Methods, devices, equipment, and storage media for determining autonomous driving decision information

Abstract

This application relates to a method, apparatus, device, and storage medium for determining decision information for autonomous driving. The method includes: first, inputting road environment data of the vehicle's location into a homogeneous, heterogeneous prediction model to predict the trajectory, obtaining multiple candidate trajectories; then, processing each candidate trajectory to obtain a target predicted trajectory and its uncertainty; and finally, inputting the target predicted trajectory and its uncertainty into a decision model to obtain decision information. This method can correct existing decision models based on trajectory prediction results based on the uncertainty of the trajectory prediction. When the prediction model may fail, it exhibits higher trajectory prediction uncertainty to form a higher collision risk estimate with the conflicting object, thereby timely detecting and avoiding safety hazards caused by prediction failure.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to a method, apparatus, device, and storage medium for determining autonomous driving decision information. Background Technology

[0002] In the field of autonomous driving technology, autonomous driving decision-making technology predicts the behavior of other road users by perceiving environmental data, thereby achieving autonomous driving decision-making. Prediction algorithms are a crucial component of autonomous driving decision-making technology, providing the decision-making system with predicted trajectory information of other road users' behavior, enabling the autonomous driving system to make driving decisions based on this predicted trajectory information.

[0003] However, due to complex traffic environments, uncertainties among traffic participants, and the occurrence of unexpected events, prediction algorithms may have inaccurate predictions of the trajectories of other traffic participants. This can lead to autonomous driving systems making incorrect decisions based on inaccurate predictions, thereby increasing the risk of traffic accidents.

[0004] Therefore, determining accurate information for autonomous driving decisions has become an urgent problem to be solved. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, device, and storage medium for accurately determining autonomous driving decision information in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a method for determining autonomous driving decision information, including:

[0007] The road environment data where the vehicle is located is input into the isomorphic heterogeneous prediction model to predict the trajectory, and multiple candidate trajectories are obtained.

[0008] Each candidate trajectory is processed to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory;

[0009] The target predicted trajectory and its uncertainty are input into the decision model to obtain decision information.

[0010] In one embodiment, the decision model includes: a collision constraint model, a road boundary model, a traffic light model, a control input limit constraint model, a reference path position prediction model, a vehicle state model, and a control variable model. The collision constraint model is used to predict collisions between the vehicle and other traffic participants; the road boundary model is used to predict collisions between the vehicle and the road boundary; the traffic light model is used to predict violations by the vehicle; the control input limit constraint model is used to predict when the vehicle exceeds a preset control variable; the reference path position prediction model is used to predict the vehicle's position on the reference path at the next moment; the vehicle state model is used to predict the vehicle's driving state at the next moment; and the control variable model is used to calculate the vehicle's control variables at the current moment.

[0011] In one embodiment, the method further includes:

[0012] A collision constraint model is constructed based on the participants' driving angle, distance parameters, trajectory parameters, and uncertainty parameters; the distance parameters represent the projection parameters of the relative distance between the center position of the vehicle and the center position of the participants onto the coordinate axes.

[0013] In one embodiment, the method further includes:

[0014] A road boundary model is constructed based on the width of the left boundary of the road, the width of the right boundary of the road, and the vehicle profile error.

[0015] In one embodiment, the method further includes:

[0016] Construct a control input limit constraint model based on the maximum and minimum parameters of the control variables and the vehicle's control variables.

[0017] In one embodiment, the method further includes:

[0018] A reference path position prediction model is constructed based on the vehicle's position on the reference path, the vehicle's speed at any given time, the vehicle's travel time, and the vehicle's position on the reference path at the next moment.

[0019] In one embodiment, the method further includes:

[0020] A vehicle state model is constructed based on the vehicle's input state at any given time, the vehicle's control variables at any given time, and the vehicle's driving state at the next time.

[0021] In one embodiment, the method further includes:

[0022] A control variable model is constructed based on the vehicle's input state at any given time, the vehicle's control variables at any given time, and the vehicle's position on the reference path point.

[0023] In one embodiment, the above-described processing of each candidate trajectory to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory includes:

[0024] The average coordinate position at the same time is calculated for all candidate trajectories to obtain the average coordinate position at the same time, and the target predicted trajectory is determined based on the average coordinate position at all times.

[0025] The variance of the coordinate positions at the same time on all candidate trajectories is calculated to obtain the variance coordinate positions at the same time, and the uncertainty of the target predicted trajectory is determined based on the variance coordinate positions at all times.

[0026] In one embodiment, the method further includes:

[0027] Acquire sample road environment data; the sample road environment data includes sample traffic participant status data, sample traffic light status data, sample map status data, and traffic participant trajectory data;

[0028] Based on road environment data, the initial prediction models with different parameters are trained to obtain the prediction models with different parameters. Each initial prediction model includes a history encoder, an interaction feature encoder, a map encoder, and a trajectory decoder.

[0029] Secondly, this application also provides a device for determining autonomous driving decision information, comprising:

[0030] The prediction module is used to input the road environment data where the vehicle is located into the isomorphic heterogeneous prediction model to predict the trajectory and obtain multiple candidate trajectories.

[0031] The processing module is used to process each candidate trajectory to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory;

[0032] The acquisition module is used to input the target predicted trajectory and the uncertainty of the target predicted trajectory into the decision model to obtain decision information.

[0033] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0034] The road environment data where the vehicle is located is input into the isomorphic heterogeneous prediction model to predict the trajectory, and multiple candidate trajectories are obtained.

[0035] Each candidate trajectory is processed to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory;

[0036] The target predicted trajectory and its uncertainty are input into the decision model to obtain decision information.

[0037] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0038] The road environment data where the vehicle is located is input into the isomorphic heterogeneous prediction model to predict the trajectory, and multiple candidate trajectories are obtained.

