Method for acquiring vehicle driving track, electronic device and storage medium

Abstract

The application discloses a vehicle driving track acquisition method, an electronic device and a storage medium, and belongs to the technical field of intelligent driving. The vehicle driving track acquisition method comprises the following steps: acquiring first feature information for indicating the mutual relationship between a first vehicle and a second vehicle and second feature information for indicating the mutual relationship between the first vehicle and a map; determining state information of the first vehicle according to the first feature information and the second feature information; determining an action probability distribution of an action space of the first vehicle under the state information, and performing at least twice sampling based on the action probability distribution; and determining the first vehicle driving track according to the sampling results of the at least twice sampling. The coupling between the state information of the vehicle and the actual state information of the vehicle can be effectively enhanced through the first feature information and the second feature information, and the reliability and timeliness of the vehicle driving track can be improved.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to a method for acquiring vehicle driving trajectory, an electronic device, and a storage medium. Background Technology

[0002] Vehicle trajectory planning requires an accurate and comprehensive understanding of the vehicle's motion characteristics, the surrounding environment, and the motion characteristics of other objects in the environment in order to determine the vehicle's state information and improve the reliability of trajectory planning.

[0003] In related technologies, there is an end-to-end network model for vehicle trajectory planning that can output the vehicle's trajectory over a future period of time based on the input environmental information, the motion state of other vehicles, and the motion state of the vehicle itself.

[0004] However, in the above methods, because the input information is relatively isolated from each other, the coupling between the network model's understanding of the vehicle's current state information generated based on the input information and the vehicle's actual state information is poor. For example, the network model has poor interpretability of the relationships between the vehicle and the environment, and between the vehicle and other vehicles. Summary of the Invention

[0005] This application provides a method for acquiring vehicle driving trajectory, an electronic device, and a storage medium to solve the technical problems existing in related technologies. Specifically, it includes the following technical solutions.

[0006] In a first aspect, this application provides a method for obtaining a vehicle's driving trajectory. The method includes: obtaining first feature information and second feature information, wherein the first feature information is used to indicate the relationship between a first vehicle and a second vehicle, and the second feature information is used to indicate the relationship between the first vehicle and a map; determining the state information of the first vehicle based on the first feature information and the second feature information; determining the action probability distribution of the first vehicle's action space under the state information, and performing at least two samplings based on the action probability distribution, wherein the action space is a set of actions performed by the first vehicle under different state information determined based on empirical data, and the sampling is used to simulate the execution of actions in the action space and update the state information; and determining the driving trajectory of the first vehicle based on the sampling results of the at least two samplings.

[0007] In some possible implementations, determining the first vehicle's trajectory based on the sampling results of the at least two samplings includes: determining a multimodal trajectory space for the first vehicle based on the sampling results, wherein the multimodal trajectory space is a set of candidate trajectories of the first vehicle corresponding to the sampling results; and determining the first vehicle's trajectory based on the multimodal trajectory space.

[0008] In some possible implementations, determining the first vehicle driving trajectory based on the multimodal trajectory space includes: determining the state value of the candidate trajectory in the multimodal trajectory space, the state value being used to indicate a prediction of the result of executing the candidate trajectory; and determining the candidate trajectory with the highest state value among the candidate trajectories as the driving trajectory.

[0009] In some possible implementations, determining the state information of the first vehicle based on the first feature information and the second feature information includes: processing the first feature information and the second feature information through a cross-attention mechanism to obtain a processing result; and determining the state information of the first vehicle based on the processing result.

[0010] In some possible implementations, the action space includes a discrete action space, which is used to indicate an action space where the number of actions is finite.

[0011] In some possible implementations, the actions in the action space are controlled based on the lateral velocity, longitudinal velocity, and angular velocity of the first vehicle.

[0012] In some possible implementations, the method further includes: smoothing the driving trajectory of the first vehicle after determining the driving trajectory based on the sampling results of the at least two samplings.

[0013] In some possible implementations, the first feature information includes a vectorized representation of the relationship between the first vehicle and the second vehicle, and the second feature information includes a vectorized representation of the relationship between the first vehicle and the map.

[0014] Secondly, this application provides an apparatus for acquiring a vehicle driving trajectory, comprising: an acquisition module for acquiring first feature information and second feature information, wherein the first feature information is used to indicate the relationship between a first vehicle and a second vehicle, and the second feature information is used to indicate the relationship between the first vehicle and a map; a first determination module for determining state information of the first vehicle based on the first feature information and the second feature information; a sampling module for determining the action probability distribution of the first vehicle's action space under the state information, and performing at least two samplings based on the action probability distribution, wherein the action space is a set of actions performed by the first vehicle under different state information determined based on empirical data, and the sampling is used to simulate the execution of actions in the action space and update the state information; and a second determination module for determining the driving trajectory of the first vehicle based on the sampling results of the at least two samplings.

