CN114793460A

CN114793460A - Method for creating an algorithm for the computer simulation of a traffic participant, method for training an algorithm of at least one control device for a motor vehicle, computer program product and motor vehicle

Info

Publication number: CN114793460A
Application number: CN202080086080.0A
Authority: CN
Inventors: U·埃贝勒; C·蒂姆
Original assignee: PSA Automobiles SA
Current assignee: PSA Automobiles SA
Priority date: 2019-12-10
Filing date: 2020-12-03
Publication date: 2022-07-26
Also published as: DE102019219241A1; EP4073778A1; WO2021115918A1

Abstract

A computer-implemented method for creating a traffic participant algorithm for the computer simulation of traffic participants is described, wherein the traffic participants belong to traffic participants of an unsuccessfully protected category, wherein data of a plurality of different, actually present traffic participants in the traffic participants of the category are detected in a real traffic environment by means of sensors arranged on the traffic participants during the execution of at least one mission, wherein a movement trajectory of the traffic participants is determined from the data, wherein an average movement trajectory for the mission and a bandwidth for deviating from the average movement trajectory are calculated from the movement trajectory. A computer-implemented method for training an algorithm of at least one control device for a motor vehicle, a computer program product and a motor vehicle are also described.

Description

Method for creating an algorithm for the computer simulation of a traffic participant, method for training an algorithm of at least one control device for a motor vehicle, computer program product and motor vehicle

Technical Field

The invention relates to a computer-implemented method for creating an algorithm for computer simulation of a traffic participant, a computer-implemented method for training an algorithm of at least one control device for a motor vehicle, a computer program product, and a motor vehicle.

Background

Methods for creating an algorithm for the computer simulation of a traffic participant, methods for training an algorithm for at least one control device for a motor vehicle, computer program products and motor vehicles of the type mentioned at the outset are known from the prior art.

In recent years, the first motor vehicle which is driven partially automatically (corresponding to SAE2 grade according to SAE J3016) has become available in series. Vehicles that are driven automatically (corresponding to SAE > -3 according to SAE J3016) or autonomously (corresponding to SAE4/5 according to SAE J3016) must be able to react automatically to unknown traffic conditions with maximum safety according to a wide variety of preset settings, such as driving objectives and compliance with common traffic regulations. Due to the unpredictable behavior of other traffic participants, in particular of other human traffic participants, the traffic reality is highly complex, so that it is considered almost impossible to program the respective control algorithms and control devices of a motor vehicle using conventional methods and on the basis of artificial rules.

Furthermore, in order to solve highly complex problems by means of computers, it is known to: algorithms are developed using machine learning or artificial intelligence methods or through self-learning neural networks. Such algorithms may react more appropriately to complex traffic conditions than conventional algorithms on the one hand. On the other hand, the algorithm can be further developed and continuously improved in principle by continuous learning during the development process and in daily life by means of artificial intelligence. Alternatively, the state of the algorithm can be frozen and used in a corresponding control unit in the motor vehicle after the training phase and the end of the manufacturer verification in the development process.

A disadvantage of the known method is that, up to now, the simulation has been carried out with traffic participants who comply with the rules. Instead, what often happens in practice is: traffic participants do not comply with rules such as driving too fast, seemingly unproblematic crossing lane markings, inattention, passing right or moving on an inaccessible trajectory, etc. Therefore, algorithms trained solely with other traffic participants who adhere to traffic regulations are not well-prepared for human driving behavior. This leads to unnatural driving behavior of a motor vehicle equipped with a correspondingly trained algorithm, since the motor vehicle reacts less flexibly and can lead to critical situations if the algorithm fails to correctly predict the behavior of other traffic participants.

A traffic simulation is disclosed by JP 2009019920 a, which generates an actual route based on a prediction of pedestrian or cyclist behaviour. The route is optimized from the perspective of a pedestrian or cyclist and also contains the risk of collision with a strange vehicle, possibly with further parameters such as weather. The resulting model integrates a large number of routes, based on overall risk, with which real traffic conditions are simulated.

