CN115221722A

CN115221722A - Simulation test method, model training method and device for automatic driving vehicle

Info

Publication number: CN115221722A
Application number: CN202210895515.7A
Authority: CN
Inventors: 卢帅; 王成法; 韩仲卿
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-07-28
Filing date: 2022-07-28
Publication date: 2022-10-21
Anticipated expiration: 2042-07-28
Also published as: CN115221722B

Abstract

The disclosure provides a simulation test method of an automatic driving vehicle, which relates to the technical field of artificial intelligence, in particular to the technical field of automatic driving, driving assistance and deep learning. The specific implementation scheme is as follows: in response to determining that a preset interaction behavior exists between the vehicle to be tested and the obstacle, determining a target deep learning model according to the type of the preset interaction behavior, wherein the obstacle moves in the simulation scene based on the preset state information, and the vehicle to be tested moves in the simulation scene based on the automatic driving mode; processing the environmental information and the preset state information of the obstacle by using a target deep learning model to obtain target state information; and controlling the movement of the obstacle using the target state information. The disclosure also provides a training method and device of the deep learning model, electronic equipment and a storage medium.

Description

Simulation test method, model training method and device for automatic driving vehicle

Technical Field

The present disclosure relates to the field of artificial intelligence technology, and more particularly, to the field of automated driving, assisted driving, and deep learning techniques. More specifically, the present disclosure provides a simulation test method of an autonomous vehicle, a training method of a deep learning model, an apparatus, an electronic device, and a storage medium.

Background

With the development of artificial intelligence technology, the application scenarios of automatic driving and assisted driving are increasing. In a simulation scenario, the validity or accuracy of the autonomous driving mode may be verified.

Disclosure of Invention

The disclosure provides a simulation test method of an autonomous vehicle, a training method of a deep learning model, a method, a device, equipment and a storage medium.

According to an aspect of the present disclosure, there is provided a simulation test method of an autonomous vehicle, the method including: in response to determining that a preset interaction behavior exists between the vehicle to be tested and the obstacle, determining a target deep learning model according to the type of the preset interaction behavior, wherein the obstacle moves in the simulation scene based on the preset state information, and the vehicle to be tested moves in the simulation scene based on the automatic driving mode; processing the environmental information and the preset state information of the obstacle by using a target deep learning model to obtain target state information; and controlling the movement of the obstacle using the target state information.

According to another aspect of the present disclosure, there is provided a training method of a deep learning model, the method including: inputting first sample sub-data of sample data into an initial deep learning model to obtain output sample state information, wherein the sample data is environment information and state information of an obstacle in a sample period, and the sample data is from road test sub-data corresponding to the type of a preset interaction behavior; determining a loss value according to the output sample state information and second sample subdata of the sample data; and adjusting parameters of the initial deep learning model by using the loss value to obtain a target deep learning model, wherein the target deep learning model corresponds to the type of a preset interaction behavior.

According to another aspect of the present disclosure, there is provided a simulation test apparatus of an autonomous vehicle, the apparatus including: the system comprises a first determining module, a target deep learning module and a target deep learning module, wherein the first determining module is used for determining a preset interactive behavior between a vehicle to be tested and an obstacle according to the type of the preset interactive behavior, the obstacle moves in a simulation scene based on preset state information, and the vehicle to be tested moves in the simulation scene based on an automatic driving mode; the processing module is used for processing the environmental information and the preset state information of the barrier by using the target deep learning model to obtain target state information; and a control module for controlling the movement of the obstacle using the target state information.

According to another aspect of the present disclosure, there is provided a training apparatus for a deep learning model, the apparatus including: the acquisition module is used for inputting first sample sub-data of sample data into the initial deep learning model to obtain output sample state information, wherein the sample data is environment information and state information of an obstacle in a sample time period, and the sample data is road test sub-data corresponding to the type of a preset interactive behavior; a fifth determining module, configured to determine a loss value according to the output sample state information and second sample sub-data of the sample data; and the adjusting module is used for adjusting parameters of the initial deep learning model by using the loss value to obtain a target deep learning model, wherein the target deep learning model corresponds to the type of a preset interaction behavior.

According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method provided in accordance with the present disclosure.

According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing a computer to perform a method provided according to the present disclosure.

According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program which, when executed by a processor, implements a method provided according to the present disclosure.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a schematic diagram of an exemplary system architecture to which the method and apparatus for simulation testing of autonomous vehicles may be applied, according to one embodiment of the present disclosure;

FIG. 2 is a flow chart of a method of simulation testing of an autonomous vehicle according to one embodiment of the present disclosure;

fig. 3A-3C are schematic diagrams of a simulation testing method of an autonomous vehicle according to one embodiment of the present disclosure.

FIG. 4 is a flow chart of a method of simulation testing of an autonomous vehicle according to another embodiment of the disclosure;

FIG. 5 is a schematic diagram of a method of simulated testing of an autonomous vehicle according to another embodiment of the present disclosure;

FIG. 6 is a flow diagram of a method of training a deep learning model according to one embodiment of the present disclosure;

FIG. 7 is a schematic diagram of a training method of a deep learning model according to one embodiment of the present disclosure;

FIG. 8 is a block diagram of a simulation test setup of an autonomous vehicle according to one embodiment of the present disclosure;

FIG. 9 is a block diagram of a training apparatus for deep learning models according to one embodiment of the present disclosure; and

FIG. 10 is a block diagram of an electronic device to which a simulation testing method and/or a training method of deep learning models of an autonomous vehicle may be applied, according to one embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The automatic driving simulation technology is a means for testing an automatic driving algorithm. The automatic driving simulation technology has the advantages of low cost and high test efficiency, and is very important for the iterative development of an automatic driving algorithm. One problem faced by the automated driving simulation technique is: the authenticity is not sufficient. For example, during an autonomous driving simulation, obstacles may not form effective, complex interactions with the autonomous vehicle, resulting in insufficient confidence in the autonomous driving simulation. For example, in the automatic driving simulation process, the behavior of the obstacle is fixed, the structure of the simulation scene is simple, and the testing capability is insufficient. For another example, in the process of automatic driving simulation, it is difficult for obstacles to reasonably react to the behavior of the automatic driving vehicle, so that a large number of interactive behaviors which rarely occur or are unreasonable in a real scene occur, and an error occurs in a test result.

