EP4073778A1

EP4073778A1 - Method for creating a road user algorithm for computer simulation of road users, method for training at least one algorithm for a control unit of a motor vehicle, computer program product, and motor vehicle

Info

Publication number: EP4073778A1
Application number: EP20820809.0A
Authority: EP
Inventors: Ulrich Eberle; Christoph THIEM
Original assignee: PSA Automobiles SA
Current assignee: Stellantis Auto SAS
Priority date: 2019-12-10
Filing date: 2020-12-03
Publication date: 2022-10-19
Anticipated expiration: 2040-12-03
Also published as: CN114793460B; DE102019219241A1; EP4073778B1; WO2021115918A1; CN114793460A

Abstract

The invention relates to a computer-implemented method for creating a road user algorithm for computer simulation of road users, wherein: the road users belong to a class of poorly protected road users; data of a plurality of different, genuinely existing road users of the class in a genuine traffic environment is detected during the implementation of at least one mission with the aid of sensors affixed to the road users; movement trajectories of the road users are determined from the data; a mean movement trajectory for the mission and ranges for deviations from the mean trajectories are calculated from the movement trajectories. The invention further relates to a computer-implemented method for training at least one algorithm for a control unit of a motor vehicle, a computer program product, and a motor vehicle.

Description

METHOD FOR CREATING A TRAFFIC SUBSCRIBER ALGORITHM FOR COMPUTER SIMULATION OF TRAFFIC SUBSCRIBERS, METHOD FOR TRAINING AT LEAST ONE ALGORITHM FOR A CONTROL UNIT OF A MOTOR VEHICLE, COMPUTER VEHICLE DRIVE

A computer-implemented method for creating a traffic user algorithm for computer simulation of traffic users, a computer-implemented method for training at least one algorithm for a control unit of a motor vehicle, a computer program product and a motor vehicle are described here.

Methods for creating a road user algorithm for computer simulation of road users, methods for training at least one algorithm for a control device of a motor vehicle, computer program products and motor vehicles of the type mentioned at the beginning are known in the prior art.

The first partially automated vehicles (corresponding to SAE Level 2 in accordance with SAE J3016) have reached series production readiness in recent years. Automated driving (corresponds to SAE level> = 3 according to SAE J3016) or autonomous driving (corresponds to SAE level 4/5 according to SAE J3016) vehicles must be able to handle unfamiliar traffic situations independently with maximum safety based on a variety of specifications, for example destination and compliance with current traffic regulations can react. Since the reality of traffic is highly complex due to the unpredictability of the behavior of other road users, especially other human road users, it is almost impossible to program corresponding control algorithms and devices for motor vehicles with conventional methods and on the basis of man-made rules.

To cope with highly complex problems using computers, it is also known to develop algorithms using machine learning or artificial intelligence methods or to have them developed using self-learning neural networks. On the one hand, such algorithms can react more moderately to complex traffic situations than traditional algorithms. On the other hand, it is in principle with the help of artificial intelligence possible to further develop and continuously improve the algorithms during the development process and in everyday life through constant learning. Alternatively, a status of the algorithm can be frozen after the completion of a training phase in the development process and a validation by the manufacturer and put into use in corresponding control devices in motor vehicles.

The disadvantage of the known method is that the simulation has so far been carried out with rule-compliant road users. In reality, however, it often happens that road users do not behave in accordance with the rules, e.g. drive too fast, drive over lane markings for no apparent reason, are inattentive, overtake to the right or move on non-obvious trajectories, etc. The algorithm is therefore less prepared for human driving behavior. This leads to an unnatural driving behavior of a motor vehicle equipped with a suitably trained algorithm, since the motor vehicle can react less flexibly and critical situations can arise if the algorithm did not correctly anticipate the behavior of the other road user.

A traffic simulation is known from JP 2009 019920 A which generates a realistic route based on the prediction of pedestrian or bicycle behavior. The route is optimized from the perspective of the pedestrian or the cyclist and also includes the risk of a collision with another vehicle and possibly other parameters such as weather. The resulting models integrate a large number of routes, depending on the overall risk, with which real traffic conditions are simulated.

This describes a situation in which a pedestrian or cyclist reacts to other parameters. However, the route is based on simulation data. The disadvantage of this is that the routes generated in this way cannot simulate the behavior of real road users.