Thus, a situation is described in which a pedestrian or a bicycle riding vehicle reacts to a further parameter. However, the route is based on analog data. The disadvantage is that the resulting route does not simulate the behavior of the real traffic participants.

Disclosure of Invention

The object of the present invention is therefore to develop a computer-implemented method of the type mentioned at the outset, a computer program product and a motor vehicle for training at least one algorithm for a control device of a motor vehicle, and to develop a method of the type mentioned at the outset, a computer-implemented method for creating an algorithm for computer simulation of traffic participants, a computer program product and a motor vehicle for training at least one algorithm for a control device of a motor vehicle such that they are better configured for reflecting real traffic conditions.

The object is achieved by a computer-implemented method for creating a traffic participant algorithm for computer simulation of traffic participants according to claim 1, a computer-implemented method for training at least one algorithm for a control device of a motor vehicle according to the parallel claim 6, a computer program product according to the parallel claim 9 and a motor vehicle according to the parallel claim 10. Further configurations and developments are the subject matter of the dependent claims.

A computer-implemented method for creating a traffic participant algorithm for the computer simulation of traffic participants is described below, wherein the traffic participants belong to traffic participants of an unsuccessfully protected category, wherein data of a plurality of different, actually present traffic participants in the traffic participants of the category are detected in a real traffic environment by means of sensors arranged on the traffic participants during the execution of at least one task, wherein a movement trajectory of the traffic participants is determined from the data, wherein an average movement trajectory and a bandwidth deviating from the average movement trajectory of the task are calculated from the movement trajectory.

A traffic participant that is unprotected energetically refers to a traffic participant whose passive safety device does not provide any protection, or provides little protection, such as by a safety suit and/or a wearable protector and/or a helmet. Examples of this category are pedestrians, skateboarders, roller skaters, bicycle riders, motorcycle riders, four-wheel vehicle riders, scooter riders, wheelchair riders and the like.

The task can be primarily a task that is addressed to the actually present traffic participants or can be derived from data that are obtained from the movement data of the traffic participants, which move between a common starting point or starting point region and a common end point or end point region without explicit tasks. The corresponding task can also be formulated more complex, for example taking one of a plurality of possible routes, or it can have intermediate objectives.

A possible source may be, for example, a location recording device such as a mobile phone or a smart watch. Many traffic participants carry mobile telephones with them, which are generally suitable for recording corresponding data.

Ideally, the number of traffic participants is so high that the trajectory data obtained therefrom is statistically significant that the trajectory can be reliably deduced and the bandwidth can be derived. Statistical significance was considered from 15 traffic participants.

The bandwidth may for example cover 95 or 99% of the individual tracks, thereby excluding only those tracks which are very far from the average track.

The average value can be determined according to various known averaging methods, for example as a geometric mean.

Thus, the respective traffic participant algorithm may produce different trajectories within the bandwidth for a given task that more closely approximates the true behavior of the traffic participant than a fixed trajectory for the same task. In one possible configuration, these differences can be achieved by various parameters.

In a first further development, the traffic participant algorithm can be designed such that it generates a randomized trajectory within the bandwidth. This randomized trajectory leads to different behavior patterns in different training periods or different traffic participants within a simulation simulated by means of the same algorithm. The result is thus that new and unpredictable situations can continue to be created which are closer to reality than situations which repeat in the same way.

In a further embodiment, it can be provided that the traffic participant is a vehicle of a given, unprotected vehicle category, wherein at least one sensor for detecting data is respectively mounted on or assigned to the vehicle, wherein the data is evaluated by the traffic participant algorithm.

In the case of vehicles in which the traffic participant is controlled by a human being, sensors, for example acceleration sensors, are often already arranged on or in the respective motor vehicle. If the resulting data is evaluated, the number of additional required sensors may be reduced or additional sensors may be avoided altogether.

According to one development, the data can be transformed and/or scaled before being evaluated by the traffic participant algorithm, so that better usable sensor data, for example accelerations, can be generated from the raw sensor data, for example voltage values.