In some embodiments, a simulation scenario may be established based on the drive test data. In this simulation scenario, the velocity, position and shape of the obstacle are determined. The reappearance of the scene where the drive test vehicle is located can be realized. In this simulation scenario, the behavior of the obstacle is fixed, which may reflect the interaction between the obstacle and the drive test vehicle. In the simulation scene, the obstacle is difficult to reasonably react according to the behavior change of the automatic driving vehicle, and further a large amount of interaction behaviors which rarely or unreasonably appear in a real scene occur, so that the test result has errors.

In some embodiments, simulation scenarios may also be designed manually. The scene designed based on manual experience can be used for testing a specific target. In this simulation scenario, the behavior of the obstacle is also fixed. For example, a particular action may be triggered in case a particular condition is met. The expression mode of the simulation scene can be a parameterized expression mode. The structure and behavior of the simulation scene is relatively simple. In the simulation scene, the behavior of the barrier is single, the diversity is poor, and other test targets are difficult to realize.

FIG. 1 is a schematic diagram of an exemplary system architecture to which the method and apparatus for simulation testing of autonomous vehicles may be applied, according to one embodiment of the present disclosure.

It should be noted that fig. 1 is only an example of a system architecture to which the embodiments of the present disclosure may be applied to help those skilled in the art understand the technical content of the present disclosure, and does not mean that the embodiments of the present disclosure may not be applied to other devices, systems, environments or scenarios.

As shown in fig. 1, a system architecture 100 according to this embodiment may include

sensors

101, 102, 103, a network 120, a server 130, and a Road Side Unit (RSU) 140. Network 120 is used to provide a medium for communication links between

sensors

101, 102, 103 and server 130. Network 120 may include various connection types, such as wired and/or wireless communication links, and so forth.

The

sensors

101, 102, 103 may interact with the server 130 over the network 120 to receive or send messages or the like.

The

sensors

101, 102, 103 may be functional elements integrated on the vehicle 110, such as infrared sensors, ultrasonic sensors, millimeter-wave radar, information acquisition devices, and the like. The

sensors

101, 102, 103 may be used to collect status data of a perception object (e.g., a pedestrian, a vehicle, an obstacle, etc.) around the vehicle 110 as well as surrounding road data.

The vehicle 110 may communicate with the roadside unit 140, receive information from the roadside unit 140, or transmit information to the roadside unit.

The roadside unit 140 may be disposed on a signal light, for example, so as to adjust the duration or frequency of the signal light.

The server 130 may be disposed at a remote end capable of establishing communication with the vehicle-mounted terminal, and may be implemented as a distributed server cluster formed by a plurality of servers, or may be implemented as a single server.

The server 130 may be a server that provides various services. For example, a map application, a data processing application, and the like may be installed on the server 130. Taking the server 130 running the data processing application as an example: the state data of the obstacle and the road data transmitted from the

sensors

101, 102, 103 are received via the network 120. One or more of the state data of the obstacle and the road data may be used as the data to be processed. And processing the data to be processed to obtain target data.

It should be noted that the simulation test method for the autonomous vehicle provided in the embodiment of the present disclosure may be generally executed by the server 130. Accordingly, the simulation test device for the autonomous vehicle provided by the embodiment of the present disclosure may also be disposed in the server 130.

It is understood that the number of sensors, networks, and servers in fig. 1 is merely illustrative. There may be any number of sensors, networks, and servers, as desired for implementation.

It should be noted that the sequence numbers of the respective operations in the following methods are merely used as a representation of the operations for description, and should not be construed as representing the execution order of the respective operations. The method need not be performed in the exact order shown, unless explicitly stated.

FIG. 2 is a flow chart of a method of simulation testing of an autonomous vehicle according to one embodiment of the present disclosure.

As shown in fig. 2, the method 200 may include operations S210 to S230.

In operation S210, in response to determining that a preset interactive behavior exists between the vehicle to be tested and the obstacle, a target deep learning model is determined according to a type of the preset interactive behavior.

In the disclosed embodiments, the obstacle is moved in the simulation scene based on the preset state information.

For example, the obstacle may be various objects on a road such as a pedestrian, a motor vehicle, a non-motor vehicle, or the like.

In an embodiment of the present disclosure, a vehicle under test is moved in a simulation scenario based on an autonomous driving mode.

For example, the vehicle under test may be an autonomous vehicle as described above.

In the embodiment of the present disclosure, each type of preset interaction behavior corresponds to one target deep learning model.

For example, there may be multiple types of preset interaction behavior, and correspondingly, there may be multiple target deep learning models.

For example, the plurality of deep learning models may include a Long Short-Term Memory (LSTM) model, a vector network (VectorNet) model, and so on.

In the embodiment of the present disclosure, the preset interaction behavior may be: and in the process that the vehicle to be tested executes the automatic driving behavior, the obstacle executes the preset behavior.

For example, one preset interaction behavior may be: and under the condition that the vehicle to be detected moves straight, the obstacle makes a turning behavior.

In operation S220, the environment information and the preset state information of the obstacle are processed using the target deep learning model to obtain target state information.

In the embodiment of the present disclosure, the environment information and the preset state information of the obstacle may be input into the target deep learning model to obtain the target state information.

For example, the environment information may include lane line information, traffic indication information. The traffic indicating information may be information of a traffic light.