Thus, the task arises of computer-implemented methods for training at least one algorithm for a control unit of a motor vehicle, computer program products and motor vehicles of the type mentioned at the beginning Road user algorithm for computer simulation of road users, computer-implemented methods for training at least one algorithm for a control device of a motor vehicle, computer program products and motor vehicles of the type mentioned above to the effect that they are better set up to map real traffic conditions.

The object is achieved by a computer-implemented method for creating a road user algorithm for computer simulation of road users according to claim 1, a computer-implemented method for training at least one algorithm for a control unit of a motor vehicle according to the independent claim 6, a computer program product according to the independent claim 9 and a motor vehicle according to the independent claim 10. Further designs and developments are the subject of the dependent claims.

The following is a computer-implemented method for creating a traffic participant algorithm for computer simulation of road users, the road users belonging to a class of poorly protected road users, with data from a plurality of different, real existing road users of the class in a real traffic environment with the help of sensors attached to the road users are detected during the implementation of at least one mission, with movement trajectories of the road users being determined from the data, with an average movement trajectory for the mission and bandwidths for deviations from the average movement trajectory being calculated from the movement trajectories.

Badly protected road users are those whose passive safety equipment offers no, marginal or only little protection, e.g. through safety clothing and / or body-worn protectors and / or helmets. This class includes, for example, pedestrians, skateboarders, roller skaters, cyclists, motorcyclists, quad riders, scooter riders, wheelchair users and the like.

The mission can, among other things, be a mission presented to actually existing road users, or the mission can be derived from data obtained from movement data of the road users who move between a common starting point or area and a common end point or end without an explicit mission. move area. A corresponding mission can also be formulated in a more complex manner, for example to take a certain one of several possible routes or it can have intermediate destinations.

Possible sources can be, for example, position recording devices such as cell phones or smart watches. Many road users carry around cell phones, which are usually suitable for recording relevant data.

The majority of road users are ideally so large that the trajectory data obtained are statistically significant, so that reliable statements about the trajectories and bandwidths can be derived. Statistical significance can be assumed for 15 or more road users.

The bandwidths can cover, for example, 95 or 99% of the individual trajectories, so that only trajectories that are very far away from the mean trajectory are excluded.

The mean value can be determined using various known mean value methods, e.g. as a geometric mean.

A corresponding road user algorithm can thus generate different trajectories for a given mission within the bandwidths, which are closer to a real behavior of road users than a fixed trajectory for the same mission. The differences can be achieved in a possible embodiment through various parameters.

In a first further refinement, the traffic user algorithm can be designed in such a way that it generates randomized trajectories within the bandwidths. Such randomized trajectories lead to different behavioral patterns in different training cycles or in different road users within a simulation that are simulated using the same algorithm. As a result, new and unpredictable situations can be created that are closer to reality than uniformly repetitive situations. In another further refinement, it can be provided that the traffic participants represent vehicles of a given, poorly protected vehicle class, with at least one sensor for recording data being attached to the vehicles or assigned to the vehicles, the data being evaluated by the traffic participant algorithm become.

In the event that the road users are vehicles that are controlled by people, sensors are often arranged on or in the corresponding motor vehicles anyway, e.g. acceleration sensors. If the data generated from this are evaluated, the number of additionally required sensors can be reduced or additional sensors can be completely avoided.

Before the data is evaluated by the traffic user algorithm, it can be converted and / or converted according to a further development so that more usable sensor data, e.g. acceleration, can be generated from raw sensor data, e.g. voltage values.

In a further refinement, it can be provided that the at least one sensor arranged on the road user or the vehicle is a camera, a GPS sensor, a lidar and / or a radar sensor.

With the help of GPS sensors, absolute positions can be recorded, from which movement patterns and trajectories can be derived. Relative movement data can be determined with the aid of acceleration sensors. Cameras, lidar or radar sensors can be used to record the surroundings and can therefore generate the trajectories starting from a known starting point or at least make them plausible. This can be helpful, for example, in order to recognize that a certain trajectory contains an evasive movement in response to a stationary or moving obstacle. Such a trajectory can then be classified accordingly.

In a further refinement, it can be provided that at least one sensor which detects the road users is arranged in the traffic environment, the data being evaluated and incorporated into the road user algorithm.