[0015] In some possible implementations, when the second determining module determines the driving trajectory of the first vehicle based on the sampling results of the at least two samplings, it is used to determine the multimodal trajectory space of the first vehicle based on the sampling results, wherein the multimodal trajectory space is a set of candidate trajectories of the first vehicle corresponding to the sampling results; and to determine the driving trajectory of the first vehicle based on the multimodal trajectory space.

[0016] In some possible implementations, when the second determining module determines the first vehicle driving trajectory based on the multimodal trajectory space, it is used to determine the state value of the candidate trajectory in the multimodal trajectory space, the state value being used to indicate the prediction of the result of executing the candidate trajectory; and the candidate trajectory with the highest state value among the candidate trajectories is determined as the driving trajectory.

[0017] In some possible implementations, when the first determining module determines the state information of the first vehicle based on the first feature information and the second feature information, it processes the first feature information and the second feature information through a cross-attention mechanism to obtain a processing result; and determines the state information of the first vehicle based on the processing result.

[0018] In some possible implementations, the action space includes a discrete action space, which is used to indicate an action space where the number of actions is finite.

[0019] In some possible implementations, control is performed in the motion space based on the lateral velocity, longitudinal velocity, and angular velocity of the first vehicle.

[0020] In some possible implementations, the second determining module is further configured to smooth the driving trajectory after determining the first vehicle driving trajectory based on the sampling results of the at least two samplings.

[0021] In some possible implementations, the first feature information includes a vectorized representation of the relationship between the first vehicle and the second vehicle, and the second feature information includes a vectorized representation of the relationship between the first vehicle and the map.

[0022] In a third aspect, this application provides an electronic device for acquiring a vehicle's driving trajectory, comprising: a memory storing at least one program instruction for acquiring a vehicle's driving trajectory; and a processor, wherein when the program instruction is executed by the processor, the vehicle implements the method of the first aspect of this application or any possible implementation thereof.

[0023] In a fourth aspect, this application provides a computer program (product) comprising computer program / instructions, which are executed by a processor to cause a vehicle to implement the method of the first aspect of this application or any possible implementation thereof.

[0024] In a fifth aspect, this application provides a computer-readable storage medium having stored program instructions thereon for acquiring a vehicle's driving trajectory, which, when executed by one or more processors, cause the vehicle to implement the method of the first aspect or any possible implementation of the first aspect of this application.

[0025] The beneficial effects of the technical solution provided in this application include at least the following:

[0026] The technical solution provided in this application can effectively enhance the coupling between the vehicle's state information and the vehicle's actual state information by using second feature information indicating the relationship between the first vehicle and the environment and first feature information indicating the relationship between the first vehicle and the second vehicle. This allows the method provided in this application to more accurately describe the real state between the first vehicle and the environment, as well as between the first vehicle and the second vehicle, which is beneficial to ensuring the reliability and timeliness of the vehicle's driving trajectory. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of an implementation scenario provided in the embodiments of this application;

[0029] Figure 2 This is a flowchart of a method for obtaining vehicle driving trajectory provided in an embodiment of this application;

[0030] Figure 3 This is a schematic diagram illustrating the determination of the multimodal trajectory space of the first vehicle through sampling, provided in an embodiment of this application.

[0031] Figure 4 This is a schematic diagram of the structure of the device for obtaining vehicle driving trajectory provided in the embodiments of this application;

[0032] Figure 5 This is a schematic diagram of the structure of an electronic device for acquiring vehicle driving trajectory provided in an embodiment of this application. Detailed Implementation

[0033] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0034] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0035] An end-to-end network model is an artificial intelligence model architecture that directly maps inputs to outputs. In vehicle trajectory planning, an end-to-end network model can directly map sensor input data to the vehicle's trajectory planning control behavior, which helps reduce the complexity of the trajectory planning control system and improve the computation speed of trajectory planning.

[0036] However, in related technologies, environmental information, the vehicle's motion state, and the motion state of other vehicles are usually treated as separate elements and processed into vectors in isolation during the trajectory planning process. This results in the neglect of some important connections and interrelationships, leading to poor coupling between the understanding of the vehicle's state information and the actual state information, which in turn affects the reliability of the vehicle's trajectory planning.

