectory dimension, the map dimension, and the internal and external parameter dimension of the perception device. It should be understood that these description dimensions are merely exemplary and this embodiment is not limited thereto. In practical applications, more description dimensions may be selected to more fully and comprehensively describe the driving environment. The environmental prompt dimension can be used to describe the overall attributes of the driving environment. The data type selected for this environmental prompt dimension can be text as described above. The descriptive data in the environmental description dimension can be environmental prompt words. Several exemplary environmental prompt words may include: sunny day, rainy day, intersection, front camera view, rear camera view, underground garage, mountain road, or school. Further examples are not provided here. Furthermore, in actual applications, the descriptive data under the environmental prompt dimension may be manually input or automatically generated through methods such as automatic enumeration, which is not limited in this embodiment.The vehicle trajectory dimension can be used to describe vehicle trajectories in the driving environment. The data type selected for this dimension can be the aforementioned trajectory, or a trajectory file. Trajectory file formats include, but are not limited to, GPX, FIT, TCX, CSV, KML, or any custom format, and are not limited here. The descriptive data in the vehicle trajectory dimension can be trajectory data. Trajectory data can include data describing the trajectories of the host vehicle and other vehicles in the driving environment. Trajectory data can include, but are not limited to, vehicle spacing, obstacle attributes, obstacle locations, and congestion data, and further examples are not provided here. The host vehicle mentioned in the context of this embodiment can be understood as the vehicle in the driving environment intended for autonomous driving, while other vehicles in the driving environment can be understood as vehicles that may enter the host vehicle's field of view during driving. The map dimension can be used to describe the map in the driving environment. The data type selected for this dimension can be the aforementioned map, or a map file. Map file formats include, but are not limited to, EPS, KML, SHP, or CAD, and are not limited here. The descriptive data in the map dimension can be map data. Map data may include, but is not limited to, road direction, intersection shape, road width, road curvature, road flatness, or road slope, among other information. Further examples are not provided here. The internal and external parameters dimension of the perception device is used to describe the internal and external parameters of the perception device on the host vehicle in the driving environment. The data type selected for the internal and external parameters of the perception device can be text or a data table, without limitation. The descriptive data in the internal and external parameters dimension of the perception device is the internal and external parameters of the perception device. The internal parameters of a perception device are typically factory-set parameters, such as camera resolution and focal length, while the external parameters of a perception device are adjustable parameters, such as camera mounting position and camera pitch angle. Here, the internal and external parameters dimension of the perception device primarily constructs the internal and external parameters corresponding to the perception device installed on the host vehicle in the driving environment. Furthermore, in practical applications, the descriptive data in the internal and external parameters dimension of the perception device can be selected based on vehicle model, selecting the internal and external parameters of the perception device published for existing vehicle models. In other words, the internal and external parameters of the sensing devices of the desired vehicle model can be found through publicly available data to describe the host vehicle in the target driving environment. It should be understood that the above description dimensions are merely illustrative and this embodiment is not limited thereto. Other description dimensions that can describe the driving environment are also applicable to this embodiment.Furthermore, in this embodiment, the descriptive data under each description dimension can be implemented as structured data, which facilitates data construction and processing. Continuing with Figure 1, in step 101, the constructed descriptive data can be input into a pre-set generative model. A generative model is a machine learning model that, by learning the underlying representation and distribution of real data, can generate new content that is similar to the real data. Based on this capability of the generative model, the pre-set generative model in this embodiment incorporates knowledge of generating driving perception data for the described driving environment based on the descriptive data. This knowledge can be acquired by the generative model through model training. In an exemplary training scheme, descriptive data can be extracted from real driving perception data under the aforementioned description dimensions. The real driving perception data and the extracted descriptive data are then fed into the generative model for supervised training of the generative model. The real driving perception data serves as supervision during the training process. In this way, the generative model can learn what type of driving perception data should be generated when receiving a certain combination of descriptive data. Figure 3 is a logical diagram of an optional implementation of a method for generating autonomous driving training samples, provided in an exemplary embodiment of the present disclosure. Referring to Figure 3, the generation module in this embodiment preferably employs a controllable diffusion model (DM). Diffusion models (DMs) are probability-based generation models that generate new data by gradually adding noise to data and then gradually removing it. A controllable diffusion model can be understood as a diffusion model that uses input as a constraint. Based on this, in step 101, the description data in different description dimensions can be vectorized separately to generate representation vectors. Specifically, a representation vector can be generated for each description dimension. These representation vectors are then input into the controllable diffusion model as constraints for the model. Different vector representation rules can be employed for different description dimensions. In practical applications, pre-trained models can be used to perform vectorization operations, and different types of pre-trained models can be employed for different description dimensions. For example, the aforementioned