[0039] Each candidate trajectory is processed to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory;

[0040] The target predicted trajectory and its uncertainty are input into the decision model to obtain decision information.

[0041] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0042] The road environment data where the vehicle is located is input into the isomorphic heterogeneous prediction model to predict the trajectory, and multiple candidate trajectories are obtained.

[0043] Each candidate trajectory is processed to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory;

[0044] The target predicted trajectory and its uncertainty are input into the decision model to obtain decision information.

[0045] The aforementioned method, apparatus, device, and storage medium for determining autonomous driving decision information first inputs the road environment data of the vehicle into a homogeneous but heterogeneous prediction model for trajectory prediction, obtaining multiple candidate trajectories. Then, each candidate trajectory is processed to obtain the target predicted trajectory and its uncertainty. Finally, the target predicted trajectory and its uncertainty are input into a decision model to obtain decision information. This method considers the uncertainty of trajectory prediction in the prediction model and obtains the final decision information based on the trajectory prediction uncertainty and the decision model, providing more accurate decision information for autonomous vehicles. Compared with existing technologies, this method modifies existing decision models based on trajectory prediction results based on the uncertainty of trajectory prediction. When the prediction model may fail, it exhibits higher trajectory prediction uncertainty to form a higher collision risk estimate with the conflicting object, thereby timely detecting and avoiding safety hazards caused by prediction failure. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies 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.

[0047] Figure 1 This is an application environment diagram of a method for determining autonomous driving decision information in one embodiment;

[0048] Figure 2 This is a flowchart illustrating a method for determining autonomous driving decision information in one embodiment;

[0049] Figure 3 This is a schematic diagram of the vehicle's kinematics model;

[0050] Figure 4 The simplified model diagram of the vehicle shape is used;

[0051] Figure 5 A simplified model diagram of the traffic participants used;

[0052] Figure 6 This is a flowchart illustrating a method for determining autonomous driving decision information in another embodiment;

[0053] Figure 7 This is a flowchart illustrating a method for determining autonomous driving decision information in another embodiment;

[0054] Figure 8 This is a framework diagram of the decision-making model constructed in this application;

[0055] Figure 9 This is a flowchart illustrating a method for determining autonomous driving decision information in another embodiment;

[0056] Figure 10 This is a structural block diagram of a device for determining autonomous driving decision information in one embodiment. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0059] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0060] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0061] The rapid development of autonomous driving technology is leading a transformation in the future of transportation. The realization of autonomous vehicles requires highly accurate and reliable decision-making methods to ensure safe and efficient driving operations. Autonomous driving decision-making technology predicts the behavior of other road users by perceiving environmental data, achieving autonomous driving decision-making. Among these, artificial intelligence-based prediction algorithms are a crucial component of autonomous driving decision-making technology. Based on perceived data and environmental information, they provide decision-making technology with predicted trajectory information of other road users' behaviors, enabling the autonomous driving system to make driving decisions based on this predicted trajectory information.

[0062] However, due to complex traffic environments, uncertainties among traffic participants, and unexpected events—for example, on congested roads, traffic participants may exhibit unusual behaviors such as changing lanes, stopping, or suddenly accelerating; or changes in weather conditions, road conditions, and traffic rules—prediction algorithms may produce inaccurate predictions of other traffic participants' trajectories. This can lead to autonomous driving systems making incorrect decisions based on inaccurate predictions, thereby increasing the risk of traffic accidents. Therefore, determining accurate autonomous driving decision-making information is a pressing issue. This application aims to address this problem.

[0063] Having described the background technology of the method for determining autonomous driving decision information provided in the embodiments of this application, the implementation environment involved in the method for determining autonomous driving decision information provided in the embodiments of this application will be briefly described below. The method for determining autonomous driving decision information provided in the embodiments of this application can be applied to, for example... Figure 1 The computer device shown includes a processor and a memory connected via a system bus. The memory stores a computer program, and the processor executes the computer program to perform the steps described in the method embodiments below. Optionally, the computer device may also include an input / output interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to communicate with external terminals via a network connection. Optionally, the computer device may be a server, a personal computer, a personal digital assistant, or other terminal devices, such as tablet computers, mobile phones, etc., or it may be a cloud or remote server. This application embodiment does not limit the specific form of the computer device.

[0064] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0065] After introducing the application scenarios of the method for determining autonomous driving decision information provided in the embodiments of this application above, the following focuses on the method for determining autonomous driving decision information described in this application.

[0066] In one embodiment, such as Figure 2 As shown, a method for determining autonomous driving decision information is provided, which is then applied to... Figure 1 Taking a computer device as an example, the explanation includes the following steps:

[0067] S201. Input the road environment data where the vehicle is located into the isomorphic heterogeneous prediction model to predict the trajectory and obtain multiple candidate trajectories.

[0068] Here, "vehicle" refers to an autonomous vehicle, and the road environment data includes: the historical status of other traffic participants around the vehicle, the status of traffic lights around the vehicle, and map information about the surrounding area. The prediction model can be a multi-agent prediction network model; it is used to predict the future trajectories of other traffic participants (e.g., vehicles, pedestrians, and non-motorized vehicles) based on the road environment data of the vehicle. The prediction model includes: a traffic participant history encoder, an interaction feature encoder, a map encoder, and a future trajectory decoder.

[0069] Among them, the isomorphic heteroparametric prediction model can be multiple prediction models with the same structure but different parameters. The process of obtaining the isomorphic heteroparametric prediction model can include:

[0070] Step 1: Randomly initialize the initial parameters in the prediction model;

[0071] Step 2: Input the road environment data where the vehicle is located and the trajectory information of other traffic participants into the initialized prediction model, train the initialized prediction model, and obtain the trained prediction model.