[0037] In view of this, and addressing the technical problem of poor coupling between the understanding of vehicle state information and the actual state information of the vehicle in vehicle trajectory planning in related technologies, this application provides a method for vehicle trajectory planning. By performing special processing on the input information of vehicle trajectory planning, the connection relationships and mutual relationships between the vehicle and the environment, and between the vehicle and other vehicles, can be preserved to ensure the reliability of vehicle trajectory planning.

[0038] Figure 1 This is a schematic diagram of an implementation scenario provided in an embodiment of this application. (Reference) Figure 1 The implementation scenarios provided in this application include a vehicle 11 and a control unit 12.

[0039] The vehicle 11 is equipped with a variety of sensors for detecting the environment and target objects around the vehicle. These sensors can establish a communication connection with the control unit 12 via wired or wireless means, so that the control unit 12 can determine the status information of the vehicle 11 through the sensing data of the various sensors and plan the trajectory of the vehicle 11 based on the status information of the vehicle 11.

[0040] Optionally, the control unit 12 can be a single server or a server cluster consisting of multiple servers. The control unit 12 can also be an in-vehicle terminal capable of performing control functions. The various sensors installed in the vehicle 11 can include any sensor capable of detecting the surrounding environment and the motion information of other vehicles in the environment, such as lidar and millimeter-wave radar; this application makes no limitations in this regard.

[0041] Those skilled in the art should understand that the above-described vehicle 11 and control unit 12 are merely examples, and other existing or future vehicles or control units that are applicable to this application should also be included within the scope of protection of this application, and are hereby incorporated by reference.

[0042] Figure 2 This is a flowchart illustrating a method for obtaining a vehicle's driving trajectory according to an embodiment of this application. This method can, for example, be executed by a control unit mounted in the vehicle, and this application makes no limitations in this regard. See also... Figure 2 The method for vehicle trajectory planning provided in this application embodiment may include the following steps.

[0043] Step S210: Obtain first feature information and second feature information. The first feature information is used to indicate the relationship between the first vehicle and the second vehicle, and the second feature information is used to indicate the relationship between the first vehicle and the environment.

[0044] Optionally, the relationship between the first vehicle and the second vehicle may include, for example, spatial, kinematic, and semantic relationships, and this application does not impose any limitations in this regard. Specifically, the spatial relationship between the first vehicle and the second vehicle may include, for example, positional and angular relationships; the kinematic relationship may include, for example, directional relationships; and the semantic relationship may include, for example, whether the relationship between the first vehicle and the second vehicle is one of following, overtaking, or yielding.

[0045] In some embodiments, the method for obtaining the first feature information includes, for example, determining the first feature information based on the motion characteristics of a first vehicle and the motion characteristics of a second vehicle. The motion characteristics of the first vehicle include, for example, information such as the position, speed, angle, and heading of the first vehicle obtained by sensors mounted on the first vehicle; the motion characteristics of the second vehicle include, for example, information such as the position, speed, angle, and heading of the first vehicle obtained by sensors mounted on the second vehicle, etc., and this application does not impose any limitations in this regard. The method for obtaining the first feature information is, for example, calculating the relative speed and relative position between the first vehicle and the second vehicle based on the motion characteristics of the first vehicle and the motion characteristics of the second vehicle.

[0046] Similarly, the relationship between the first vehicle and the environment includes, for example, the spatial relationship, motion relationship, and semantic relationship between the first vehicle and the environment, and this application does not impose any limitations in this regard. Specifically, the spatial relationship between the first vehicle and the environment includes, for example, the global coordinate position, relative coordinate position, and angular coordinates of the first vehicle on the map; the motion relationship between the first vehicle and the environment includes, for example, the heading angle of the first vehicle; and the semantic relationship between the first vehicle and the environment includes, for example, the road type, lane attributes, and whether the first vehicle complies with the corresponding traffic rules.

[0047] In some embodiments, the method for obtaining the second feature information includes, for example, determining the second feature information based on the motion characteristics of the first vehicle and map information related to the environment in which the first vehicle is located. For example, the relative speed and relative position of the first vehicle on the map are calculated based on the motion characteristics of the first vehicle and the map information. The motion characteristics of the first vehicle include, for example, information such as the position, speed, angle, and heading of the first vehicle obtained by sensors mounted on the first vehicle; the map information includes, for example, map information generated from data obtained by sensors mounted on the first vehicle, or map information obtained through interaction with the cloud. This application does not impose any limitations in this regard.

[0048] Optionally, to reduce the data complexity of the first and second feature information, the first feature information is a vectorized representation of the relationship between the first and second vehicles, and the second feature information is a vectorized representation of the relationship between the first vehicle and the environment. Vectorization is a method of converting complex data into a set of numerical vectors.