scene prompt dimension uses text as the data type. Therefore, a pre-trained model suitable for text, such as clip, can be employed to vectorize the description data in the scene prompt dimension. For another example, the data type used in the aforementioned vehicle trajectory dimension is trajectory text. Therefore, a pre-trained model such as QcNET can be used to vectorize the description data in the vehicle trajectory dimension.It should be understood that this embodiment does not limit the type of pre-trained model used for different description dimensions; it only needs to be applicable to the data type within the corresponding description dimension. Referring to Figure 3 , after vectorization, in step 101, representation vectors corresponding to each description dimension of the target driving environment are generated. These representation vectors can be input into the constrained diffusion model shown in Figure 3 as constraints for the constrained diffusion model. Input can be performed according to the conventional input requirements of the constrained diffusion model. In one exemplary input implementation, the representation vectors corresponding to the aforementioned vehicle trajectory dimension, map dimension, and internal and external parameter dimension of the perception device can be input into the constrained diffusion model's encoding module. Referring to Figure 3 , within the constrained diffusion model, after encoding by the encoding module, the encoding result can be passed to the constrained diffusion model's decoding module, ultimately outputting the driving perception data required in this embodiment. Referring to Figure 3 , the representation vector corresponding to the aforementioned scene cue dimension does not need to be input into the constrained diffusion model's encoding module. Instead, it can be directly combined with the encoding result output by the encoding module and then passed to the constrained diffusion model's decoding module as an encoding result. It should be understood that the above-described input implementation scheme is merely exemplary and not limited to this embodiment, and further explanation is omitted here. Continuing with Figures 1 and 2, in step 102, the generative model can be used to generate driving perception data for the target driving environment as autonomous driving training samples. As mentioned above, the generative model in this embodiment has the ability to generate driving perception data based on descriptive data. Therefore, in step 101, after inputting descriptive data according to preset descriptive dimensions into the generative model, the generative model is triggered to output driving perception data corresponding to the target driving environment. The driving perception data output in step 102 in this embodiment can be in a format such as images or videos, which is not limited here. In practical applications, the driving perception data output in step 102 of this embodiment can be set to a training sample format supported by the autonomous driving model. In this way, the training samples output in step 102 can be directly provided to the autonomous driving model for training. In summary, this embodiment proposes description dimensions for describing the driving environment. Based on these dimensions, for a desired target driving environment, description data can be constructed within the description dimensions to describe the target driving environment. The constructed description data can then be input into a preset generative model to leverage the generative model's generative capabilities to generate driving perception data corresponding to the target driving environment, serving as autonomous driving training samples.Accordingly, in this embodiment, by continuously transforming the descriptive data under various description dimensions, a rich set of driving environments can be combined to describe them. In particular, various driving scenarios not commonly encountered in daily driving can be described, and driving perception data can be automatically generated for each described driving scenario. Compared to traditional solutions that require the construction of real test scenarios and simulated driving environments, this approach can effectively reduce the cost of acquiring driving perception data and provide richer, more comprehensive training samples for autonomous driving training, thereby improving the decision-making performance of the autonomous driving model. In the above and following embodiments, descriptive data can be constructed using various implementations. In one optional implementation, existing reference data for any description dimension can be selected and modified to generate descriptive data for the target driving environment under that dimension. In this optional implementation, descriptive data for the target driving environment can be generated by flexibly modifying the reference data. In this way, descriptive data for the target driving environment can be constructed through simple data modification operations, eliminating the need to construct a real or simulated driving environment as in traditional solutions. Clearly, the cost of this simple data modification operation is far lower than that of traditional solutions, and it very efficiently achieves the desired target driving environment description. In this optional implementation, the reference data for any description dimension can come from a variety of sources. One exemplary source can be pre-extracting description data for any description dimension from real driving perception data to serve as reference data for that description dimension. Pre-extracting description data from real driving perception data is not limited to this implementation. In practice, it can be extracted using a machine learning module or manually compiled through observation of real driving perception data. Further implementation examples are not provided here; it is sufficient to ensure that the extracted description data is consistent with the format of the description data specified for the corresponding description dimension in this embodiment. Furthermore, when multiple sets of real driving perception data are used, multiple sets of pre-extracted description data for any description dimension may be available. Therefore, when constructing description data for a target driving environment, multiple sets of reference data for that description dimension can be selected. In practical applications, the reference data set that most closely matches the target driving environment or contains the most data items required for the target driving environment can be selected to reduce the number of modification operations, although this is not a limitation. Another exemplary source of data could be using description data constructed for other driving environments and corresponding to that description dimension as reference