[0072] Step 3: Repeat Step 1 and Step 2 M times (each time the initial parameters of the prediction model are different) to obtain M trained isomorphic heterogeneous models.

[0073] In this embodiment, when it is necessary to determine the vehicle's decision information based on the road environment data surrounding the vehicle, the road environment data where the vehicle is located can be obtained first. This data is then input into multiple pre-trained, homogeneous, heterogeneous prediction models to predict the trajectory, resulting in candidate trajectories predicted by each traffic participant in each prediction model. For example, candidate trajectory 1 for pedestrian 1 in prediction model 1, candidate trajectory 2 for electric vehicle 2 in prediction model 1, candidate trajectory 3 for vehicle 3 in prediction model 1, and candidate trajectory 4 for pedestrian 1 in prediction model 2, candidate trajectory 5 for electric vehicle 2 in prediction model 2, and candidate trajectory 6 for vehicle 3 in prediction model 2 are obtained.

[0074] S202. Process each candidate trajectory to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory.

[0075] In this embodiment, after obtaining the candidate trajectories predicted by each traffic participant in each prediction model, the candidate trajectories predicted by any traffic participant in each prediction model are processed to obtain the target predicted trajectory of the traffic participant, and the uncertainty of the candidate trajectories predicted by any traffic participant in each prediction model is processed to obtain the uncertainty of the target predicted trajectory of the traffic participant. Optionally, a weighted average is performed on the candidate trajectories predicted by any traffic participant in each prediction model to obtain the target predicted trajectory of the traffic participant, and the difference is performed on the candidate trajectories predicted by any traffic participant in each prediction model to obtain the uncertainty of the target predicted trajectory of the traffic participant.

[0076] S203. Input the target predicted trajectory and the uncertainty of the target predicted trajectory into the decision model to obtain decision information.

[0077] The decision model refers to a model that can determine a vehicle's decisions based on the trajectories of other traffic participants. The decision model includes an objective function and constraints. The decision model determines decision information by minimizing the objective function while satisfying the constraints, thereby determining the decision information for the autonomous vehicle. The objective function includes: the contour error of the vehicle's lateral deviation from the reference path, the hysteresis error of the vehicle's longitudinal deviation from the reference path, the deviation of the vehicle's speed from the reference speed, the magnitude of the vehicle's acceleration, and the angle of the vehicle's front wheels relative to the horizontal direction. Constraints include: collision constraints with other traffic participants, traffic light constraints, speed constraints, road boundary constraints, and control input limit constraints.

[0078] In this embodiment of the application, after obtaining the target predicted trajectory and the uncertainty of the target predicted trajectory, the target predicted trajectory and the uncertainty of the target predicted trajectory can be input into the decision model for calculation to obtain the vehicle's decision information, and the decision information is sent to the vehicle to instruct the vehicle to make a decision based on the decision information.

[0079] The method for determining autonomous driving decision information provided in this application first inputs the road environment data of the vehicle into a homogeneous but heterogeneous prediction model for trajectory prediction, obtaining multiple candidate trajectories. Then, each candidate trajectory is processed to obtain the target predicted trajectory and its uncertainty. Finally, the target predicted trajectory and its uncertainty are input into a decision model to obtain decision information. This method considers the uncertainty of trajectory prediction in the prediction model and obtains the final decision information based on the trajectory prediction uncertainty and the decision model, providing more accurate decision information for autonomous vehicles. Compared with existing technologies, this method modifies existing decision models based on trajectory prediction results based on the uncertainty of trajectory prediction. When the prediction model may fail, it exhibits higher trajectory prediction uncertainty to form a higher collision risk estimate with the conflicting object, thereby timely detecting and avoiding safety hazards caused by prediction failure.

[0080] In one embodiment, a decision model structure is provided, wherein the decision model includes: a collision constraint model, a road boundary model, a traffic light model, a control input limit constraint model, a reference path position prediction model, a vehicle state model, and a control variable model.

[0081] Among them, the collision constraint model is used to predict the collision between the vehicle and other traffic participants; the road boundary model is used to predict the collision between the vehicle and the road boundary; the traffic light model is used to predict the vehicle's violation; the control input limit constraint model is used to predict the vehicle's exceeding the preset control variables; the reference path position prediction model is used to predict the vehicle's position on the reference path at the next moment; the vehicle state model is used to predict the vehicle's driving state at the next moment; and the control variable model is used to calculate the vehicle's control variables at the current moment.

[0082] Among them, the following is adopted. Figure 3 The model shown describes the motion of an autonomous vehicle, such as... Figure 3 As shown, the model includes the following elements: vehicle center of gravity position (p x p y ) and its distance l from the front and rear axles f and l r Vehicle yaw angle φ, center of gravity sideslip angle β, front wheel steering angle δ f Vehicle speed v and acceleration a, etc.

[0083] The following sections will introduce the collision constraint model, road boundary model, traffic light model, control input limit constraint model, reference path position prediction model, vehicle state model, and control variable model respectively:

[0084] Example 1: The collision constraint model can be constructed based on the following method: the collision constraint model is constructed according to the driving angle, distance parameters, trajectory parameters and uncertainty parameters of the traffic participants.

[0085] Among them, the driving angle of the traffic participant refers to the angle between the traffic participant's orientation and the horizontal direction; the distance parameter represents the projection parameter of the relative distance between the center position of the vehicle and the center position of the traffic participant on the coordinate axis, including: the projection parameter of the relative distance between the center position of the vehicle and the center position of the traffic participant on the horizontal coordinate axis (x-axis), and the projection parameter of the relative distance between the center position of the vehicle and the center position of the traffic participant on the vertical coordinate axis (y-axis); the trajectory parameter refers to the position coordinate parameter of each point on the target predicted trajectory; the uncertainty parameter refers to the uncertainty parameter of the position coordinate of each point on the target predicted trajectory.