[0049] In some embodiments, the representation of the first feature information / second feature information can refer to the following formula (1):

[0050] [x1, x2, ... x n (1)

[0051] Where n is an integer greater than or equal to 1, x1, x2, ..., xn n The first feature information consists of multiple components in the spatial relationship dimension, motion relationship dimension, and semantic relationship dimension between the first vehicle and the second vehicle, and the second feature information consists of multiple components in the spatial relationship dimension, motion relationship dimension, and semantic relationship dimension between the first vehicle and the environment.

[0052] By vectorizing the first and second feature information, the embodiments of this application can greatly reduce the complexity of the data. When applied to an end-to-end network model, the end-to-end network model can directly map the vectorized data to the control behavior of vehicle trajectory planning, thereby further improving the calculation speed of vehicle trajectory planning.

[0053] Step S220: Determine the status information of the first vehicle based on the first feature information and the second feature information.

[0054] For example, the state information of the first vehicle is the sum of the environment in which the first vehicle is located at any one travel moment in a trip (i.e., a set of multiple travel moments), the motion information of other vehicles in the environment, and the motion information of the vehicle itself. It can be used, but is not limited to, to provide a decision basis for the trajectory planning of the first vehicle.

[0055] Considering that in practical applications, to gain a more comprehensive and accurate understanding of the first vehicle's state information, the acquisition of the first and second feature information aims to ensure the breadth and depth of information coverage, potentially leading to some redundant information in both feature information. Therefore, to avoid interference from redundant information in vehicle trajectory planning and to improve the computational speed of trajectory planning, important information can be selectively filtered and redundant information removed when determining the first vehicle's state information. Important information refers to information in the first and second feature information that is helpful for vehicle trajectory planning, while redundant information refers to information in the first and second feature information that is not helpful for vehicle trajectory planning.

[0056] In some embodiments, determining the state information of a first vehicle based on first feature information and second feature information includes, for example, processing the first feature information and second feature information through a cross-attention mechanism to obtain a processing result; and determining the state information of the first vehicle based on the processing result. The cross-attention mechanism can be used, but is not limited to, to identify important and redundant information during the process of determining the state information of the first vehicle, and to highlight important information and suppress redundant information by dynamically allocating weights.

[0057] Step S230: Determine the action probability distribution of the first vehicle's action space under the state information, and perform at least two samplings based on the action probability distribution. The action space is a set of actions performed by the first vehicle under different state information, determined based on empirical data. The sampling is used to simulate the execution of actions in the action space and update the state information.

[0058] For example, empirical data may include data obtained by testing actions that the vehicle may perform under different state information. The action space may include, for example, a set of all actions that the vehicle may perform under different state information, as determined based on empirical data, and may be used, but is not limited to, providing empirical guidance for vehicle trajectory planning.

[0059] The actions performed by the first vehicle under different state information are usually accomplished by controlling the speed, acceleration, angular velocity, and other control signals of the first vehicle. Therefore, the actions in the action space of the first vehicle can be represented by the speed, acceleration, angular velocity, and other control signals of the first vehicle. For example, the actions in the action space are controlled based on the lateral speed, longitudinal speed, and angular velocity of the first vehicle. The representation of the action space can be referenced in the following formula (2):

[0060] a = [v x ,v y ,ω] (2)

[0061] Where a represents the action in the motion space of the first vehicle, and v x Let v be the lateral velocity of the first vehicle. y Let ω be the longitudinal velocity of the first vehicle, and ω be the angular velocity of the first vehicle. The lateral velocity is, for example, the velocity in the horizontal plane perpendicular to the heading of the first vehicle, and the longitudinal velocity is, for example, the velocity in the horizontal plane parallel to the heading of the first vehicle.

[0062] Considering the need to improve data processing and computation speed in the trajectory planning scenario of the first vehicle to ensure the timeliness of the trajectory planning, the action space of the first vehicle may include, for example, a discrete action space. This discrete action space indicates a finite number of actions. By setting a discretized action space, the first vehicle can sample and simulate a limited number of actions, thereby improving the computational efficiency of the first vehicle's trajectory planning.

[0063] When the motion space of the first vehicle is a discrete motion space, and the actions in the motion space are controlled based on the lateral velocity, longitudinal velocity, and angular velocity of the first vehicle, it is necessary to discretize the control signal corresponding to each action. For example, the control signal is made to take values ​​within a specified range at specified intervals to obtain discretized control signals. Then, multiple discretized control signals are arranged and combined to obtain a finite number of actions in the discretized motion space. The value range can be adjusted according to the motion performance of the first vehicle, and the specified interval can be adjusted according to the performance of the control unit installed in the first vehicle. This application does not impose any limitations in this regard.