data for that description dimension.During their research, the inventors discovered that many driving environments are not significantly different from one another, nor are the descriptive data within the same descriptive dimension. Therefore, a driving environment for which descriptive data has already been constructed can serve as a reference environment for the target driving environment. That is, when constructing descriptive data for a target driving environment, the descriptive data from other driving environments can be borrowed and, after simple modifications, generated to describe the target driving environment. It is worth noting that the various sources of reference data provided above are merely illustrative, and this embodiment is not limited thereto, and further examples are not provided here. Based on the abundant reference data, in this optional implementation, appropriate reference data can be selected within any descriptive dimension for the target driving environment, further reducing the number of required modification operations and enabling more efficient and accurate construction of descriptive data for the target driving environment. It should be understood that, while multiple descriptive dimensions are pre-defined in this embodiment, this does not necessarily require modification operations on the reference data for each dimension to generate the required descriptive data for the target driving environment. This is because the descriptive data for some descriptive dimensions can directly use the corresponding reference data without modification. For example, if the target driving environment and a previously described driving environment share the same environmental cue words, the environmental cue words in that driving environment can be directly used as the descriptive data for the target driving environment within the environmental cue dimension. Based on the aforementioned technical concept of constructing the required descriptive data for a driving environment by simply modifying reference data, this embodiment proposes a practical application scheme: The descriptive data for some of the description dimensions created for the target driving environment are modified, while the descriptive data for the remaining description dimensions remain unchanged, to generate a set of descriptive data capable of describing the new driving environment. The generated descriptive data is then input into a generative model, which then uses the generative model to output driving perception data corresponding to the new driving environment. It is understood that this practical application scheme proposes that a new driving environment can be described by partially modifying the descriptive data corresponding to an already described driving environment. This allows for the generation of richer and more extensive driving environments based on the already described driving environment, thereby more quickly, cost-effectively, and simply increasing the richness of autonomous driving training samples. In another optional implementation, new description data may be created within any description dimension to generate description data for describing the target driving environment within that description dimension. The creation operation in this optional implementation may be understood as a creation operation from scratch.As mentioned above, the description data in this embodiment can be structured data. Therefore, the required description data can be created by manually compiling the required structured data items according to the description dimension. It is worth noting that the two aforementioned implementation methods for constructing description data can be used interchangeably for a specific driving environment. The appropriate implementation method can be selected for different description dimensions. In summary, in this embodiment, multiple implementation methods can be used to construct description data for the target driving environment to be described, transforming the problem of constructing the driving environment into the problem of constructing description data. Therefore, through simple data operations such as modification or creation, the desired driving environment can be accurately described, with low cost, high efficiency, and no reliance on professional knowledge. Thus, in this embodiment, simple data operations can flexibly describe a variety of driving environments. Theoretically, all possible driving environments can be covered, especially those less common in daily driving, as mentioned above. This effectively improves coverage of driving environments and, in turn, enhances the richness of autonomous driving training samples. In the above and following embodiments, the content of the description data required to be constructed varies depending on the description dimension. The following describes, using several exemplary description dimensions as examples, a detailed description data construction scheme for each dimension. This embodiment proposes: In the environmental prompt dimension, scenario prompt words can be constructed to describe the target driving environment; in the vehicle trajectory dimension, trajectory data for each vehicle within the target driving environment can be constructed; in the map dimension, road attribute data within the target driving environment can be constructed; and in the perception device intrinsic and extrinsic parameter dimension, intrinsic and extrinsic parameters of the perception device can be set for the host vehicle within the target driving environment. As mentioned above, trajectory data here may include, but is not limited to, vehicle spacing, obstacle attributes, obstacle location, or congestion data. Road attribute data may include, but is not limited to, road direction, intersection shape, road width, road curvature, road flatness, or road slope. Intrinsic parameters of the perception device may include, but are not limited to, camera resolution and focal length, while extrinsic parameters of the perception device may include, but are not limited to, camera mounting position and camera pitch angle. Based on the various data items preset for describing data in different description dimensions, in this embodiment, values ​​under various preset data items can be constructed for the target driving environment in different description dimensions by modifying or creating the aforementioned embodiments, thereby generating the required description data.Taking the construction of descriptive data using a modification method (i.e., modifying selected reference data) as an example: within the vehicle trajectory dimension, the distance between the host vehicle and adjacent vehicles can be adjusted; and / or obstacles can be added around the host vehicle and their attributes set; and / or the congestion level can be adjusted, etc. Specifically, for each data item set within the vehicle