In a further embodiment, it can be provided that the at least one sensor arranged on the traffic participant or on the vehicle is a camera, a GPS sensor, an acceleration sensor, a lidar sensor and/or a radar sensor.

The absolute position can be detected by means of an FPS sensor, from which the movement pattern and trajectory can be derived. The relative movement data can be determined by means of an acceleration sensor. A camera, lidar sensor or radar sensor may be used to detect the environment and may therefore generate a trajectory based on a known starting point or at least perform a plausibility verification. This may be helpful to identify: the determined trajectory comprises an avoidance movement in response to a fixed or moving obstacle. Such trajectories can then be classified accordingly.

In a further embodiment, it can be provided that at least one sensor for detecting a traffic participant is arranged in the traffic environment, wherein the data are evaluated and are introduced into the traffic participant algorithm.

The sensors arranged in the infrastructure may be, for example, stationary, for example, traffic monitoring cameras, which can detect the entire traffic event and thus provide additional data to verify the plausibility of the behavior of the traffic participants. From this, in particular the expected behavior of the relevant traffic participant can be deduced.

In a further development, it can be provided that a self-learning neural network is provided, wherein the data are provided to the self-learning neural network, wherein the traffic participant algorithm is trained by the self-learning neural network.

By means of the self-learning neural network, a traffic participant algorithm can be provided which is able to react in a human-like manner to unknown situations. It is thus possible to provide a universally applicable algorithm for traffic simulation, by means of which a simulation of a respective traffic participant belonging to an unsuccessfully protected category can be generated.

A first independent subject matter relates to a computer-implemented method for training an algorithm of at least one control device for a motor vehicle, wherein the control device is provided for implementing an automated or autonomous driving function by predicting an assembly of the motor vehicle using the at least one algorithm based on input data, wherein the algorithm is trained by a self-learning neural network, comprising the following steps:

a) providing a computer program product module for an automated or autonomous driving function, wherein the computer program product module contains the algorithm to be trained and the self-learning neural network;

b) providing a simulation environment with simulation parameters, wherein the simulation environment contains map data of a real-existing application area, motor vehicles and at least one further, simulated traffic participant as a proxy, wherein the behavior of the at least one further traffic participant is determined by a traffic participant algorithm generated in the above-described manner;

c) providing a mission for a motor vehicle; and

d) the task is performed and the algorithm is trained.

Since the algorithm of the control device is trained with the aid of an agent which simulates a more natural behavior of traffic participants of the category to be protected against force, such as pedestrians, bicycle riders, motorcycle riders or the like, than an agent programmed in a conventional manner, a more natural simulation environment can be provided, so that the corresponding algorithm can already reach a high degree of maturity in a pure simulation.

In a first further development, it can be provided that at least one of the further, simulated road users generates a trajectory which is subjected to parameter changes in a randomized or random manner.

In each simulation, trajectories that undergo parameter changes randomly or randomly are reproduced in situ, so that the behavior of the respective traffic participant or agent is not predetermined. This leads to particularly complex challenges for the algorithm to be trained and to more robust algorithms.

In a further development, it can be provided that a plurality of further, simulated road users are provided, which move at least partially along the mean trajectory partially within the bandwidth.

If the traffic participants exhibit different behaviors, i.e. some do a "normal" behavior and others do an abnormal behavior, a large number of agents may thus be provided, which are able to simulate a very real traffic event.

A further independent subject matter relates to a device for creating a traffic participant algorithm for the computer simulation of traffic participants, wherein the traffic participants belong to an unsuccessfully protected category of traffic participants, wherein sensors are arranged on a plurality of different actually existing traffic participants in the category of traffic participants, with which sensors data can be detected in an actual traffic environment during the implementation of at least one task, wherein means for determining a movement trajectory from the data are provided, wherein the means are provided for calculating an average movement trajectory and a bandwidth deviating from the average movement trajectory for the task from the movement trajectory.