[0064] In some embodiments, when the control signals corresponding to the actions in the discrete action space are the lateral velocity, longitudinal velocity, and angular velocity of the first vehicle, the specified range for discretization of the lateral velocity is -1.5 m / s to 1.5 m / s, the specified interval for discretization of the lateral velocity is 0.1, and the number of discretized lateral velocity values ​​is 30; the specified range for discretization of the longitudinal velocity is -0 m / s to 3 m / s, the specified interval for discretization of the longitudinal velocity is 0.1, and the number of discretized longitudinal velocity values ​​is 30; the specified range for discretization of the angular velocity is -1.5 rad / s to 1.5 rad / s, the specified interval for discretization of the angular velocity is 0.1, and the number of discretized angular velocity values ​​is 30. The 30 lateral velocities, 30 longitudinal velocities, and 30 angular velocities are arranged and combined to obtain the discrete action space of the first vehicle.

[0065] The probability distribution of actions in the action space of the first vehicle under the given state information can be used, but is not limited to, to indicate the probability of the first vehicle performing an action in the action space under the given state information at a certain driving time. Specifically, the sum of the probabilities of all actions performed by the first vehicle under the given state information at a certain driving time is 1.

[0066] Considering that the actions a first vehicle needs to perform under the same state information may differ across different journeys, in other words, the actions that empirical data indicates a vehicle needs to perform under different state information may vary depending on the vehicle's journey. Therefore, the method provided in this application can determine the action probability distribution of the first vehicle's action space based on the specific state information of the first vehicle in the current journey, and sample the actions in the action space at least twice under the guidance of the action probability distribution. This ensures that the actions a first vehicle needs to perform under a specific state information in the current journey are supported by empirical data and can also exhibit some variability depending on the journey, thus better reflecting the actual situation of the first vehicle during its journey.

[0067] As mentioned above, vectorized representation can reduce the complexity of data. In view of this, the state information of the first vehicle includes, for example, a state vector obtained by performing vectorization processing on the state information.

[0068] In some embodiments, the operation of determining the action probability distribution of the action space of the first vehicle according to the state information and performing at least two samplings in the action space under the guidance of the action probability distribution is performed by a policy network model. Wherein, the policy network model can be used for, but not limited to, outputting the probability distribution of the first vehicle in the action space according to the input current state information of the first vehicle, and for the expression form of the policy network model, reference can be made to the following formula (3) for example:

[0069] P(A=a | S=s t )=π(a | s t , θ)(3)

[0070] Wherein, A is the action space of the first vehicle, S is the set of state information of the first vehicle, a is an action in the action space of the first vehicle, s t is the state information of the first vehicle at time t, θ is the network parameter of the policy network model, P is the probability distribution of the first vehicle in the action space A when the state information is s t .

[0071] For example, performing at least two samplings based on the action probability distribution includes: randomly determining one action in the action space as a sampling action based on the action probability distribution, and performing at least two samplings on the sampling action.

[0072] When the action space of the first vehicle is a discrete action space, the method for randomly determining one action in the action space as a sampling action based on the action probability distribution includes, for example: dividing the interval [0, 1] into sub-intervals of different lengths according to the probability of each action in the discrete action space, taking a random number in the interval [0, 1], and the action corresponding to the sub-interval to which the random number falls is the sampling action.

[0073] As mentioned above, sampling can be used for, but not limited to, simulating the execution of actions in the action space and updating state information. After randomly determining one action in the action space as a sampling action based on the action probability distribution, the state information of the first vehicle can be updated according to the result obtained after simulating the first vehicle executing the sampling action and driving for a preset period of time.

[0074] Considering that in the trajectory planning of the first vehicle, when simulating the first vehicle performing sampling actions over a preset time period, it is generally desirable that the preset time period be long enough, and that the sampling actions simulated during the first vehicle's driving should more closely resemble the actual driving scenario of the first vehicle within the preset time period. This would allow the first vehicle sufficient warning time for potential future driving scenarios to cope with possible emergencies. Therefore, the preset time period can be divided into multiple sampling steps at specified sampling intervals. In a sampling process where at least two sampling actions are performed, the sampling process may include multiple sampling steps.

[0075] When the sampling process includes multiple sampling steps, the motion probability distribution of the first vehicle's motion space can be determined again based on the updated state information after at least two samplings. The state information of the first vehicle is then updated to the next sampling step based on the updated motion probability distribution, until the trajectory planning target of the first vehicle in the preset time period is completed, or until the number of sampling steps is N, where N is an integer ≥1. The specific value of N can be adjusted according to the actual application.