trajectory dimension, trajectory data for the target driving environment is constructed within the vehicle trajectory dimension by modifying the corresponding data item within the reference data. Within the map dimension, the width, curvature, flatness, and / or slope of the road involved in the trajectory data can be adjusted; and / or the intersection shape can be modified, triggering the simultaneous modification of the trajectory route in the trajectory data corresponding to the target driving environment to adapt to the modified intersection shape; and / or the road direction can be modified, triggering the simultaneous modification of the trajectory route in the trajectory data corresponding to the target driving environment to adapt to the modified road direction, etc. Specifically, for each data item set in the map dimension, trajectory data is constructed for the target driving environment in the map dimension by modifying the corresponding data item in the reference data. Furthermore, it should be understood that in practical applications, for the same driving environment, the aforementioned vehicle trajectory dimension and map dimension must be compatible. That is, the vehicle trajectory described in the vehicle trajectory dimension must be able to adapt to the road described in the map dimension. Only in this way can road attributes and other information in the map dimension be accurately incorporated into the trajectory data, thereby accurately describing the desired driving environment. In summary, this embodiment provides description data construction schemes for several exemplary description dimensions. The various data items set in the description data are combined to accurately describe the desired driving environment. Therefore, by designing comprehensive and reasonable data items, it is possible to ensure that the constructed description data accurately describes the desired driving environment. It should be noted that some of the processes described in the above embodiments and accompanying drawings include multiple operations that appear in a specific order. However, it should be understood that these operations may be executed in a different order than the order in which they appear herein or in parallel. Operation numbers such as 101 and 102 are merely used to distinguish between different operations and do not represent any execution order. FIG4 is a schematic diagram of the structure of a computing device provided in another exemplary embodiment of the present disclosure. As shown in FIG4 , the computing device includes a memory 40, a processor 41, and a communication component 42.The processor 41 is coupled to the memory 40 and the communication component 42 and is configured to execute a computer program in the memory 40 to: construct description data for a desired target driving environment within preset description dimensions; input the constructed description data into a preset generative model, wherein the generative model includes knowledge for generating driving perception data for the described driving environment based on the description data; and utilize the generative model to generate driving perception data for the target driving environment as autonomous driving training samples. In an optional embodiment, when constructing description data within the preset description dimensions, the processor 41 may specifically be configured to: for any description dimension, select and modify existing reference data within the description dimension to generate description data for describing the target driving environment within the description dimension; or create new description data within the description dimension to generate description data for describing the target driving environment within the description dimension. In an optional embodiment, the processor 41 may further be configured to: pre-extract description data for the description dimension from real driving perception data to serve as reference data for the description dimension; and / or use description data constructed for other driving environments and corresponding to the description dimension as reference data for the description dimension. In an optional embodiment, the description data is structured data, and the description dimensions include an environmental prompt dimension, a vehicle trajectory dimension, a map dimension, and / or an internal and external parameter dimension of a perception device; and the modalities used in the constructed description data include text, trajectory, and / or map. In an optional embodiment, when constructing description data within a preset description dimension, the processor 41 may be specifically configured to: construct environmental prompt words describing the target driving environment within the environmental prompt dimension; and / or construct trajectory data for each vehicle within the target driving environment within the vehicle trajectory dimension; and / or construct road attribute data within the target driving environment within the map dimension; and / or set internal and external parameters of the perception device for the host vehicle within the target driving environment within the perception device internal and external parameters dimension. In an optional embodiment, the trajectory data includes vehicle spacing, obstacle attributes, obstacle location, and / or congestion data; and the road attribute data includes road direction, intersection shape, road width, road curvature, road flatness, and / or road slope.In an optional embodiment, when constructing trajectory data for each vehicle within the target driving environment in the vehicle trajectory dimension, the processor 41 may be specifically configured to: adjust the distance between the host vehicle and adjacent vehicles in the reference data selected in the vehicle trajectory dimension; and / or add obstacles around the host vehicle and set obstacle attributes; and / or adjust the congestion level. In an optional embodiment, when constructing road attribute data within the target driving environment in the map dimension, the processor 41 may be specifically configured to: adjust the width, curvature, flatness, and / or slope of the road involved in the trajectory data in the reference data selected in the map dimension; and / or modify the intersection shape and trigger the simultaneous modification of the trajectory route in the trajectory data corresponding to the target driving environment to adapt to the modified intersection shape; and / or modify the road direction and trigger the simultaneous modification of the trajectory route in the trajectory data corresponding to the target driving environment to adapt to the modified road direction. In an optional embodiment, the processor 41 may further be configured to: modify the description data for some description dimensions created for the target driving environment, while keeping