[0086] For example, the collision constraint model constructed above based on the driving angle, distance parameters, trajectory parameters, and uncertainty parameters of traffic participants can be expressed by the following formula (1):

[0087]

[0088] in, This represents the constraint conditions of the collision constraint model. This indicates that no collision occurred. This indicates that a collision has occurred. This represents the basic state of the vehicle at time t, including the vehicle's specific position coordinates and driving angle at time t. The driving angle refers to the angle between the vehicle's orientation and the horizontal direction. This represents the basic state of other traffic participant i at time t, including the specific location coordinates and driving angle of other traffic participant i at time t. The driving angle of other traffic participant i refers to the angle between the orientation of other traffic participant i and the horizontal direction.

[0089] in, Let c represent the center of the circle at time t. j The projection of the relative distance between the center of the ellipse representing traffic participant i and the center of the ellipse onto the x-axis, and Let c represent the center of the circle at time t. j The projection of the relative distance between the center of the ellipse representing traffic participant i and the center of the ellipse onto the y-coordinate axis, R(Φ i ) represents a rotation matrix, where the counterclockwise direction is the positive direction, Φ i This represents the driving angle of traffic participant i. To simplify the description of the collision constraint model, in this embodiment, n circles are used to approximate the space occupied by the vehicle, such as... Figure 4 As shown, using nc = 3 circles are used to approximate the space occupied by the vehicle. The black solid square in the figure represents the real vehicle. The length of the real vehicle is L and the width of the real vehicle is W. The three dashed circles in the figure represent the approximate vehicle. Then, the radius r of the approximate circle can be expressed by the following formula (2):

[0090]

[0091] The coordinates of the center of the approximate circle, c j It can be expressed by the following formula (3):

[0092]

[0093] Where p represents the center of the real vehicle, j represents the j-th circle, and c j Let φ represent the center of the j-th circle, and φ represent the vehicle's travel angle.

[0094] After describing the approximate graphic of the vehicle above, the approximate graphic of traffic participant i will be described below, as follows: Figure 5 As shown, an ellipse is used to approximate the space occupied by traffic participant i. The solid black box in the figure represents the actual traffic participant i vehicle, and the dashed ellipse represents the approximate traffic participant i, where a i Let b represent the semi-major axis of the ellipse that approximates the space occupied by traffic participant i. i Let represent the semi-minor axis of the ellipse that represents the space occupied by the approximate traffic participant i, where in the above formula (1) . This represents the semi-major axis of the ellipse representing the space occupied by traffic participant i, after considering uncertain parameters. Let represent the semi-minor axis of the ellipse representing the space occupied by traffic participant i after considering uncertain parameters, and and This can be expressed by the following formulas (4) and (5):

[0095]

[0096]

[0097] Among them, in the above formula Let represent the projection of the uncertainty parameter of traffic participant i at time t onto the x-axis. Let represent the projection of the uncertainty parameter of traffic participant i at time t onto the y-axis, η represent the sensitivity coefficient of the semi-major axis of the ellipse of the space occupied by traffic participant i after approximation, and μ represent the sensitivity coefficient of the semi-minor axis of the ellipse of the space occupied by traffic participant i after approximation.

[0098] Example 2: The road boundary model can be constructed based on the following method: the width of the left boundary of the road, the width of the right boundary of the road, and the vehicle's profile error.

[0099] The width of the left boundary of the road refers to the width between the inner and outer boundaries of the left road, and the width of the right boundary of the road refers to the width between the inner and outer boundaries of the right road. The vehicle profile error refers to the profile error of the vehicle's lateral deviation from the reference path at any given time. The width of the left boundary of the road and the width of the right boundary of the road can be the same or different. For example, the width of both the left boundary of the road and the width of the right boundary of the road can be 0.25m.

[0100] For example, the road boundary model constructed based on the width of the left boundary of the road, the width of the right boundary of the road, and the vehicle profile error can be expressed by the following formula (6):

[0101]

[0102] in, The width of the left boundary of the road. The width of the right boundary of the road. This represents the input state of the vehicle at time t. This represents the contour error corresponding to the vehicle's input state at time t.

[0103] Example 3: The control input limit constraint model can be constructed based on the following method: the control input limit constraint model is constructed based on the maximum parameter of the control variable, the minimum parameter of the control variable, and the control variables of the vehicle.

[0104] The maximum parameter of the control variable can be either the maximum value of the vehicle's acceleration or the maximum value of the vehicle's steering angle. The minimum parameter of the control variable can be either the minimum value of the vehicle's acceleration or the minimum value of the vehicle's steering angle. The vehicle's control variables can include the vehicle's acceleration and steering angle information at any given time. It should be noted that in this embodiment, the maximum value of acceleration is set to 2.8 m / s², the maximum value of steering angle is -4 m / s², the maximum value of steering angle is 0.22π, and the minimum value of steering angle is -0.22π.

[0105] For example, the control input limit constraint model constructed above based on the maximum parameter of the control variable, the minimum parameter of the control variable, and the control variables of the vehicle can be expressed by the following formula (7):

[0106]

[0107] Among them, U min U represents the smallest parameter of the control variable. max This represents the maximum parameter of the control variable. This represents the vehicle's control variables at time t, and the control variables include the vehicle's acceleration information and steering angle information.

[0108] Example 4: The reference path position prediction model can be constructed based on the following: the vehicle's position on the reference path, the vehicle's speed at any time, the vehicle's travel time, and the vehicle's position on the reference path at the next time.

[0109] Among them, the vehicle's position on the reference path indicates the vehicle's specific location on the reference path at the current moment, the vehicle's speed at any moment indicates the vehicle's speed on the travel path at any moment, the vehicle's travel time indicates the time taken for the vehicle to travel from its position at the previous moment to its position at the next moment, and the vehicle's position on the reference path at the next moment indicates the vehicle's specific location on the reference path at the next moment.