For example, the preset state information may be obtained from drive test data. The drive test data may be collected by the drive test vehicle in a real scene.

In operation S230, the obstacle movement is controlled using the target state information.

For example, the obstacle may be controlled to switch to the state indicated by the target state information.

According to the method and the device, the state of the barrier is determined by utilizing the deep learning model corresponding to the type of the preset interaction behavior, the intelligent level of the barrier is improved, and the rationality of the simulation test is ensured.

In some embodiments, the method 200 described above may further include: and determining the test result of the vehicle to be tested according to the vehicle to be tested and the obstacle controlled by the target behavior information.

For example, the end position of the vehicle under test in the simulation scenario may be set. In the process that the vehicle to be detected moves to the end point position in the simulation scene, if a preset interaction behavior is generated between the obstacle and the vehicle to be detected, the obstacle can be controlled to move by using the target state information. After the vehicle to be tested moves to the end position, the test result of the vehicle to be tested can be determined. It can be understood that the obstacle is adjusted, the behavior of the obstacle is not completely fixed any more, and the interaction behavior generated between the obstacle and the vehicle to be tested is basically a reasonable interaction behavior, so that the test result of the vehicle to be tested is more reasonable and accurate. In the embodiment of the present disclosure, the autopilot controller of the vehicle to be tested may output various instructions, track information, position information, speed information, and other various information as the test result.

In some embodiments, the type of preset interaction behavior comprises at least one of: under the condition that the vehicle to be detected moves straight, the obstacle turns; under the condition that the vehicle to be detected turns, the barrier moves straight; under the condition that the vehicle to be detected moves straight, the barrier enters a lane where the vehicle to be detected is located; and the obstacle follows the vehicle to be detected.

For example, in the case where the vehicle under test is traveling straight, the obstacle steering may include: in the simulation scenario, lane 1 is to the left of lane 2. The vehicle to be tested moves on the lane 1, the obstacle moves on the lane 2, and the obstacle turns left to enter the lane 1 and moves behind the vehicle to be tested.

For example, in the case where the vehicle under test is turning, the obstacle straight-ahead includes: in the simulation scenario, lane 1 is to the left of lane 2. The vehicle to be measured moves on the lane 2, and the obstacle moves on the lane 1. When the vehicle to be tested turns left and enters the lane 1, the barrier continues to go straight on the lane 1 and moves behind the vehicle to be tested.

For example, in the case where the vehicle to be measured travels straight, the obstacle entering the lane in which the vehicle to be measured is located includes: in the simulation scenario, lane 1 is to the left of lane 2. The vehicle to be tested moves on the lane 1, the obstacle moves on the lane 2, the obstacle turns left to enter the lane 1 and moves in front of the vehicle to be tested.

For example, the obstacle following the vehicle under test includes: the track heights of the obstacles and the vehicle to be detected are consistent. In the simulation scenario, lane 1 is to the left of lane 2. The vehicle to be measured moves on the lane 2, and the obstacle moves on the lane 2. After the vehicle to be tested turns left and enters the lane 1, the obstacle also turns left and enters the lane 1.

In some embodiments, the type of preset interactive behavior further comprises other preset interactive behaviors.

For example, the other preset interaction behavior may be other abnormal interaction behavior.

For another example, for other preset interactive behaviors, a generalized target deep learning model may be used to process the environmental information and the preset state information of the relevant obstacle, so as to obtain the target state information.

In some embodiments, the preset state information of the obstacle includes at least one of: position information of the obstacle, velocity information of the obstacle, acceleration information of the obstacle, shape information of the obstacle, and type information of the obstacle.

For example, the position information of the obstacle may indicate a position where the obstacle is located in the simulation scene, and the position information of the obstacle may be coordinates of the obstacle.

For example, the speed information of the obstacle may indicate a moving speed of the obstacle in the simulation scene. In a simulation scene, the barrier can move at a constant speed or at a variable speed.

For example, the acceleration information of the obstacle may indicate an acceleration of the obstacle. In the simulation scene, the obstacles can move with even acceleration or with variable acceleration.

For example, the shape information of the obstacle may indicate a size of the obstacle. The dimensions of the obstacle may include, for example, length, width, height, and the like.

For example, the type information of the obstacle may indicate the type of the obstacle. The types of obstacles may include pedestrians, non-motor vehicles, and the like.

In some embodiments, the environmental information includes at least one of: lane line information, traffic direction information, road network topology information, and rules information.

For example, lane line information may indicate the location and type of lane line. The types of lane lines may include long solid lines, dashed lines, and the like.

For example, the traffic indication information may be information of a traffic light. Based on the traffic indication information, it may be determined whether to perform operations such as left turn, right turn, straight running, and the like.

For example, the road network topology information may be a topological relationship indicating roads and road junctions within the simulation scene.

For example, the rule information is related to traffic rules. The rule information may include: the highest speed of movement, the lowest speed of movement, and so on.

In the disclosed embodiment, the environmental information further includes sidewalk information. For example, the sidewalk information may indicate a location of the sidewalk.

According to the embodiment of the disclosure, the target state information is determined according to the road network topology information, the rule information and the like, so that the behavior of the barrier is more reasonable and meets the regulation of the relevant traffic rules. And is also helpful for obtaining more accurate and reasonable test results.

In some embodiments, in some implementations of operation S230, the target state information includes target speed information, and controlling the obstacle movement using the target state information includes: and controlling the barrier to move according to the speed indicated by the target speed information. The following will be described in detail with reference to fig. 3A to 3C.

Fig. 3A-3C are schematic diagrams of a simulation testing method of an autonomous vehicle according to one embodiment of the disclosure.