In a further development, it can be provided that the road user is a vehicle of a given, unprotected vehicle category, wherein in each case at least one sensor for detecting data is attached to or assigned to the vehicle, wherein the road user algorithm is designed to evaluate the data of the at least one sensor.

In a further embodiment, it can be provided that at least one sensor for detecting a traffic participant is arranged in the traffic environment, wherein means are provided for evaluating the data and introducing it into the traffic participant algorithm.

In a further embodiment, it can be provided that a self-learning neural network is provided, data being provided to the self-learning neural network, the self-learning neural network being provided for training the traffic participant algorithm.

A further independent subject matter relates to a device for training an algorithm of at least one control unit for a motor vehicle, wherein the control unit is provided for carrying out an automated or autonomous driving function by predicting an assembly of the motor vehicle using the at least one algorithm on the basis of input data, wherein a self-learning neural network is provided for training the algorithm, wherein:

a) providing a computer program product module for automated or autonomous driving functions, wherein the computer program product module contains the algorithm to be trained and the self-learning neural network;

b) providing a simulation environment with simulation parameters, wherein the simulation environment contains map data of a real-existing application area, motor vehicles and at least one further, simulated traffic participant as a proxy, wherein the behavior of the at least one further traffic participant is determined by a traffic participant algorithm generated in the manner described above;

c) providing a mission for a motor vehicle; and

d) means are provided for performing the task and training the algorithm.

In a first further development, provision can be made for a means for randomizing or randomly varying a parameter of the trajectory of at least one of the further, simulated road users.

In a further development, provision can be made for a plurality of further, simulated traffic participants to be provided, which are provided for movement at least partially along the mean trajectory partially within the bandwidth.

A further independent subject matter relates to a computer program product having a computer-readable storage medium on which are embedded instructions which, when executed by at least one computing unit, cause the at least one computing unit to be arranged for implementing a method of the above-mentioned type.

The method can be implemented distributed over one or more computing units, so that certain method steps are implemented on the one computing unit and further method steps are implemented on at least one further computing unit, wherein, if necessary, the computed data can be transmitted between the computing units.

Another independent subject matter relates to a motor vehicle having a computer program product of the above-mentioned type.

Drawings

Further features and details emerge from the following description, in which (with reference to the drawings, if appropriate) at least one embodiment is described in detail. The described and/or illustrated features form the subject matter themselves or in any meaningful combination, if appropriate independently of the claims, and in particular additionally also form the subject matter of one or more independent applications. Identical, similar and/or functionally identical components are provided with the same reference signs. Here, schematically:

fig. 1 shows a motor vehicle, which is provided for autonomous driving;

FIG. 2 illustrates a computer program product for the motor vehicle of FIG. 1;

FIG. 3 shows a site with the motor vehicle and other traffic participants of FIG. 1;

FIG. 4 shows a representation of a traffic participant in a category that is unprotected from force;

FIG. 5 shows a flow diagram of a method;

FIG. 6 shows a flow chart of another method;

FIG. 7 illustrates a schematic diagram of generating a virtual agent; and

fig. 8 shows a schematic diagram of an algorithm for generating an algorithm for controlling a motor vehicle control device of the motor vehicle in fig. 1.

Detailed Description

Fig. 1 shows a motor vehicle 2, which is provided for autonomous driving.

The motor vehicle 2 has a motor vehicle control device 4, which has a computing unit 6 and a memory 8. In the memory 8 a computer program product is stored which is described in detail later, in particular in connection with fig. 2, 3 and 8.

The vehicle control device 4 is connected on the one hand to a series of environmental sensors which allow the current position of the vehicle 2 and the corresponding traffic situation to be detected. To this end,

environment sensors

10, 12 are located in front of motor vehicle 2,

environment sensors

14, 16 are located in the rear of motor vehicle 2, a camera 18 and a GPS module 20. Depending on the configuration, further sensors, for example wheel speed sensors, acceleration sensors, etc., can be provided, which are connected to the vehicle control device 4.