[0110] For example, the reference path position prediction model constructed above based on the vehicle's position on the reference path, the vehicle's speed at any time, the vehicle's travel time, and the vehicle's position on the reference path at the next time can be expressed by the following formula (8):

[0111] λ t+1 =λ t +v t Δ t (8);

[0112] Where, λ t+1 λ represents the vehicle's position on the reference path at the next moment. t This indicates the vehicle's current position on the reference path, v t Δ represents the vehicle's speed at the current moment. t This indicates the time required for the vehicle to move from its current position on the reference path to its next position on the reference path.

[0113] Example 5: The vehicle state model can be constructed based on the vehicle's input state at any time, the vehicle's control variables at any time, and the vehicle's driving state at the next time.

[0114] Here, the vehicle's input state at any given time represents the input state received by the autonomous vehicle from the decision-making system at any given time; the vehicle's control variables at any given time represent the control variables received by the autonomous vehicle from the decision-making system at any given time; and the vehicle's driving state at the next moment represents the input state received by the autonomous vehicle from the decision-making system at the next moment.

[0115] For example, the vehicle state model constructed based on the vehicle's input state at any time, the vehicle's control variables at any time, and the vehicle's driving state at the next time can be expressed by the following formula (9):

[0116]

[0117] in, This indicates the vehicle's driving status at the next moment. This represents the input state of the vehicle at time t. This represents the vehicle's control variables at time t, including the vehicle's acceleration and steering angle.

[0118] Example 6: The control variable model can be constructed based on the vehicle's input state at any time, the vehicle's control variables at any time, and the vehicle's position on the reference path point.

[0119] Among them, the vehicle's input state at any time represents the input state received by the autonomous vehicle from the decision system at any time, the vehicle's control variables at any time represent the control variables received by the autonomous vehicle from the decision system at any time, and the vehicle's position on the reference path point represents the vehicle's current position on the reference path.

[0120] For example, the control variable model constructed based on the vehicle's input state at any time, the vehicle's control variables at any time, and the vehicle's position on the reference path point can be expressed by the following formula (10):

[0121]

[0122] in, Represents the vehicle's control variables. This represents the input state of the vehicle at time t. Let λ represent the vehicle's control variables at time t, including the vehicle's acceleration and steering angle. t This indicates the vehicle's current position on the reference path.

[0123] Among them, in the above formula (10) It can also be expressed by the following formula (11):

[0124]

[0125] in, Indicates hysteresis error. The profile error represents the lateral and longitudinal deviations between the vehicle's trajectory point at time t and the corresponding position on the reference path, respectively. ref Indicates the reference speed, at and δ ft Let Q represent the vehicle's acceleration and front wheel steering angle at time t, respectively. c Q l Q v Q a and Q δ The values ​​represent weighting factors, which are set to 130, 130, 4, 6, and 100 in this embodiment.

[0126] In addition, in this embodiment of the application, the decision model also includes a traffic light model. For example, the traffic light model can be represented by the following formula (12):

[0127] tl_cons(s t ,s stop ,traffic_light) (12);

[0128] Among them, s t and s stop These represent the positions of the vehicle and the stop line, respectively. traffic_light indicates the traffic light status; when the light is red, the vehicle cannot move beyond the stop line.

[0129] The method for determining the decision model provided in this application improves the reliability of the decision model and the safety of the system by introducing uncertainty to construct a collision constraint model and determining the decision model based on the environmental relationship between vehicles and traffic participants.

[0130] In one embodiment, the process of processing each candidate trajectory to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory can also be described, i.e., S202 above, "processing each candidate trajectory to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory", such as Figure 6 As shown, it includes:

[0131] S301. Calculate the average coordinate position of all candidate trajectories at the same time to obtain the average coordinate position at the same time, and determine the target predicted trajectory based on the average coordinate position at all times.

[0132] In this embodiment of the application, after obtaining multiple candidate trajectories in S201, the coordinate positions corresponding to the same time on all candidate trajectories can be averaged to obtain the average coordinate position at the same time, as well as the average coordinate position at each time on each candidate trajectory, and the target predicted trajectory can be determined based on the average coordinate position at all times.

[0133] For example, after trajectory prediction using multiple isomorphic and heterogeneous prediction models, the m-th candidate trajectory can be represented by the following formula (13):

[0134] y m ~P(y|x,W m (13);

[0135] Where x represents the input data of the prediction model, W m Let y represent the parameters of the m-th trajectory prediction network, and y be the trajectory of the traffic participant being predicted. m This represents the prediction result of the m-th prediction network on the future trajectories of traffic participants based on the input information, which includes the predicted future trajectory T of traffic participant i. f Location information at each moment, and the predicted future traffic participant i. f The location information at each moment can be represented by the following formula (14):

[0136]

[0137] in, Indicates the predicted traffic participant i in the future T f Location information at any given time Indicates the predicted traffic participant i in the future T f The horizontal coordinate information at each moment, Indicates the predicted traffic participant i in the future T f The ordinate information at each time point.

[0138] Furthermore, after obtaining the coordinate positions on different candidate trajectories at the same time, the coordinate positions on different candidate trajectories at the same time are averaged to obtain the average coordinate position at that time. The average coordinate position at time t can be expressed by the following formula (15):

[0139]

[0140] in, This represents the average coordinate position at time t. This represents the average x-coordinate position at time t. This represents the average ordinate position at time t. and They can be expressed by the following formulas (16) and (17):

[0141]

[0142]

[0143] in, Indicates the predicted traffic participant i in the future T f The horizontal coordinate information at each moment, Indicates the predicted traffic participant i in the future T fThe vertical coordinate information at each time point, where M represents the total number of isomorphic and heterogeneous prediction models.