A sensor arranged in the infrastructure, which can be static, for example, for A traffic surveillance camera, for example, can record the entire traffic situation and thus provide additional data for checking the plausibility of the behavior of road users. In this way, among other things, anticipatory behavior of the road users concerned can be derived.

In a further refinement, provision can be made for a self-learning neural network to be provided, the data being provided to the self-learning neural network, the traffic user algorithm being trained by the self-learning neural network.

With the help of a self-learning neural network, a traffic user algorithm can be created that can react to unknown situations in a human-like manner. It is thus possible to create a universally replaceable algorithm for traffic simulations, with the aid of which a simulation of a corresponding traffic participant who belongs to a poorly protected class can be generated.

A first independent subject relates to a computer-implemented method for training at least one algorithm for a control unit of a motor vehicle, the control unit being provided for implementing an automated or autonomous driving function by intervening in units of the motor vehicle on the basis of input data using the at least one algorithm , the algorithm being trained by a self-learning neural network, comprising the following steps: a) providing a computer program product module for the automated or autonomous driving function, the computer program product module containing 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 operational area, the motor vehicle and, as an agent, at least one further simulated road user, where e in the behavior of the at least one other road user is determined by a road user algorithm that was generated according to the type described above; c) providing a mission for the motor vehicle, and d) carrying out the mission and training the algorithm.

The fact that the algorithm of the control unit is trained with the help of agents that a more natural behavior of road users of a poorly protected class, for example pedestrians, cyclists, motorcyclists or the like, simulate than conventionally programmed agents, a more lifelike simulation environment can be specified so that a corresponding algorithm can already reach a high level of maturity in a pure simulation.

In a first further refinement, it can be provided that at least one of the further simulated road users generates a trajectory that is randomized or stochastically parameter-varied.

A randomized or stochastically parameter-varied trajectory is newly generated in situ for each simulation so that the behavior of the relevant road user or agent is not determined in advance. This leads to particularly complex challenges for the algorithm to be trained and to a more robust algorithm.

In a further refinement, provision can be made for a plurality of further simulated road users to be provided, who move at least partly along the central trajectory and partly within the bandwidths.

If the road users behave differently, i.e. some behave "normally" and others behave abnormally, a large number of agents can be created from this, who are able to simulate very realistic traffic events.

Another independent subject relates to a device for creating a traffic user algorithm for computer simulation of road users, the road users belonging to a class of poorly protected road users, with sensors being attached to a plurality of different, real existing road users of the class, with which data is transferred to a real Traffic environment can be detected during the implementation of at least one mission, means for determining movement trajectories from the data are provided, with the means being set up to generate a mean movement trajectory from the movement trajectories. torie for the mission and bandwidths for deviations from the mean movement trajectory.

In a first further refinement, it can be provided that the traffic participants represent vehicles of a given, poorly protected vehicle class, with at least one sensor for recording data being attached to the vehicles or assigned to the vehicles, the traffic participant algorithm being designed to do this To evaluate data from the at least one sensor.

In a further refinement, it can be provided that at least one sensor which detects the road users is arranged in the traffic environment, with means for evaluating the data and for introducing it into the road user algorithm.

In a further refinement, a self-learning neural network can be provided, the data being provided to the self-learning neural network, the self-learning neural network being set up to train the road user algorithm.

Another independent subject matter relates to a device for training at least one algorithm for a control unit of a motor vehicle, the control unit being provided for implementing an automated or autonomous driving function by intervening in units of the motor vehicle on the basis of input data using the at least one algorithm , wherein a self-learning neural network is provided for training the algorithm, wherein: a) a computer program product module for the automated or autonomous driving function is available, the computer program product module containing the algorithm to be trained and the self-learning neural network, b) a simulation environment with simulation parameters stands, wherein the simulation environment contains map data of a real existing operational area, the motor vehicle and, as an agent, at least one further simulated road user, with a behavior of the at least one further road users are determined by a road user algorithm that was generated according to the type described above; c) a mission is ready for the motor vehicle, and d) means for performing the mission and training the algorithm are provided.

In a first further refinement, provision can be made for means for randomizing or for stochastically varying parameters of the trajectory of the at least one of the further simulated road users to be provided.

In a further refinement, means for providing a plurality of further simulated road users can be provided, which are set up to move at least partially along the central trajectory, partially within the bandwidths.