During operation of the motor vehicle 2, the computing unit 6 loads a computer program product stored in the memory 8 and executes the computer program product. Based on the algorithm and the input signals, the computing unit 6 determines a control of the motor vehicle 2, which control can be effected by the computing unit 6 by intervening on a steering device 22, an engine control device 24 and a brake 26, which are each connected to the motor vehicle control device 4.

Fig. 2 shows a computer program product 28 with computer program product modules 30.

The computer program product 30 has a self-learning neural network 32 that trains an algorithm 34. The self-learning neural network 32 learns according to a reinforcement learning method, that is, the neural network 32 attempts to obtain a benefit for improved behavior, i.e., improve the algorithm 34, by altering the algorithm 34 according to one or more criteria or metrics. Alternatively, however, known supervised and unsupervised learning methods and combinations of these learning methods may be used.

The algorithm 34 may essentially consist of a complex filter with a matrix of values (usually called weights) which define a filter function which determines the behavior of the algorithm 34 on the basis of the input variables sensed by the ambient sensors 10 to 20 and generates control signals for controlling the motor vehicle 2.

The monitoring of the quality of the algorithm 34 is carried out by a further computer program product module 36 which monitors the input variables and the output variables, determines therefrom an index and checks on the basis of this index whether the function corresponds to the quality. At the same time, the computer program product module 36 may provide negative and positive benefits to the neural network 32.

Fig. 3 shows the site 36.

The motor vehicle 2 is driving on a street 38 which intersects a street 40 at a street intersection 42.

A plurality of motorcycle riders, such as motorcycle rider 44 shown in fig. 3, have the task 50 of driving from a starting point 46 on street 40 to a target point 48 on street 38. The route requires a transition from street 40 to street 38. The motorcycle rider 44 travels the determined trajectory 52.1 during this task 50.

This task 50 is repeated several times by different motorcycle riders, all having the task of traveling from origin 46 to target point 48. In practice, the starting and target points may be defined as areas or corridors, for example as two-dimensional areas or as starting and target lines.

In this case, the other motorcycle riders ride further trajectories 52.2, 52.3 and 52.4 which are respectively different from one another and which are due, on the one hand, to the driving behavior of the respective traffic participant and, on the other hand, to the respectively different traffic conditions or specific situations which exist.

With the aid of the traffic participant algorithm described here for generating the class for motorcycle riders, an average trajectory 54 and bandwidth boundaries 56.1,56.2 are generated therefrom, which describe the usual behavior and the usual deviations from the average trajectory 54.

The bandwidth boundaries 56.1,56.2 cover the majority of the traveled trajectories in each section, i.e. the individual trajectories of the trajectories 52.1, 52.2 do not need to be completely within the bandwidth boundaries 56.1,56.2 from the start point to the end point.

In addition, fixedly positioned traffic monitoring cameras 58 may be provided at the street intersections 42, which observe traffic events and thereby record the behavior of the motorcycle rider 44 and the trajectory 52.1 of his travel, as well as other traffic events such as the movement of pedestrians 60.

The algorithm 34 for the control device 4 of the motor vehicle 2 is then trained in a simulation of the location 36, as described below, wherein the simulation of the motorcycle rider 44 is carried out as an agent, which may have the same tasks 50 within the framework of the simulation.

Fig. 4 shows a motorcycle rider 44.

The motorcycle rider 44 has a GPS sensor 62 that continuously detects data regarding the position of the motorcycle rider 44. The GPS sensor 62 may be mounted in a mobile phone carried by the motorcycle rider 44, for example.

Furthermore, a control device 64 is provided, which can control, for example, a motorcycle 65 of the motorcycle rider 44 and can detect data.

In addition, a camera 66 is connected to the control device 64, which camera records traffic events in front of the motorcycle rider 44. Furthermore, the control device 64 may detect data about the motorcycle, such as throttle attitude, engaged gear, speed, tilt angle, and the like.

Fig. 5 shows a flow chart for generating a traffic participant algorithm.

In a first step software is provided.

Neural networks and algorithms are then provided.