As shown in fig. 3A, in a real scene, a vehicle 301 'is traveling on a lane 310', and a drive test vehicle 302 'is traveling on a lane 320'. During travel, the drive test vehicle 302' may utilize various sensors to collect drive test data. In a real scenario, the vehicle 301 'is traveling straight on the lane 310'. The drive test vehicle 302 'is traveling straight on the lane 320'. The speed of drive test vehicle 302 'may be the same as the speed of vehicle 301'. The acceleration of the drive test vehicle 302 'may be the same as the acceleration of the vehicle 301'.

As shown in fig. 3B, in the simulation scenario, at time T1, the vehicle 301 moves on the lane 310, and the vehicle 303 under test moves on the lane 320. The speed of the vehicle 301 may be less than the speed of the vehicle 303 under test. The acceleration of the vehicle 303 under test may be less than the acceleration of the vehicle 301. The state information of the vehicle 303 under test is determined by the autonomous driving controller.

As shown in fig. 3C, in the simulation scenario, at a time T2 next to the time T1, the vehicle 303 to be tested turns left to enter the lane 310, and the vehicle 301 still moves on the lane 310. If the vehicle 301 and the vehicle 303 to be detected still keep the original speed and acceleration to move, the vehicle 303 to be detected and the vehicle 301 will collide (or end-to-end), and an unreasonable interaction behavior is generated. In this case, a target deep learning model may be determined according to the type of interactive behavior (in the case where the vehicle under test is turning, the obstacle is going straight). The deep learning model is constructed based on a long-term and short-term memory model, for example. The state information and the environmental information of the vehicle 301 are processed by using the target deep learning model to obtain target state information. For example, the target state information includes target speed information. The vehicle 301 may be controlled to move (e.g., decelerate) at a speed indicated by the target speed information to avoid collision with the vehicle 303 to be measured. Therefore, unreasonable interaction behaviors are reduced, and the test result of the vehicle to be tested is more accurate.

In some embodiments, the target state information includes target position information and target velocity information, and controlling the movement of the obstacle using the target state information includes: and controlling the obstacle to move to the position indicated by the target position information according to the speed indicated by the target speed information.

For example, lane 1 is located on the left side of lane 2 in the simulation scenario. At time T1, the obstacle moves on the lane 1, and the vehicle to be measured moves on the lane 2. The speed of the obstacle may be less than the speed of the vehicle under test. The acceleration of the obstacle may be greater than the acceleration of the vehicle under test. The state information of the vehicle to be tested is determined by the automatic driving controller.

In a simulation scene, at a next time T2 of the time T1, the vehicle to be tested turns left and wants to enter the lane 1, and the obstacle still runs on the lane 1. If the obstacle and the vehicle to be detected still keep the original speed and acceleration to move, the vehicle to be detected and the obstacle collide with each other, and unreasonable interactive behaviors are generated. In this case, a target deep learning model may be determined according to the type of the interactive behavior (in the case where the vehicle under test is turning, the obstacle goes straight). The target deep learning module is constructed based on a vector network model, for example. And processing preset state information and environment information of the barrier by using the target deep learning model to obtain target state information. For example, the target state information includes target speed information and target position information. The obstacle may be controlled to move (e.g., decelerate) at a speed indicated by the target speed information and to a target position (e.g., lane 2) to avoid collision with the vehicle to be measured. Therefore, unreasonable interaction behaviors are sufficiently reduced, and the test result of the vehicle to be tested is more accurate.

It will be appreciated that the above described unreasonable interaction behavior is merely an example. The autopilot controller may generate a variety of commands to control the vehicle under test.

FIG. 4 is a flow chart of a method of simulation testing of an autonomous vehicle according to another embodiment of the disclosure.

As shown in fig. 4, the method 400 may include operations S401 to S404. It is understood that S401 to S404 may be performed before the above operation S210.

In operation S401, environment information and preset state information of a plurality of obstacles are determined according to the drive test data.

For example, the environmental information and the preset state information of the plurality of obstacles may be extracted from the drive test data.

In operation S402, a simulation scenario is established according to the environmental information and preset state information of the plurality of obstacles.

For example, scene reconstruction may be performed in time and space by using various simulation tools according to the environmental information and the preset state information of the obstacle, so as to obtain a simulation scene.

In operation S403, a vehicle under test is added to the simulation scenario.

For example, parameters such as speed, acceleration, starting position, and ending position of the vehicle to be tested in the simulation scene may be set. Based on the automatic driving mode, in a simulation scene, the vehicle to be tested moves from the initial position to the end position according to the set speed and acceleration. The acceleration may be an average acceleration and the velocity may be an average velocity. During the moving process, the automatic driving controller of the vehicle to be tested can adjust the speed, the acceleration and the position of the vehicle to be tested according to the information such as the position of the obstacle.

In operation S404, a type of an interactive behavior between the vehicle under test and the obstacle is determined.

In the embodiment of the present disclosure, the trajectory information of the obstacle is determined according to the preset state information of the obstacle.

For example, the preset state information includes position information and speed information, whereby trajectory information of the obstacle can be determined.

In the embodiment of the disclosure, the type of the interaction behavior between the vehicle to be tested and the obstacle is determined according to the track information of the vehicle to be tested and the track information of the obstacle.

For example, in an autonomous driving mode, an autonomous driving controller of the vehicle under test may determine the speed and position of the vehicle under test, and thus the trajectory of the vehicle under test. According to the track information of the vehicle to be detected and the track information of the obstacle, the type of the interaction behavior between the vehicle to be detected and the obstacle can be determined.

For example, the interactive behaviors may include preset interactive behaviors and non-preset interactive behaviors.

It is to be understood that the simulation test method of an autonomous vehicle according to the present disclosure is described above in detail, and the principle of the simulation test method of an autonomous vehicle according to the present disclosure will be described in detail with reference to the related embodiments.

FIG. 5 is a schematic diagram of a method for simulation testing of an autonomous vehicle according to another embodiment of the present disclosure.