[0144] Furthermore, the average coordinate position at each time step on each candidate trajectory is calculated, and the target predicted trajectory is determined based on the average coordinate position at all times.

[0145] S302. Calculate the variance of the coordinate positions at the same time on all candidate trajectories to obtain the variance coordinate positions at the same time, and determine the uncertainty of the target predicted trajectory based on the variance coordinate positions at all times.

[0146] In this embodiment of the application, after obtaining multiple candidate trajectories in S201, the variance of the coordinate positions corresponding to the same time on all candidate trajectories can be calculated to obtain the variance coordinate positions at the same time, as well as the variance coordinate positions at each time on each candidate trajectory, and the uncertainty of the target predicted trajectory can be determined based on the variance coordinate positions at all times. For example, after obtaining the coordinate positions on different candidate trajectories at the same time, the variance of the coordinate positions on different candidate trajectories at the same time is calculated to obtain the variance coordinate positions at that time, wherein the variance of the prediction result at time t can be expressed by the following formula (18):

[0147]

[0148] in, This represents the average of all network predictions. Indicates the predicted traffic participant i in the future T f Location information at any given time.

[0149] Furthermore, the variance coordinates of each candidate trajectory at each time step are calculated, and the uncertainty of the target predicted trajectory is determined based on the variance coordinates of all time steps.

[0150] The target prediction trajectory and the method for determining the uncertainty of the target prediction trajectory provided in the embodiments of this application lay the foundation for the decision model determined based on the uncertainty of the target prediction trajectory, and are an indispensable step of this application.

[0151] In one embodiment, in the above Figure 2 Based on the illustrated embodiments, as Figure 7 As shown, the above method includes:

[0152] S204. Obtain sample road environment data.

[0153] The sample road environment data includes sample traffic participant status data, sample traffic light status data, sample map status data, and traffic participant trajectory data.

[0154] In this embodiment of the application, before inputting the road environment data of the vehicle into the isomorphic heterogeneous prediction model for trajectory prediction in step S201, it is necessary to train the initial prediction model to obtain the isomorphic heterogeneous prediction model. During the training of the initial prediction model, sample road environment data at a certain moment can be obtained from the environment database of the computer device. It should be noted that the environment database can be pre-stored in the computer device to store road environment data at various times.

[0155] S205. Based on road environment data, train the initial prediction model with different parameters for the same structure to obtain the prediction model with different parameters.

[0156] Each initial prediction model includes a history encoder, an interaction feature encoder, a map encoder, and a trajectory decoder. The interaction feature encoder uses a graph neural network design, and the map encoder uses a convolutional neural network design. The history encoder uses a network based on gated recurrent units to process the historical states of traffic participants. The interaction feature encoder between traffic participants and traffic light states uses a network based on directed edge feature heterogeneous graphs to handle the interaction states between traffic participants and traffic light states. The map encoder uses an adaptive map selection method. The trajectory decoder is constructed using a long short-term memory network.

[0157] The following describes the training methods for any isomorphic heterogeneous prediction model, including:

[0158] Step 1: Randomly initialize the parameters of the initial prediction model; for example, design a random number seed to randomly initialize the parameters of the initial model; among them, for typical urban signalized intersection scenarios, a trajectory prediction model that considers the historical status of traffic participants, traffic light status, and scene map information can be constructed.

[0159] Step 2: Randomly shuffle the sample road environment data obtained in step S204 above, and use the randomly shuffled sample road environment data to train the initial prediction model until the training loss converges, and obtain a trained prediction model.

[0160] Step 3: Repeat Step 1 and Step 2 M times to obtain M isomorphic trajectory prediction models with different parameters. It should be noted that the parameters of the initial prediction model are different each time.

[0161] In one embodiment, such as Figure 8As shown, a process is provided to obtain a predicted trajectory based on a trained prediction model (i.e., the trajectory prediction model in the figure), and to obtain different predicted trajectories based on different trained isomorphic heterogeneous prediction models (i.e., the trajectory prediction models in the figure). The process is to perform calculations based on the different predicted trajectories to obtain the target predicted trajectory (i.e., the final predicted trajectory in the figure) and the uncertainty of the target predicted trajectory (i.e., the uncertainty in the figure). Finally, a decision model (i.e., the safety decision model sensitive to the risk of prediction failure in the figure) is constructed based on the target predicted trajectory and the uncertainty of the target predicted trajectory. It should be noted that the input to the trained prediction model is the state information of traffic participants at a certain moment, the state information of traffic lights, and the map information at a certain moment. After this information is input into the trained prediction model, the historical state encoding, interaction feature encoding, map information encoding, and decoder in the trained prediction model process this information to obtain the predicted trajectory. Furthermore, different predicted trajectories are obtained based on different trained isomorphic predictive models (i.e., the trajectory prediction model in the figure), and the target predicted trajectory (i.e., the final predicted trajectory in the figure) and the uncertainty of the target predicted trajectory (i.e., the uncertainty in the figure) are calculated based on the different predicted trajectories. A decision model is then constructed based on the target predicted trajectory and the uncertainty of the target predicted trajectory, and the vehicle's decision information is obtained based on this decision model.

[0162] The training method for the prediction model provided in this application, based on different initialization parameters of the prediction model, yields multiple isomorphic and heterogeneous prediction models, laying the foundation for obtaining multiple prediction trajectories based on multiple isomorphic and heterogeneous prediction models, and is an indispensable part of this application.

[0163] In one embodiment, such as Figure 9 As shown, a complete method for determining autonomous driving decision information is also provided, including:

[0164] S10. Obtain sample road environment data;

[0165] S11. Based on road environment data, train the initial prediction model with different parameters for the same structure to obtain the prediction model with different parameters.