the description data for the remaining description dimensions unchanged, to generate a set of description data capable of describing the new driving environment; and input the generated description data into the generation model to utilize the generation model to output driving perception data corresponding to the new driving environment. In an optional embodiment, the generation module utilizes a constrained diffusion model. When inputting the constructed description data into the preset generation model, the processor 41 may be specifically configured to: vectorize the description data for different description dimensions to generate representation vectors; and input the representation vectors into the constrained diffusion model to serve as constraints for the constrained diffusion model. Furthermore, as shown in FIG4 , the computing device also includes other components, such as a power supply component 43. FIG4 schematically illustrates only some components and does not imply that the computing device comprises only those shown in FIG4 . It is worth noting that the technical details of the computing device embodiments described above can be found in the relevant descriptions of the aforementioned method embodiments. To save space, they will not be repeated here, but this should not diminish the scope of protection of the present disclosure. Accordingly, embodiments of the present disclosure also provide a computer-readable storage medium storing a computer program. When the computer program is executed, it can implement the steps of the aforementioned method embodiments. Accordingly, embodiments of the present disclosure also provide a computer program product. When the computer program is executed, it can implement the steps of the aforementioned method embodiments.The memory in FIG. 4 is used to store computer programs and can be configured to store various other data to support operations on the computing platform. Examples of such data include instructions for any application or method operating on the computing platform, contact data, phone book data, messages, images, videos, and the like. The memory can be implemented using any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disks, or optical disks. The communication component in FIG. 4 is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G / LTE, 5G, and other mobile communication networks, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies. The power supply assembly in Figure 4 provides power to various components of the device in which the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply assembly is located. Those skilled in the art will appreciate that embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code. The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions.These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, such that the instructions, when executed by the processor of the computer or other programmable data processing device, produce means for implementing the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram. These computer program instructions can also be stored in a computer-readable memory capable of directing the computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means, which implement the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram. These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, such that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram. It should also be noted that the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, product, or device comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, product, or device. Without further limitation, elements defined by the phrase "comprising a..." do not preclude the presence of other identical elements in the process, method, product, or device comprising the elements. 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, etc.) referred to in 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 countries and regions, and corresponding operation portals are provided for users to choose to authorize or deny. The foregoing description is merely an embodiment of the present disclosure and is not intended to limit the present disclosure. Those skilled in the art will readily appreciate that various modifications and variations of the present disclosure are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this disclosure should be included in the scope of protection of this disclosure.

Claims

Claims Curvature, flatness and / or slope; and / or, modifying the intersection shape and triggering the synchronous modification of the trajectory route in the trajectory data corresponding to the target driving environment to adapt to the modified intersection shape; and / or, modifying the road direction and triggering the synchronous modification of the trajectory route in the trajectory data corresponding to the target driving environment to adapt to the modified road direction.

9. The method according to any one of claims 1 to 8, wherein: The method further includes: modifying the descriptive data under some description dimensions created for the target driving environment, while keeping the descriptive data under the remaining description dimensions unchanged, to generate a set of descriptive data capable of describing the new driving environment; and inputting the generated descriptive data into the generation model to output driving perception data corresponding to the new driving environment using the generation model.

10. The method according to any one of claims 1 to 9, wherein: The generation module adopts a constrained diffusion model, and inputting the constructed description data into a preset generation model includes: vectorizing the description data under different description dimensions to generate representation vectors; and inputting the representation vectors into the constrained diffusion model to serve as constraints of the constrained diffusion model.

11. A computing device, wherein: Comprises a memory, a processor and a communication component; the memory is used to store one or more computer instructions; the processor is coupled with the memory and the communication component, and is used to execute the one or more computer instructions, so as to execute the method for generating autonomous driving training samples as described in any one of claims 1 to 0.

12. A computer-readable storage medium storing a computer program, wherein: When the computer program is executed by one or more processors, it causes the one or more processors to execute the method for generating autonomous driving training samples according to any one of claims 1 to 10.

13. A computer program product, wherein: The method comprises a computer program, which, when executed by one or more processors, causes the one or more processors to execute the method for generating autonomous driving training samples according to any one of claims 1 to 10.

Images