As shown in fig. 5, according to the drive test data 5001, environmental information and preset state information of a plurality of obstacles can be determined. From the environmental information and the preset state information, a simulated context 5002 can be established. After the vehicle under test is added to the simulated context 5002, the monitoring module 5003 can be utilized to monitor the interaction behavior between the obstacle and the vehicle under test in real time so as to determine the type of the interaction behavior between the vehicle under test and the obstacle.

After the fact that the preset interactive behaviors exist between the vehicle to be tested and the obstacle is determined, a target deep learning Model is determined from the target deep learning models Model _1, 5004, the target deep learning Model _ N5005 and the target deep learning Model _ All 5006 according to the type of the preset interactive behaviors. N is an integer greater than 1. In this embodiment, the determined target deep learning Model may be the target deep learning Model _1 5004. The target state information 5007 can be obtained by processing the environmental information and the preset state information of the obstacles by using the target deep learning Model _1 5004. The movement of obstacles in the simulation scene is controlled using the target state information 5007.

It is to be appreciated that the target deep learning Model _ All 5006 can be a generalized target deep learning Model as described above.

It will be appreciated that the above description details the simulation test method for an autonomous vehicle. The manner in which the target deep learning model is obtained will be described in detail below with reference to related embodiments.

FIG. 6 is a flow diagram of a method of training a deep learning model according to one embodiment of the present disclosure.

As shown in fig. 6, the method 600 may include operations S610 through S630.

In operation S610, first sample sub-data of sample data is input to the initial deep learning model, and output sample state information is obtained.

In the embodiment of the present disclosure, the sample data is the environmental information and the state information of the obstacle in one sample period.

For example, the length of the sample period may be 6 seconds.

In the disclosed embodiments, the first sample sub-data corresponds to a first sample sub-period of the sample period.

For example, the first sample sub-period may be a period corresponding to 1 st to 3 rd seconds in the sample period.

In an embodiment of the present disclosure, the sample data is derived from drive test data corresponding to a type of a preset interaction behavior.

For example, the drive test data is derived from drive test data collected by a drive test vehicle. Drive test vehicles may be deployed with various sensors to collect drive test data.

For example, the deep learning model may be at least one of an long-term memory model and a vector network model.

In the embodiment of the present disclosure, each type of preset interaction behavior corresponds to one drive test data.

For example, there may be multiple types of preset interaction behavior, and correspondingly, there may be multiple drive test sub-data.

In the embodiment of the present disclosure, the preset interaction behavior may be: the obstacle performs a predetermined behavior in a process in which the drive test vehicle performs a driving behavior.

For example, one preset interaction behavior may be: under the condition that the road test vehicle moves straight, the obstacle makes a turning behavior.

In operation S620, a loss value is determined according to the output sample state information and the second sample sub-data of the sample data.

In an embodiment of the present disclosure, the second sample sub-data corresponds to a second sample sub-period of the sample period.

For example, the second sample subinterval may be a period corresponding to 4 th to 6 th seconds in the sample period.

In embodiments of the present disclosure, various loss functions may be utilized to determine a loss value between the output sample state information and the second sample sub-data.

For example, the various loss functions may include, for example, an L1 loss function.

For example, the output sample state information includes output sample position information and output sample velocity information. The second sample sub-data includes position information and speed information of the obstacle. The second sample sub-data may be a label of the first sample sub-data. The label may also be referred to as a true value. The output sample state information may be an output value of the model. Thus, various loss functions may be utilized to determine the loss value based on the difference between the output sample state information and the second sample sub-data.

In operation S630, parameters of the initial deep learning model are adjusted using the loss value, resulting in a target deep learning model.

In the embodiment of the present disclosure, the target deep learning model corresponds to a type of a preset interaction behavior.

For example, the parameters of the initial deep learning model may be adjusted by using a gradient descent or the like according to the loss value to obtain the target deep learning model.

In some embodiments, the type of preset interaction behavior comprises at least one of: in the case of a straight-ahead drive of the road test vehicle, the obstacle is steered; in the case of a drive test vehicle turning, the obstacle is moving straight; under the condition that the road test vehicle runs straight, the obstacle enters a lane where the road test vehicle is located; and an obstacle following drive test vehicle.

For example, in the case of a drive test vehicle traveling straight, obstacle steering may include: in the real scene, lane 1 is to the left of lane 2. The drive test vehicle moves on lane 1, the obstacle moves on lane 2, the obstacle turns left into lane 1 and moves behind the drive test vehicle.

For example, in the case of a drive test vehicle turning, the obstacle straight-ahead may include: in the real scene, lane 1 is to the left of lane 2. The drive test vehicle moves on the lane 2 and the obstacle moves on the lane 1. When the drive test vehicle turns left to enter the lane 1, the obstacle continues to go straight on the lane 1 and is located behind the drive test vehicle.

For example, in the case where the drive test vehicle is traveling straight, the obstacle entering the lane in which the drive test vehicle is located may include: in the real scene, lane 1 is to the left of lane 2. The drive test vehicle moves on the lane 1, the obstacle moves on the lane 2, and the obstacle turns left to enter the lane 1 and moves in front of the drive test vehicle.

For example, the obstacle following drive test vehicle may include: the track heights of the obstacles and the drive test vehicle are consistent. In the real scene, lane 1 is to the left of lane 2. The drive test vehicle moves on the lane 2, and the obstacle moves on the lane 2. After the drive test vehicle has turned left into lane 1, the obstacle has also turned left into lane 1.

It can be understood that, in the case where the type of the preset interactive behavior between the vehicle to be tested and the obstacle is "in the case of straight-ahead driving of the vehicle to be tested, the target deep learning model trained by the road test sub-data in which the type of the preset interactive behavior is" in the case of straight-ahead driving of the vehicle to be tested, obstacle steering "may be used to process the environmental information and the state information of the obstacle.