[0166] S12. Input the road environment data where the vehicle is located into the isomorphic heterogeneous prediction model to predict the trajectory and obtain multiple candidate trajectories;

[0167] S13. Calculate the average coordinate position of all candidate trajectories at the same time to obtain the average coordinate position at the same time, and determine the target predicted trajectory based on the average coordinate position at all times.

[0168] S14. Calculate the variance of the coordinate positions at the same time on all candidate trajectories to obtain the variance coordinate positions at the same time, and determine the uncertainty of the target predicted trajectory based on the variance coordinate positions at all times.

[0169] S15. Input the target predicted trajectory and the uncertainty of the target predicted trajectory into the decision model to obtain decision information.

[0170] The method for determining autonomous driving decision information provided in this application first inputs the road environment data of the vehicle into a homogeneous but heterogeneous prediction model for trajectory prediction, obtaining multiple candidate trajectories. Then, each candidate trajectory is processed to obtain the target predicted trajectory and its uncertainty. Finally, the target predicted trajectory and its uncertainty are input into a decision model to obtain decision information. This method considers the uncertainty of trajectory prediction in the prediction model and obtains the final decision information based on the trajectory prediction uncertainty and the decision model, providing more accurate decision information for autonomous vehicles. Compared with existing technologies, this method modifies existing decision models based on trajectory prediction results based on the uncertainty of trajectory prediction. When the prediction model may fail, it exhibits higher trajectory prediction uncertainty to form a higher collision risk estimate with the conflicting object, thereby timely detecting and avoiding safety hazards caused by prediction failure.

[0171] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0172] Based on the same inventive concept, this application also provides an apparatus for determining autonomous driving decision information to implement the method for determining autonomous driving decision information described above. The solution provided by this apparatus is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the decision information determination apparatus provided below can be found in the limitations of the decision information determination method described above, and will not be repeated here.

[0173] In one exemplary embodiment, such as Figure 10As shown, an apparatus for determining autonomous driving decision information is provided, comprising: a prediction module 10, a processing module 11, and an acquisition module 12, wherein:

[0174] The prediction module 10 is used to input the road environment data where the vehicle is located into the isomorphic but heterogeneous prediction model to predict the trajectory and obtain multiple candidate trajectories.

[0175] The processing module 11 is used to process each candidate trajectory to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory.

[0176] The acquisition module 12 is used to input the target predicted trajectory and the uncertainty of the target predicted trajectory into the decision model to obtain decision information.

[0177] In an exemplary embodiment, the decision model includes: a collision constraint model, a road boundary model, a traffic light model, a control input limit constraint model, a reference path position prediction model, a vehicle state model, and a control variable model. The collision constraint model is used to predict collisions between the vehicle and other traffic participants; the road boundary model is used to predict collisions between the vehicle and the road boundary; the traffic light model is used to predict violations by the vehicle; the control input limit constraint model is used to predict when the vehicle exceeds a preset control variable; the reference path position prediction model is used to predict the vehicle's position on the reference path at the next moment; the vehicle state model is used to predict the vehicle's driving state at the next moment; and the control variable model is used to calculate the vehicle's control variables at the current moment.

[0178] In an exemplary embodiment, the above-described apparatus further includes: a first construction unit, specifically configured to construct a collision constraint model based on the participant's driving angle, distance parameters, trajectory parameters, and uncertainty parameters; the distance parameters represent the projection parameters of the relative distance between the center position of the vehicle and the center position of the participant onto the coordinate axes.

[0179] In an exemplary embodiment, the above-described apparatus further includes a second construction unit, specifically configured to construct a road boundary model based on the width of the left boundary of the road, the width of the right boundary of the road, and the vehicle profile error.

[0180] In an exemplary embodiment, the above-described apparatus further includes a third construction unit, specifically configured to construct a control input limit constraint model based on the maximum parameter of the control variable, the minimum parameter of the control variable, and the control variables of the vehicle.

[0181] In an exemplary embodiment, the above-described apparatus further includes a fourth construction unit, specifically configured to construct a reference path position prediction model based on the vehicle's position on the reference path, the vehicle's speed at any given time, the vehicle's travel time, and the vehicle's position on the reference path at the next moment.

[0182] In an exemplary embodiment, the above-described apparatus further includes a fifth construction unit, specifically configured to construct a vehicle state model based on the vehicle's input state at any given time, the vehicle's control variables at any given time, and the vehicle's driving state at the next time.

[0183] In an exemplary embodiment, the above-described apparatus further includes a sixth construction unit, specifically configured to construct a control variable model based on the vehicle's input state at any given time, the vehicle's control variables at any given time, and the vehicle's position on a reference path point.

[0184] In an exemplary embodiment, the processing module 11 includes: a first computing unit and a second computing unit, wherein:

[0185] The first calculation unit is specifically used to calculate the average coordinate position of all candidate trajectories at the same time, obtain the average coordinate position at the same time, and determine the target predicted trajectory based on the average coordinate position at all times.

[0186] The second calculation unit is specifically used to calculate the variance of the coordinate positions corresponding to the same time on all candidate trajectories, obtain the variance coordinate positions at the same time, and determine the uncertainty of the target predicted trajectory based on the variance coordinate positions at all times.

[0187] In an exemplary embodiment, the above apparatus further includes: an acquisition module and a training module, wherein:

[0188] The acquisition module is used to acquire sample road environment data; the sample road environment data includes sample traffic participant status data, sample traffic light status data, sample map status data, and traffic participant trajectory data;

[0189] The training module is used to train the initial prediction models with different parameters based on road environment data to obtain the prediction models with different parameters. Each initial prediction model includes a history encoder, an interaction feature encoder, a map encoder, and a trajectory decoder.