A task, such as task 50 shown in FIG. 3, is then determined as the motorcycle rider travels from start point 46 to target point 48.

Sensor data, such as GPS sensor 62, control device 64, and camera 66 and traffic monitoring camera 58, are then detected by the motorcycle rider 44 and other motorcycle riders, respectively. The sensor data is fed into a neural network where the average trajectory and bandwidth are found.

Next, a random module is added, by means of which the trajectory in the simulation can be made to vary randomly.

An algorithm is generated and output therefrom, which can be used in another traffic simulation.

Fig. 6 shows a flow chart for training the algorithm 34.

In a first step, a computer program product module (software) for a control unit 4 of a motor vehicle 2 to be trained is provided.

In a subsequent step, map data of the location 38 is provided.

The task for the algorithm 34 is then defined, for example, as the motor vehicle 2 reaching the target location from the starting location in a defined traffic situation.

Other traffic participants, such as motorcycle riders, are present in the traffic situation. To this end, virtual objects and agents are defined and provided. At least one of these agents acts in accordance with the algorithm generated in figure 5. The relevant agent may have a task 50.

One or more different traffic conditions with the virtual object and the agent are then simulated.

The algorithm is then trained in a control unit of the motor vehicle on the basis of the proposed conditions.

By observing the vehicle to be trained and the feedback with the simulation environment, the corresponding information can be used to influence the simulation.

A large number of different traffic participants, such as cars, trucks, motorcycles, cyclists, pedestrians, can be reflected by means of the method described here. Also, real infrastructure such as guideboards, traffic lights, street signs, traffic markings, etc. may be displayed.

By means of this method, complex traffic maneuvers can be actually studied within the framework of cooperative flows, for example driving at intersections with a plurality of traffic participants and communication between the traffic participants, in order for example not to endanger the traffic participants.

Fig. 7 shows another schematic diagram of a generation agent 70, here simulating a motorcycle rider 44.

The starting point here is that the agent 50 is generated by an algorithm 72 developed by means of a self-learning method. The algorithm 74 running in the computer 74 is provided with data for the GPS sensor 62, the control device 64, the camera 66 and the traffic monitoring camera 58.

The self-learning neural network 76 optimizes the algorithm 62 by changing the algorithm and presenting an algorithm 72 'and checks whether the modified algorithm 72' works better than the original algorithm 72. Once a certain quality index is met, the algorithm 72 is frozen and the virtual agent 70 is provided by means of the compiler 78, which can be used in the simulation environment.

Fig. 8 illustrates a simulation environment 80.

The simulated environment 80 is provided by map data of the location 34 and simulated motor vehicles 2 and other traffic participants, such as pedestrians 60 and virtual agents 70.

Then, a task is set up which the algorithm 34 controlling the motor vehicle 2 should carry out. This is processed (as described in connection with fig. 7, for example) by the neural network 32, which alters the algorithm 34 until certain quality criteria are met.

The computer program product module 30 for the control device 4 of the motor vehicle 2 is then generated by means of the compiler 82.

Although the subject matter is illustrated and described in detail by way of embodiments, the invention is not limited to the disclosed examples and other variants can be derived therefrom by the person skilled in the art. It is therefore evident that a large number of variant possibilities exist. Furthermore, the exemplary embodiments described are expressly intended only as examples and they are not to be construed in any way as limiting the scope of protection, the possibilities of application or the configuration of the invention, for example. Rather, the foregoing description and the accompanying drawings may enable one skilled in the art to practice the exemplary embodiments with particularity, wherein various changes may be made in the function and arrangement of elements described in each individual exemplary embodiment without departing from the scope of protection defined by the claims and their legal equivalents (e.g., as further explained in the specification) while knowing the inventive concepts disclosed.