It can be understood that in a real scene, when a preset interactive behavior is generated between the obstacle and the drive test vehicle, the obstacle takes a corresponding action to avoid a traffic event, and the state information of the obstacle is adjusted. Thus, the state (e.g., speed, position, etc.) of the obstacle can be efficiently adjusted using the target deep learning model trained using the relevant drive test data.

For another example, a generalized target deep learning model may be trained using roadside data associated with other predetermined interaction behaviors.

In some embodiments, the status information of the obstacle includes at least one of: position information of the obstacle, velocity information of the obstacle, acceleration information of the obstacle, shape information of the obstacle, and type information of the obstacle.

For example, the position information of the obstacle may indicate a position where the obstacle is located in the real scene, and the position information of the obstacle may be coordinates of the obstacle.

For example, the speed information of the obstacle may indicate a moving speed of the obstacle in the real scene. In a real scene, the barrier can move at a constant speed or at a variable speed.

For example, the acceleration information of the obstacle may indicate an acceleration of the obstacle. In a real scene, the obstacle can move with even acceleration or with variable acceleration.

In some embodiments, the environmental information includes at least one of: lane line information, traffic indication information, road network topology information, and regulation information.

For example, the road network topology information may be a topological relation indicating roads and road junctions in a real scene.

For example, the rule information is related to traffic rules. The rule information may include: highest speed of movement, lowest speed of movement, and the like.

According to the embodiment of the disclosure, the state information of the output sample is determined according to the road network topology information, the rule information and the like, and the deep learning model can be efficiently trained. The state information determined by the deep learning model can be more accurate and reasonable, and accords with the regulation of related traffic rules.

In some embodiments, inputting a first sample sub-data of the sample data into the initial deep learning model, and obtaining the output sample state information comprises: and classifying the drive test data to obtain a plurality of drive test sub-data.

In the embodiment of the present disclosure, the drive test sub-data corresponds to a type of a preset interaction behavior.

For example, a classification model may be used to classify the drive test data, so as to obtain the drive test data corresponding to the type of each preset interaction behavior.

In some embodiments, the drive test data may be classified according to semantic tags of the drive test data.

In an embodiment of the present disclosure, the semantic tag of the drive test data includes: the method comprises the following steps of behavior semantic tags of road test vehicles, structure semantic tags of road networks, behavior semantic tags of obstacles and type semantic tags of the obstacles.

For example, the behavior semantic tags of the drive test vehicle include: straight going, left turning, right turning, turning around, converging, etc.

For example, the structural semantic labels of road network include: straight, curved, crossing, etc.

For example, the behavioral semantic tags of the obstacles include: cutting, following, turning left, turning right, etc.

For example, the type semantic tags for the obstacles include: automotive, pedestrian, non-automotive, etc.

In the embodiment of the disclosure, the semantic label of the drive test data in a period of time may be input into the classification model to obtain the output interaction behavior type. And training a classification model according to the difference between the output interactive behavior type and the interactive behavior type label.

For example, the interactive behavior type tags may be manually determined. The drive test data of multiple time periods can be labeled manually to obtain multiple interactive behavior type labels so as to train a classification model. And classifying the road test data by using the trained classification model to obtain the road test sub-data.

In some embodiments, inputting a first sample sub-data of the sample data into the initial deep learning model, and obtaining the output sample state information comprises: dividing the drive test sub data to obtain a plurality of sample data; dividing the sample data to obtain first sample sub-data and second sample sub-data; and inputting the first sample sub-data into the initial deep learning model to obtain output sample state information.

For example, the time duration corresponding to the drive test data may be 600 seconds, for example. The road test data is divided, and 100 sample data can be obtained. The length of the sample period corresponding to each sample data may be 6 seconds.

For another example, the sample data corresponds to a sample period having a length of 6 seconds. The first sample sub-period corresponding to the first sample sub-data may be 1 st second to 3 rd second of the sample period. The second sample sub-period corresponding to the second sample sub-data may be 4 th to 6 th seconds of the sample period.

The principles of the deep learning model training method of the present disclosure will be described in detail below with reference to specific embodiments.

FIG. 7 is a schematic diagram of a training method of a deep learning model according to one embodiment of the present disclosure.

As shown in fig. 7, the classification model may be used to classify the drive test data Sample _ Total 7008, so as to obtain a plurality of drive test data. The plurality of drive test data includes drive test data Sample _1 7009, drive test data Sample _ N7010, and drive test data Sample _ All 7011.

The drive test sub-data can be divided to obtain a plurality of sample data. And training the initial deep learning model by using the sample data to obtain a target deep learning model. The plurality of drive test sub-data correspond to the plurality of target deep learning models. The plurality of target deep learning models includes: target deep learning Model _1 7004,... A target deep learning Model _ N7005, and target deep learning Model _ All 7006.

FIG. 8 is a block diagram of a simulation test setup of an autonomous vehicle according to one embodiment of the present disclosure.

As shown in fig. 8, the apparatus 800 may include a first determination module 810, a processing module 820, and a control module 830.

The first determining module 810 is configured to determine, in response to determining that a preset interactive behavior exists between the vehicle to be tested and the obstacle, a target deep learning model according to a type of the preset interactive behavior. For example, the obstacle is moved in the simulation scene based on the preset state information, and the vehicle under test is moved in the simulation scene based on the automatic driving mode.

And the processing module 820 is configured to process the environment information and the preset state information of the obstacle by using the target deep learning model to obtain target state information.

And a control module 830 for controlling the movement of the obstacle using the target state information.

In some embodiments, the preset state information of the obstacle includes at least one of: position information of the obstacle, velocity information of the obstacle, acceleration information of the obstacle, shape information of the obstacle, and type information of the obstacle, the environment information including at least one of: lane line information, traffic indication information, road network topology information, and regulation information.

In some embodiments, the target state information includes target speed information, and the control module includes: and the first control sub-module is used for controlling the barrier to move according to the speed indicated by the target speed information.