[0076] The method for determining the sampling interval includes, for example, determining the sampling interval based on the travel time of the first vehicle, such as sampling at each fixed time interval; or determining the sampling interval based on the travel distance, such as sampling once within each fixed travel distance interval. The size of the sampling interval (e.g., the size of the travel time interval or the size of the travel distance interval) can be adjusted according to the actual application scenario. For example, when the current state information of the first vehicle is relatively stable (e.g., the road, pedestrians around the vehicle, and changes in the size of the second vehicle), the sampling interval can be set to be larger; when the current state information of the first vehicle is unstable (e.g., the road, pedestrians around the vehicle, and changes in the size of the second vehicle are relatively small), the sampling interval can be set to be smaller.

[0077] Step S240: Determine the trajectory of the first vehicle based on the sampling results of at least two samplings.

[0078] For example, "at least two" could be two, three, four, or other numbers greater than two, and this application makes no limitation in this regard. The sampling results can be used, but are not limited to, indicating multiple state information corresponding to multiple sampling steps during a sampling process involving at least two samplings. As mentioned above, the state information of the first vehicle is the sum of the environment in which the first vehicle is located at any given moment during a trip, the motion information of other vehicles in the environment, and the motion information of the vehicle itself (i.e., including all the aforementioned information). Therefore, the multiple state information obtained during the sampling process can indicate the position and motion information of the first vehicle at multiple moments during the trip, and thus determine the travel trajectory of the first vehicle for a certain period after the current moment of travel based on the multiple position and motion information.

[0079] As mentioned earlier, sampling under the guidance of the action probability distribution requires at least two sampling operations. In other words, the method provided in this application can determine at least two sampling actions under the guidance of the action probability distribution, and obtain multiple candidate trajectories of the first vehicle based on the sampling results of the at least two sampling actions. In some embodiments, determining the driving trajectory of the first vehicle based on the sampling results of at least two sampling operations includes: determining the multimodal trajectory space of the first vehicle based on the sampling results, wherein the multimodal trajectory space is a set of candidate trajectories of the first vehicle corresponding to the sampling results; and determining the driving trajectory of the first vehicle based on the multimodal trajectory space.

[0080] Figure 3 This is a schematic diagram of a method for determining the multimodal trajectory space of a first vehicle through sampling, as provided in an embodiment of this application.

[0081] like Figure 3 As shown, the state information of the first vehicle at time t is s. t Based on the state information s of the first vehicle at time t t The motion space of the first vehicle is determined in state information s t After obtaining the action probability distribution, three samplings can be performed under the guidance of the action probability distribution. The sampling actions corresponding to the three samplings are a, ... t a t '、a t ".

[0082] The sampling actions corresponding to the three samplings are simulated and executed respectively, and the state information of the first vehicle is updated to time t+1 to obtain the state information s of the first vehicle at time t+1. t+1、 s t+1 '、s t+1 Then, based on the state information s of the first vehicle at time t+1... t+1 The motion space of the first vehicle is determined in state information s t+1 After obtaining the action probability distribution, three samplings can be performed based on the action probability distribution to obtain a. t+1 The sampling action includes the sampling process; based on the state information s of the first vehicle at time t+1. t+1 'Determine the motion space of the first vehicle in state information s' t+1 After obtaining the action probability distribution, three samplings can be performed based on the action probability distribution to obtain a. t+1 The sampling action includes 'sampling operations; based on the state information s of the first vehicle at time t+1'. t+1 "The motion space of the first vehicle is determined to be in state information s" t+1 After obtaining the action probability distribution, three samplings can be performed based on the action probability distribution to obtain a. t+1The sampling process continues until the trajectory planning target of the first vehicle within the preset time period is achieved, or the number of sampling steps is N, at time t+n, the data including s is obtained. t+n s t+n '、s t+n Once the state information, including the state information, is obtained, sampling stops. Each sampling yields candidate trajectories containing multiple state information, and the multiple candidate trajectories corresponding to multiple samplings constitute the multimodal trajectory space of the first vehicle.

[0083] In some embodiments, if the condition for stopping sampling is completing the trajectory planning target of the first vehicle within a preset time period, the method for determining the driving trajectory of the first vehicle based on the multimodal trajectory space includes, for example, determining the candidate trajectory in the multimodal trajectory space that has the smallest difference from the trajectory planning target of the first vehicle as the driving trajectory of the first vehicle. The candidate trajectory that has the smallest difference from the trajectory planning target of the first vehicle is thus a candidate trajectory that can complete the trajectory planning target of the first vehicle. For example, the candidate trajectory in the multimodal trajectory space whose endpoint position has the smallest difference from the endpoint position of the trajectory planning target is determined as the driving trajectory.