Another independent subject matter relates to a computer program product with a computer-readable storage medium on which instructions are embedded which, when executed by at least one processing unit, have the effect that the at least one processing unit is set up to execute the method of the type described above.

The method can be carried out on one or more processing units distributed so that certain method steps are carried out on one processing unit and other process steps are carried out on at least one other processing unit, with calculated data being able to be transmitted between the processing units if necessary.

Another independent subject matter relates to a motor vehicle with a computer program product of the type described above.

Further features and details emerge from the following description in which - if necessary with reference to the drawing - at least one Ausführungsbei game is described in detail. Described and / or graphically represented features form the object by themselves or in any meaningful combination, possibly also independently of the claims, and can in particular also include counter- be one or more separate registrations. Identical, similar and / or functionally identical parts are provided with the same reference numerals. They show schematically:

1 shows a motor vehicle which is set up for autonomous driving;

FIG. 2 shows a computer program product for the motor vehicle from FIG. 1; FIG.

3 shows a location with the motor vehicle from FIG. 1 and other traffic participants;

4 shows a representative of a road user of a class of poorly protected road users;

5 shows a flow chart of a method;

6 shows a flow chart of a further method;

7 shows a schematic diagram of the generation of a virtual agent, as well as

8 shows a schematic diagram of the generation of an algorithm for controlling a

Motor vehicle control device of the motor vehicle from FIG. 1.

1 shows a motor vehicle 2 which is set up for autonomous driving.

The motor vehicle 2 has a motor vehicle control device 4 with a computing unit 6 and a memory 8. A computer program product is stored in the memory 8 and is described in more detail below in particular in connection with FIGS. 2, 3 and 8.

The motor vehicle control device 4 is connected, on the one hand, to a series of environmental sensors which allow the current position of the motor vehicle 2 and the respective traffic situation to be recorded. These include environmental sensors 10, 12 at the front of the motor vehicle 2, environmental sensors 14, 16 at the rear of the motor vehicle 2, a camera 18 and a GPS module 20. Depending on the configuration, further sensors can be provided, for example wheel speed sensors, acceleration sensors, etc., which are connected to the motor vehicle control unit 4. During the operation of the motor vehicle 2, the computing unit 6 has loaded the computer program product stored in the memory 8 and executes it. On the basis of an algorithm and the input signals, the computing unit 6 decides on the control of the motor vehicle 2, which the computing unit 6 can achieve by intervening in the steering 22, engine control 24 and brakes 26, which are each connected to the motor vehicle control unit 4.

2 shows a computer program product 28 with a computer program product module 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 methods of reinforcement learning, d. H. By varying the algorithm 34, the neural network 32 tries to obtain rewards for improved behavior in accordance with one or more criteria or standards, i.e. for improvements to the algorithm 34. Alternatively, known learning methods of monitored and unsupervised learning and combinations can also be used this learning method can be used.

The algorithm 34 can essentially consist of a complex filter with a matrix of values, often called weights, which define a filter function that determines the behavior of the algorithm 34 as a function of input variables that are presently recorded by the environmental sensors 10 to 20 and control signals for controlling the motor vehicle 2 are generated.

The quality of the algorithm 34 is monitored by a further computer program product module 36, which monitors input variables and output variables, determines metrics therefrom and controls compliance with the quality by the functions on the basis of the metrics. At the same time, the computer program product module 36 can give negative as well as positive rewards for the neural network 32.

3 shows a location 36.

The motor vehicle 2 travels on a road 38 which intersects with a road 40 at a street intersection 42.

Several motorcyclists such as motorcyclist 44 shown in FIG. 3 have a mission 50 to drive from a starting point 46 on the road 40 to a destination point 48 on the road 38. The route requires turning from road 40 onto road 38. Motorcyclist 44 will drive a certain trajectory 52.1 during mission 50.

Mission 50 is repeated several times by different motorcyclists who all have the task of driving from starting point 46 to destination point 48. In practice, the start and finish points can be defined as areas or corridors, e.g. as a two-dimensional area or as start and finish lines.

Other motorcyclists will drive other trajectories 52.2, 52.3 and 52.4, which differ from one another and which are based on the one hand in the driving behavior of the corresponding road users and on the other hand in the respective prevailing, different traffic conditions or specific situations.