In some embodiments, the target state information includes target position information and target velocity information, and the control module includes: and the second control submodule is used for controlling the barrier to move to the position indicated by the target position information according to the speed indicated by the target speed information.

In some embodiments, the apparatus 800 may further comprise: the second determining module is used for determining environmental information and preset state information of a plurality of obstacles according to the drive test data; the establishing module is used for establishing a simulation scene according to the environment information and the preset state information of the plurality of obstacles; the adding module is used for adding the vehicle to be tested into the simulation scene; and the third determination module is used for determining the type of the interaction behavior between the vehicle to be detected and the obstacle.

In some embodiments, the third determining module comprises: the first determining submodule is used for determining the track information of the obstacle according to the preset state information of the obstacle; and the second determining submodule is used for determining the type of the interactive behavior between the vehicle to be detected and the obstacle according to the track information of the vehicle to be detected and the track information of the obstacle.

In some embodiments, the apparatus 800 may further comprise: and the fourth determining module is used for determining the test result of the vehicle to be tested according to the vehicle to be tested and the obstacle controlled by the target behavior information.

Fig. 9 is a block diagram of a training apparatus for a deep learning model according to another embodiment of the present disclosure.

As shown in fig. 9, the apparatus 900 may include an obtaining module 910, a fifth determining module 920, and an adjusting module 930.

An obtaining module 910, configured to input the first sample sub-data of the sample data into the initial deep learning model, to obtain state information of the output sample. For example, the sample data is environmental information and state information of an obstacle in a sample period, and the sample data is from drive test data corresponding to the type of a preset interactive behavior;

a fifth determining module 920, configured to determine a loss value according to the output sample state information and the second sample sub-data of the sample data.

An adjusting module 930, configured to adjust parameters of the initial deep learning model by using the loss value to obtain a target deep learning model, where the target deep learning model corresponds to a type of a preset interaction behavior.

In some embodiments, the obtaining module comprises: the classification submodule is used for classifying the drive test data to obtain a plurality of drive test subdata, wherein the drive test subdata corresponds to the type of a preset interaction behavior; the first dividing module is used for dividing the road test sub data to obtain a plurality of sample data; the second division submodule is used for dividing the sample data to obtain first sample sub-data and second sample sub-data, wherein the first sample sub-data corresponds to a first sample sub-time period of a sample time period, and the second sample sub-data corresponds to a second sample sub-time period of the sample time period; and the obtaining submodule is used for inputting the first sample sub-data into the initial deep learning model to obtain the output sample state information.

In the technical scheme of the disclosure, the collection, storage, use, processing, transmission, provision, disclosure and other processing of the personal information of the related user are all in accordance with the regulations of related laws and regulations and do not violate the good customs of the public order.

The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.

FIG. 10 illustrates a schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 10, the apparatus 1000 includes a computing unit 1001 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a Random Access Memory (RAM) 1003. In the RAM 1003, various programs and data necessary for the operation of the device 1000 can be stored. The calculation unit 1001, the ROM 1002, and the RAM 1003 are connected to each other by a bus 1004. An input/output (I/O) interface 1005 is also connected to bus 1004.

A number of components in device 1000 are connected to I/O interface 1005, including: an input unit 1006 such as a keyboard, a mouse, and the like; an output unit 1007 such as various types of displays, speakers, and the like; a storage unit 1008 such as a magnetic disk, an optical disk, or the like; and a communication unit 1009 such as a network card, a modem, a wireless communication transceiver, or the like. The communication unit 1009 allows the device 1000 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunication networks.

Computing unit 1001 may be a variety of general and/or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 1001 executes the respective methods and processes described above, such as a simulation test method of an autonomous vehicle and/or a training method of a deep learning model. For example, in some embodiments, the simulation testing method of an autonomous vehicle and/or the training method of a deep learning model may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 1000 via ROM 1002 and/or communications unit 1009. When the computer program is loaded into RAM 1003 and executed by computing unit 1001, one or more steps of the above-described simulation testing method of an autonomous vehicle and/or training method of a deep learning model may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured in any other suitable manner (e.g., by means of firmware) to perform a simulation test method and/or a training method of a deep learning model of the autonomous vehicle.

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) display or an LCD (liquid crystal display)) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

It should be understood that various forms of the flows shown above, reordering, adding or deleting steps, may be used. For example, the steps described in the present disclosure may be executed in parallel or sequentially or in different orders, and are not limited herein as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved.

The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made, depending on design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A simulation test method of an autonomous vehicle, comprising:

in response to determining that a preset interactive behavior exists between a vehicle to be tested and an obstacle, determining a target deep learning model according to the type of the preset interactive behavior, wherein the obstacle moves in a simulation scene based on preset state information, and the vehicle to be tested moves in the simulation scene based on an automatic driving mode;

processing environmental information and preset state information of the barrier by using the target deep learning model to obtain target state information; and

and controlling the barrier to move by using the target state information.

2. The method of claim 1, wherein the type of preset interaction behavior comprises at least one of:

under the condition that the vehicle to be detected moves straight, the obstacle turns;

under the condition that the vehicle to be detected turns, the obstacle moves straight;

under the condition that the vehicle to be detected runs straight, the obstacle enters a lane where the vehicle to be detected is located; and

the obstacle follows the vehicle to be measured.

3. The method of claim 1, wherein the preset status information of the obstacle comprises at least one of: position information of the obstacle, speed information of the obstacle, acceleration information of the obstacle, shape information of the obstacle, and type information of the obstacle,

the environmental information includes at least one of: lane line information, traffic indication information, road network topology information, and regulation information.

4. The method of claim 1, wherein the target state information includes target speed information,

the controlling the obstacle to move using the target state information includes:

and controlling the barrier to move according to the speed indicated by the target speed information.