[0084] Considering that the vehicle trajectory planning objective may constantly change based on the vehicle's actual state information, the number of sampling steps will also be frequently adjusted according to the trajectory planning objective. Furthermore, when the driving distance or time corresponding to the trajectory planning objective is long, the number of sampling steps in a single trajectory planning objective may be large, placing significant pressure on the vehicle's hardware and software resources. Therefore, to reduce the pressure on the vehicle's hardware and software resources, a fixed number of sampling steps can be used for each sampling.

[0085] Since the number of sampling steps is fixed, if the vehicle fails to complete the trajectory planning target after executing a fixed number of sampling steps, the driving trajectory cannot be determined based on the difference between different candidate trajectories in the multimodal trajectory space and the trajectory planning target. As mentioned above, each sampling corresponds to a candidate trajectory that includes multiple state information. The method provided in this application embodiment can also filter candidate trajectories based on the value of multiple state information to determine the driving trajectory. In some embodiments, the state values ​​of candidate trajectories in the multimodal trajectory space are determined; the candidate trajectory with the highest state value is determined as the driving trajectory. The state value is used to indicate the prediction of the result of executing the candidate trajectory; for example, the state value of the candidate trajectory indicates whether the first vehicle can complete the trajectory planning target after executing the candidate trajectory.

[0086] In some embodiments, after determining the driving trajectory of the first vehicle based on the sampling results of at least two samplings, the method provided in this application can further smooth the driving trajectory to improve the comfort and safety of the first vehicle when driving according to the driving trajectory. For example, the driving trajectory of the first vehicle can be smoothed by interpolation.

[0087] The technical solution provided in this application can effectively enhance the coupling between the vehicle's state information and the vehicle's actual state information by using the second characteristic information of the relationship between the first vehicle and the environment and the first characteristic information of the relationship between the first vehicle and the second vehicle. This allows the method provided in this application to more accurately describe the real state between the first vehicle and the environment, as well as between the first vehicle and the second vehicle, which is beneficial to ensuring the reliability and timeliness of the vehicle's driving trajectory.

[0088] In some other possible implementations, this application also provides an apparatus for acquiring a vehicle's driving trajectory. Figure 4 This is a schematic diagram of the device for acquiring vehicle driving trajectory provided in an embodiment of this application. See also... Figure 4 The vehicle trajectory planning device provided in this application includes the following modules.

[0089] The acquisition module 410 is used to acquire first feature information and second feature information. The first feature information is used to indicate the relationship between the first vehicle and the second vehicle, and the second feature information is used to indicate the relationship between the first vehicle and the map.

[0090] The first determining module 420 is used to determine the state information of the first vehicle based on the first feature information and the second feature information.

[0091] The sampling module 430 is used to determine the action probability distribution of the first vehicle's action space under the state information, and to perform at least two samplings based on the action probability distribution. The action space is a set of actions performed by the first vehicle under different state information, determined based on empirical data. The sampling is used to simulate the execution of actions in the action space and update the state information.

[0092] The second determining module 440 is used to determine the driving trajectory of the first vehicle based on the sampling results of the at least two samplings.

[0093] In some embodiments, when determining the driving trajectory of the first vehicle based on the sampling results of the at least two samplings, the second determining module 440 is used to determine the multimodal trajectory space of the first vehicle based on the sampling results, wherein the multimodal trajectory space is a set of candidate trajectories of the first vehicle corresponding to the sampling results; and to determine the driving trajectory of the first vehicle based on the multimodal trajectory space.

[0094] In some embodiments, when the second determining module 440 determines the first vehicle driving trajectory based on the multimodal trajectory space, it is used to determine the state value of the candidate trajectory in the multimodal trajectory space, the state value being used to indicate the prediction of the result of executing the candidate trajectory; and to determine the candidate trajectory with the highest state value among the candidate trajectories as the driving trajectory.

[0095] In some embodiments, when the first determining module 420 determines the state information of the first vehicle based on the first feature information and the second feature information, it is used to process the first feature information and the second feature information through a cross-attention mechanism to obtain a processing result; and to determine the state information of the first vehicle based on the processing result.

[0096] In some embodiments, the action space includes a discrete action space, which is used to indicate an action space where the number of actions is finite.

[0097] In some embodiments, the actions in the action space are controlled based on the lateral velocity, longitudinal velocity, and angular velocity of the first vehicle.