With the aid of the method described here for generating a road user algorithm for the class of motorcyclists, a mean trajectory 54 and bandwidth limits 56.1, 56.2 are generated therefrom, which describe a usual behavior and usual deviations from the mean trajectory 54.

The bandwidth limits 56.1, 56.2 cover most of the trajectories driven in each section, i.e. it is not necessary for a single one of the trajectories 52.1 to 52.4 to lie completely within the bandwidth limits 56.1, 56.2 from start to finish.

A stationary traffic surveillance camera 58 can also be provided at the intersection 42, which observes the traffic situation and thus records the behavior of the motorcyclist 44 and his driven trajectory 52.1 as well as other traffic events, for example the movement of a pedestrian 60.

Later, as described below, the algorithm 34 for the control unit 4 of the motor vehicle 2 is trained in a simulation of the location 36, the motorcyclist 44 being simulated as an agent who can have the same mission 50 as part of the simulation.

4 shows the motorcyclist 44. The motorcyclist 44 has a GPS sensor 62 which continuously records data about the position of the motorcyclist 44. The GPS sensor 62 can be installed, for example, in a mobile phone carried by the motorcyclist 44.

Furthermore, a controller 64 is provided which, for example, controls a motorcycle 65 of the motorcyclist 44 and which can acquire data.

In addition, a camera 66 is connected to the controller 64 and records the traffic happening in front of the motorcyclist 44. The controller 64 can also acquire data about the motorcycle, for example a throttle valve position, gear engaged, speed, lean angle and the like.

5 shows a flow chart for generating the traffic user algorithm.

In a first step, the software is provided.

A neural network and an algorithm are then provided.

A mission is then determined, for example the mission 50 shown in FIG. 3, to drive as a motorcyclist from the starting point 46 to the destination point 48.

Then, from the motorcyclist 44 and the other motorcyclists, each sensor data is recorded, for example from the GPS sensor 62, the controller 64 as well as the camera 66 and the traffic surveillance camera 58. The sensor data is fed into the neural network, there an average trajectory and bandwidths determined.

Then a stochastic module is added, with the help of which a trajectory can be varied randomly in the simulation.

From this, an algorithm is generated and output that can be used in a further traffic simulation.

6 shows a flow chart of the method for training the algorithm 34. In a first step, a computer program product module (software) to be trained is provided for the control device 4 of the motor vehicle 2.

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

A mission is then defined for the algorithm 34, for example as a motor vehicle 2 to get from a starting point to a destination in a certain traffic situation.

In the traffic situation there are other road users, e.g. a motorcyclist. For this purpose, virtual objects and agents are defined and made available. At least one of the agents behaves according to the algorithm generated in FIG. 5. The agent in question can have mission 50.

Then one or more different traffic situations are simulated with the virtual objects and agents.

The algorithm is then trained in the control unit of the motor vehicle on the basis of the situations posed.

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

With the help of the method described here, a large number of different road users can be mapped, for example cars, trucks, motorcycles, cyclists, pedestrians. Real infrastructure, for example signs, traffic lights, street symbols, guidelines, etc. can also be displayed.

With the help of the method, complex traffic maneuvers can be realistically examined within the framework of cooperative mobility, for example driving at an intersection where there are several road users and communication between road users, for example so as not to endanger any road user.

7 shows a further schematic diagram of the generation of an agent 70, here a simulation of the motorcyclist 44.

It is assumed here that an agent 50 has been trained by means of a self-learning Method refined algorithm 72 is generated. The algorithm 72, which runs in a computer 74, is supplied with the data from the GPS sensor 62, the controller 64, the camera 66 and the traffic monitoring camera 58.

A self-learning neural network 76 optimizes the algorithm 62 by varying the same and specifying an algorithm 72 'and checking whether the modified algorithm 72' works better than the original algorithm 72. As soon as certain quality metrics are met, the algorithm 72 is frozen and used a compiler 78, the virtual agent 70 is created, which can be used in simulation environments.

8 shows a simulation environment 80

The simulation environment 80 is provided by map data of the location 34 as well as simulations of the motor vehicle 2 and other road users such as the pedestrian 60 and the virtual agent 70.

A mission is then set up, which algorithm 34, which controls motor vehicle 2, is to carry out. This is, as described in connection with FIG. 7, processed by the neural network 32, which varies the algorithm 34 until certain quality standards are met.