List of reference numerals

2 Motor vehicle

4 control device

6 calculating unit

8 memory

10 environmental sensor

11 environmental sensor

12 environment sensor

13 Environment sensor

14 vidicon

15 GPA module

16 wheel speed sensor

18 acceleration sensor

20 pedal sensor

22 steering device

24 motor control device

26 brake

28 computer program product

30 computer program product module

32 neural network

34, 34' algorithm

36 sites

38 street

40 street

42 street intersection

44 motorcycle rider

46 starting point

48 target points

50 tasks

52.1-52.4 tracks

54 mean trajectory

56.1,56.2 Bandwidth boundary

58 traffic monitoring camera

60 pedestrian

62 GPS sensor

64 control device

65 motorcycle

66 vidicon

70 virtual agent

72, 72' Algorithm

72' modified Algorithm

74 computer

76 self-learning neural network

78 compiler

80 simulation environment

82 a compiler.

Claims

1. A computer-implemented method for creating a traffic participant algorithm (72) for the computer simulation of traffic participants (44, 60), wherein the traffic participants belong to traffic participants (44, 60) of an unsuccessfully protected category, wherein data of a plurality of different, actually present traffic participants (44, 60) of the traffic participants of the category are detected in a real traffic environment (36) by means of sensors (62, 64, 66, 58) arranged on the traffic participants (44, 60) during the execution of at least one mission, wherein a movement trajectory (52.1-52.4) of the traffic participants (44, 60) is determined from the data, wherein an average movement trajectory (54) for the mission and a bandwidth (56.1) for deviating from the average movement trajectory (54) are calculated from the movement trajectories (52.1-52.4, 56.2).

2. The method according to claim 1, wherein the traffic participant is a vehicle (44) of a given, unsuccessfully protected vehicle category, wherein at least one sensor (62, 64, 66) for detecting data is respectively mounted on the vehicle (44) or assigned to the vehicle (44), wherein the data is evaluated by the traffic participant algorithm (72).

3. Method according to claim 1 or 2, wherein the at least one sensor (62, 64, 66) arranged on the traffic participant (44, 60) or on the vehicle (65) is a camera (66), a GPS sensor (62), an acceleration sensor, a lidar sensor and/or a radar sensor.

4. The method according to one of the preceding claims, wherein at least one sensor (58) detecting a traffic participant (44, 60) is arranged in the traffic environment (36), wherein the data is evaluated and is introduced into the traffic participant algorithm (72).

5. The method according to any one of the preceding claims, wherein a self-learning neural network (76) is provided, wherein the data is provided to the self-learning neural network (76), wherein the traffic participant algorithm (72) is trained by the self-learning neural network (76).

6. A computer-implemented method for training at least one algorithm (34) for a control unit (4) of a motor vehicle (2), wherein the control unit (4) is provided for implementing an automated or autonomous driving function by intervening in assemblies (22, 24, 26) of the motor vehicle (2) using the at least one algorithm (34) on the basis of input data, wherein the algorithm (34) is trained by means of a self-learning neural network (32), comprising the following steps:

a) providing a computer program product module (30) for an automated or autonomous driving function, wherein the computer program product module (30) contains the algorithm (34) to be trained and the self-learning neural network (32);

b) providing a simulation environment (36) with simulation parameters, wherein the simulation environment (36) contains map data (38) of the real-existing application domain, the motor vehicle (2) and at least one further, simulated traffic participant (48, 50) as agents, wherein the behavior of the at least one further traffic participant (48, 50) is determined by a traffic participant algorithm (72) generated according to any one of claims 1 to 5;

c) providing a task for a motor vehicle (2); and

d) the task is performed and the algorithm is trained (34).

7. The method according to claim 6, wherein at least one of the further, simulated traffic participants (48, 50) generates a trajectory (54, 56.1, 56.2) which is subjected to parameter variations randomly or randomly.

8. Method according to claim 6 or 7, wherein a plurality of further, simulated traffic participants (48, 50) are provided, which move at least partially along the average trajectory (54) partially within the bandwidth (56.1, 56.2).

9. A computer program product having a computer-readable storage medium (8) on which are embedded instructions that, when executed by at least one computing unit (6), cause the at least one computing unit (6) to be arranged for implementing a method according to any one of the preceding claims.

10. A motor vehicle having a computer program product according to claim 9.