[0098] In some embodiments, the second determining module 440 is further configured to smooth the driving trajectory after determining the first vehicle driving trajectory based on the sampling results of the at least two samplings.

[0099] In some embodiments, the first feature information includes a vectorized representation of the relationship between the first vehicle and the second vehicle, and the second feature information includes a vectorized representation of the relationship between the first vehicle and the map.

[0100] It should be understood that the above Figure 4 The device for acquiring vehicle driving trajectory provided is illustrated only by the above-described division of functional modules. In practical applications, the functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The device for acquiring vehicle driving trajectory and the method for acquiring vehicle driving trajectory provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the method for acquiring vehicle driving trajectory embodiment.

[0101] In some other possible implementations, this application also provides an electronic device for acquiring vehicle driving trajectory. Figure 5 This is a schematic diagram of the structure of an electronic device for acquiring vehicle driving trajectory provided in an embodiment of this application. See also... Figure 5The electronic device for acquiring vehicle driving trajectory provided in this application embodiment includes:

[0102] The memory 510 stores at least one program instruction for obtaining the vehicle's driving trajectory.

[0103] When the processor 520 executes the above program instructions, it enables the vehicle to achieve the above-mentioned combination. Figure 2 The steps of the described method and its various embodiments are described below. Depending on the implementation, the processor 520 may be one or more types of processors, including but not limited to DSP (digital signal processor), ASIC (application specific integrated circuit), FPGA (field-programmable gate array), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., and the number of such devices can be determined according to actual needs.

[0104] In some other possible implementations, this application also provides a computer program (product) comprising computer programs / instructions, which are executed by a processor to cause the vehicle to achieve the above-described combination. Figure 2 The steps of the described method and its various embodiments.

[0105] In some other possible embodiments, this application also provides a computer-readable storage medium storing program instructions for acquiring a vehicle's driving trajectory. When these program instructions are executed by one or more processors, they cause the vehicle to achieve the above-mentioned combination. Figure 2 The steps of the described method and its various embodiments are described. The computer-readable storage medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0106] It should be noted that the electronic device in this application may also be referred to as a display device. Furthermore, all information, data (including but not limited to image data) and signals involved in this application are authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0107] It should also be noted that the terms "first," "second," etc. (if applicable) in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0108] The term "and / or" in the embodiments of this application is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0109] The above description is only for the purpose of enabling those skilled in the art to understand the technical solution of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the scope of protection of this application.

Claims

1. A method for obtaining a vehicle's driving trajectory, characterized in that, The method includes: Acquire first feature information indicating the relationship between the first vehicle and the second vehicle, and second feature information indicating the relationship between the first vehicle and the map; The status information of the first vehicle is determined based on the first feature information and the second feature information; The action space of the first vehicle is determined under the state information, and at least two samples are taken based on the action probability distribution. The action space is a set of actions performed by the first vehicle under different state information, determined based on empirical data. The sampling is used to simulate the execution of actions in the action space and update the state information. The multimodal trajectory space of the first vehicle is determined based on the sampling results of the at least two samplings, wherein the multimodal trajectory space is a set of candidate trajectories of the first vehicle corresponding to the sampling results; Determine the state value of the candidate trajectory, the state value being used to indicate a prediction of the result of executing the candidate trajectory; The candidate trajectory with the highest state value among the candidate trajectories is determined as the driving trajectory of the first vehicle.

2. The method according to claim 1, characterized in that, Determining the state information of the first vehicle based on the first feature information and the second feature information includes: The first feature information and the second feature information are processed by a cross-attention mechanism to obtain the processing result; The status information of the first vehicle is determined based on the processing result.

3. The method according to claim 1, characterized in that, The action space includes a discrete action space, which is used to indicate an action space where the number of actions is finite.

4. The method according to claim 1, characterized in that, The actions in the action space are controlled based on the lateral velocity, longitudinal velocity, and angular velocity of the first vehicle.

5. The method according to claim 1, characterized in that, The method further includes: After determining the first vehicle's trajectory based on the sampling results of the at least two samplings, the trajectory is smoothed.

6. The method according to claim 1, characterized in that, The first feature information includes a vectorized representation of the relationship between the first vehicle and the second vehicle, and the second feature information includes a vectorized representation of the relationship between the first vehicle and the map.

7. An electronic device, characterized in that, include: The memory contains program instructions for acquiring the vehicle's driving trajectory. as well as A processor, when the program instructions are executed by the processor, causes the vehicle to perform the method described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, It stores program instructions for acquiring the vehicle's driving trajectory, which, when executed by one or more processors, cause the vehicle to perform the method described in any one of claims 1-6.

Images