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

Although the subject matter was illustrated and explained in more detail by means of exemplary embodiments, the invention is not restricted by the examples disclosed and other variations can be derived therefrom by the person skilled in the art. It is therefore clear that there is a multitude of possible variations. It is also clear that the exemplary embodiments mentioned are only examples that are not to be understood in any way as a limitation, for example, of the scope of protection, the possible applications or the configuration of the invention. Rather, the preceding description and the description of the figures enable the person skilled in the art to actually implement the exemplary embodiments, with the person skilled in the art knowing the inventive concept disclosed various changes, for example with regard to the function or the arrangement of individual elements mentioned in an exemplary embodiment without departing from the scope of protection which is defined by the claims and their legal equivalents, such as a more detailed explanation in the description.

List of reference symbols

2 motor vehicle

4 control unit

6 arithmetic unit

8 memories

10 environmental sensor

11 Environment sensor

12 environmental sensor

13 Environmental sensor

14 camera

15 GPS module

16 wheel speed sensor

18 accelerometer

20 pedal sensor

22 Steering

24 Motor control

26 brakes

28 Computer program product

30 computer program product module

32 neural network

34, 34 ‘algorithm

36 place

38 street

40 street

42 intersection

44 motorcyclists

46 starting point

48 target point

50 mission

52.1 - 52.4 trajectory

54 middle trajectory

56.1, 56.2 bandwidth limit 58 Traffic surveillance camera

60 pedestrians

62 GPS sensor

64 Control 65 Motorcycle

66 camera

70 virtual agent

72, 72 ‘algorithm

72 ‘modified algorithm 74 computer

76 self-learning neural network

78 compiler

80 simulation environment

82 compilers

Claims

1. Computer-implemented method for creating a road user algorithm (72) for computer simulation of road users (44, 60), the road users belonging to a class of poorly protected road users (44, 60), with data from a plurality of different, real existing road users (44, 60) of the class in a real traffic environment (36) with the aid of sensors (62, 64, 66, 58) attached to the road users (44, 60) during the implementation of at least one mission, with movement trajectories (52.1 - 52.4) the road users (44, 60) are determined, with a mean movement trajectory (54) for the mission and bandwidths (56.1, 56.2) for deviations from the mean movement trajectory (54) from the movement trajectories (52.1 - 52.4) be calculated.

2. The method according to claim 1, wherein the road users represent vehicles (44) of a given, poorly protected vehicle class, with the Fahrzeu gene (44) in each case at least one sensor (62, 64, 66) for collecting data is introduced or the Vehicles (44) is assigned, the data being evaluated by the traffic subscriber algorithm (72).

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

4. The method according to any one of the preceding claims, wherein in the traffic environment (36) at least one the traffic participants (44, 60) detecting sensor (58) is arranged, the data being evaluated and 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, the data being provided to the self-learning neural network (76), the traffic user algorithm (72) being trained by the self-learning neural network (76) .

6. Computer-implemented method for training at least one algorithm (34) for a control unit (4) of a motor vehicle (2), the control unit (4) for implementing an automated or autonomous driving function by intervening in units (22, 24, 26) of the motor vehicle (2) on the basis of Input data is provided using the at least one algorithm (34), the algorithm (34) being trained by a self-learning neural network (32), comprising the following steps: a) Providing a computer program product module (30) for the 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) map data (38) of a real existing application area, the motor vehicle (2) and as an agent at least one further simulier th traffic participant (48, 50), with a behavior at least one it further traffic participant (48, 50) is determined by a traffic participant algorithm (72) which was generated according to one of claims 1 to 5; c) providing a mission for the motor vehicle (2), and d) carrying out the mission and training the algorithm (34).

7. The method according to claim 6, wherein at least one of the further simulated traffic participants (44, 60) generates a trajectory (54, 56.1, 56.2) which is randomized or stochastically parameter-varied.

8. The method according to claim 6 or 7, wherein a plurality of further simulated road users (44, 60) are provided, which move at least partially along the central trajectory (54), partially within the bandwidths (56.1, 56.2).

9. Computer program product with a computer-readable storage medium (8) on which instructions are embedded which, when executed by at least one processing unit (6), have the effect that the at least one processing unit (6) is directed to the method according to one of the preceding claims auszufüh ren.

10. Motor vehicle with a computer program product according to claim 9.