[0190] Each module in the aforementioned autonomous driving decision-making information determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0191] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 1As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores sample road environment data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a method for determining autonomous driving decision information.

[0192] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0193] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0194] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0195] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0196] 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0197] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0198] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0199] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for determining decision information for autonomous driving, characterized in that, The method includes: The road environment data where the vehicle is located is input into the isomorphic heterogeneous prediction model to predict the trajectory and obtain multiple candidate trajectories of other traffic participants. Each of the candidate trajectories is processed to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory; The predicted target trajectory and its uncertainty are input into the decision model to obtain decision information; The decision-making model includes: collision constraint model, road boundary model, traffic light model, control input limit constraint model, reference path position prediction model, vehicle state model, and control variable model; The collision constraint model is constructed based on the driving angle, distance parameters, trajectory parameters, and uncertainty parameters of the traffic participants, as shown in the following formula (1): in, The constraints of the collision constraint model are represented. This indicates that no collision occurred. This indicates that a collision has occurred. This represents the basic state of the vehicle at time t, including the vehicle's specific position coordinates and driving angle at time t. The driving angle refers to the angle between the vehicle's orientation and the horizontal direction. This represents the basic state of other traffic participant i at time t, including the specific location coordinates and driving angle of other traffic participant i at time t. The driving angle of other traffic participant i refers to the angle between the orientation of other traffic participant i and the horizontal direction. This indicates that the center of the circle representing the vehicle is at time t. The projection of the relative distance between the center of the ellipse representing traffic participant i and the center of the ellipse onto the x-axis, and This indicates that the center of the circle representing the vehicle is at time t. The projection of the relative distance between the center of the ellipse representing traffic participant i and the center of the ellipse on the y-coordinate axis Let represent a rotation matrix, where the counterclockwise direction is the positive direction. Indicates the driving angle of traffic participant i; This represents the semi-major axis of the ellipse representing the space occupied by traffic participant i, after considering uncertain parameters. Let represent the semi-minor axis of the ellipse representing the space occupied by traffic participant i after considering uncertain parameters, and and This can be expressed by the following formulas (2) and (3): Among them, in the above formula (2) Let represent the projection of the uncertainty parameter of traffic participant i at time t onto the x-axis. The sensitivity coefficient represents the semi-major axis of the ellipse representing the approximate occupied space of traffic participant i. To simplify the description of the collision constraint model, n circles are used to approximate the occupied space of the vehicle. The semi-major axis of the ellipse representing the space occupied by traffic participant i after approximation; In the above formula (3) Let represent the projection of the uncertainty parameter of traffic participant i at time t onto the y-axis. The sensitivity coefficient represents the semi-minor axis of the ellipse representing the space occupied by traffic participant i after approximation. Let denot be the semi-minor axis of the ellipse representing the space occupied by traffic participant i after approximation.

2. The method according to claim 1, characterized in that, The collision constraint model is used to predict collisions between the vehicle and other traffic participants; the road boundary model is used to predict collisions between the vehicle and the road boundary; the traffic light model is used to predict violations by the vehicle; the control input limit constraint model is used to predict when the vehicle exceeds preset control variables; the reference path position prediction model is used to predict the vehicle's position on the reference path at the next moment; the vehicle state model is used to predict the vehicle's driving state at the next moment; and the control variable model is used to calculate the vehicle's control variables at the current moment.

3. The method according to claim 1 or 2, characterized in that, The method further includes: The road boundary model is constructed based on the width of the left boundary of the road, the width of the right boundary of the road, and the vehicle profile error.

4. The method according to claim 1 or 2, characterized in that, The method further includes: The control input limit constraint model is constructed based on the maximum parameter of the control variable, the minimum parameter of the control variable, and the control variables of the vehicle.

5. The method according to claim 1 or 2, characterized in that, The method further includes: The reference path position prediction model is constructed based on the vehicle's position on the reference path, the vehicle's speed at any given time, the vehicle's travel time, and the vehicle's position on the reference path at the next moment.

6. The method according to claim 1 or 2, characterized in that, The method further includes: The vehicle state model is constructed based on the vehicle's input state at any given time, the vehicle's control variables at any given time, and the vehicle's driving state at the next time.

7. The method according to claim 1 or 2, characterized in that, The method further includes: The control variable model is constructed based on the vehicle's input state at any time, the vehicle's control variables at any time, and the vehicle's position on the reference path point.

8. The method according to claim 1, characterized in that, The process of processing each of the candidate trajectories to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory includes: The average coordinate position at the same time is calculated by averaging the coordinate positions at all candidate trajectories, and the target predicted trajectory is determined based on the average coordinate positions at all times. The variance of the coordinate positions at the same time on all candidate trajectories is calculated to obtain the variance coordinate positions at the same time, and the uncertainty of the target predicted trajectory is determined based on the variance coordinate positions at all times.

9. The method according to claim 1, characterized in that, The method further includes: Acquire sample road environment data; the sample road environment data includes sample traffic participant status data, sample traffic light status data, sample map status data, and traffic participant trajectory data; Based on the road environment data, the initial prediction models with different parameters are trained to obtain the prediction models with different parameters; each initial prediction model includes a history encoder, an interaction feature encoder, a map encoder, and a trajectory decoder.

10. A device for determining autonomous driving decision information, characterized in that, The apparatus, applied to the method for determining autonomous driving decision information as described in any one of claims 1-9, comprises: The prediction module is used to input the road environment data where the vehicle is located into the isomorphic but heterogeneous prediction model to predict the trajectory and obtain multiple candidate trajectories of other traffic participants. The processing module is used to process each of the candidate trajectories to obtain the target predicted trajectory and the uncertainty of the target predicted trajectory; The acquisition module is used to input the target predicted trajectory and the uncertainty of the target predicted trajectory into the decision model to obtain autonomous driving decision information.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

Images