5. The method of claim 1, wherein the target state information includes target location information and target velocity information,

and controlling the barrier to move to the position indicated by the target position information according to the speed indicated by the target speed information.

6. The method of claim 1, further comprising:

determining the environment information and preset state information of a plurality of obstacles according to drive test data;

establishing the simulation scene according to the environment information and preset state information of a plurality of obstacles;

adding the vehicle to be tested into the simulation scene; and

and determining the type of the interactive behavior between the vehicle to be tested and the obstacle.

7. The method of claim 6, wherein the determining the type of interaction between the vehicle under test and the obstacle comprises:

determining track information of the obstacle according to preset state information of the obstacle; and

and determining the type of the interactive behavior between the vehicle to be detected and the obstacle according to the track information of the vehicle to be detected and the track information of the obstacle.

8. The method of claim 1, further comprising:

and determining a test result of the vehicle to be tested according to the vehicle to be tested and the obstacle controlled by the target state information.

9. A training method of a deep learning model comprises the following steps:

inputting first sample sub-data of sample data into an initial deep learning model to obtain output sample state information, wherein the sample data is environment information and state information of an obstacle in a sample period, and the sample data is from road test sub-data corresponding to the type of a preset interaction behavior;

determining a loss value according to the output sample state information and second sample subdata of the sample data; and

and adjusting parameters of the initial deep learning model by using the loss value to obtain a target deep learning model, wherein the target deep learning model corresponds to the type of a preset interaction behavior.

10. The method of claim 9, wherein said inputting a first sample sub-data of sample data into an initial deep learning model, resulting in output sample state information comprises:

classifying the drive test data to obtain a plurality of drive test sub-data, wherein the drive test sub-data corresponds to the type of a preset interactive behavior;

dividing the drive test sub data to obtain a plurality of sample data;

dividing the sample data to obtain first sample sub-data and second sample sub-data, wherein the first sample sub-data corresponds to a first sample sub-time period of the sample time period, and the second sample sub-data corresponds to a second sample sub-time period of the sample time period; and

and inputting the first sample sub-data into the initial deep learning model to obtain the state information of the output sample.

11. A simulation test apparatus of an autonomous vehicle, comprising:

the system comprises a first determining module, a target deep learning module and a target deep learning module, wherein the first determining module is used for determining a preset interactive behavior between a vehicle to be tested and an obstacle in response to the fact that the preset interactive behavior exists, and determining a target deep learning model according to the type of the preset interactive behavior, wherein the obstacle moves in a simulation scene based on preset state information, and the vehicle to be tested moves in the simulation scene based on an automatic driving mode;

the processing module is used for processing environmental information and preset state information of the barrier by using the target deep learning model to obtain target state information; and

and the control module is used for controlling the barrier to move by utilizing the target state information.

12. The apparatus of claim 11, wherein the type of preset interaction behavior comprises at least one of:

the obstacle follows the vehicle to be tested.

13. The apparatus of claim 11, wherein the preset status information of the obstacle comprises at least one of: position information of the obstacle, speed information of the obstacle, acceleration information of the obstacle, shape information of the obstacle, and type information of the obstacle,

the environmental information includes at least one of: lane line information, traffic direction information, road network topology information, and rules information.

14. The apparatus of claim 11, wherein the target state information comprises target speed information,

the control module includes:

and the first control sub-module is used for controlling the barrier to move according to the speed indicated by the target speed information.

15. The apparatus of claim 11, wherein the target state information includes target location information and target velocity information,

the control module includes:

and the second control submodule is used for controlling the barrier to move to the position indicated by the target position information according to the speed indicated by the target speed information.

16. The apparatus of claim 11, further comprising:

the second determining module is used for determining the environment information and the preset state information of the plurality of obstacles according to the drive test data;

the establishing module is used for establishing the simulation scene according to the environment information and the preset state information of the plurality of obstacles;

the adding module is used for adding the vehicle to be tested into the simulation scene; and

and the third determination module is used for determining the type of the interactive behavior between the vehicle to be detected and the obstacle.

17. The apparatus of claim 16, wherein the third determining means comprises:

the first determining submodule is used for determining the track information of the obstacle according to the preset state information of the obstacle; and

and the second determining submodule is used for determining the type of the interactive behavior between the vehicle to be detected and the obstacle according to the track information of the vehicle to be detected and the track information of the obstacle.

18. The apparatus of claim 11, further comprising:

and the fourth determination module is used for determining the test result of the vehicle to be tested according to the vehicle to be tested and the obstacle controlled by the target state information.

19. A training apparatus for deep learning models, comprising:

the system comprises an obtaining module, a judging module and a judging module, wherein the obtaining module is used for inputting first sample sub-data of sample data into an initial deep learning model to obtain output sample state information, the sample data is environment information and barrier state information in a sample time period, and the sample data is road test sub-data corresponding to the type of a preset interactive behavior;

a fifth determining module, configured to determine a loss value according to the output sample state information and second sample sub-data of the sample data; and

and the adjusting module is used for adjusting the parameters of the initial deep learning model by using the loss value to obtain a target deep learning model, wherein the target deep learning model corresponds to the type of a preset interaction behavior.

20. The apparatus of claim 19, wherein the means for obtaining comprises:

the classification submodule is used for classifying the drive test data to obtain a plurality of drive test sub-data, wherein the drive test sub-data corresponds to the type of a preset interaction behavior;

the first dividing module is used for dividing the road test sub data to obtain a plurality of sample data;

a second division submodule, configured to divide the sample data to obtain first sample sub-data and second sample sub-data, where the first sample sub-data corresponds to a first sample sub-period of the sample period, and the second sample sub-data corresponds to a second sample sub-period of the sample period; and

and the obtaining submodule is used for inputting the first sample sub-data into the initial deep learning model to obtain the output sample state information.

21. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 10.

22. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1 to 10.

23. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 10.