DE102019216836A1 - Method for training at least one algorithm for a control unit of a motor vehicle, computer program product and motor vehicle - Google Patents

Abstract

A method is described for training at least one algorithm for a control unit of a motor vehicle, 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, b) providing a simulation environment with simulation parameters , wherein the simulation environment contains map data of a real existing application area, the motor vehicle and at least one other simulated road user, the behavior of the motor vehicle and the at least one further road user being determined by a rule set with behavior parameters, c) providing a mission for the motor vehicle, d) Modifying at least one behavior parameter of the motor vehicle so that the at least one behavior parameter lies beyond the permissible limits, e) performing a simulation of the mission.

Description

A 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, computer program products and motor vehicles of the type mentioned at the beginning are known in the prior art. The first partially automated vehicles (corresponds to SAE Level 2 in accordance with SAE J3016) have reached series production readiness in recent years. Automated (corresponds to SAE level> = 3 according to SAE J3016) or autonomous (corresponds to SAE level 4/5 according to SAE J3016) vehicles must react independently to unknown traffic situations with maximum safety based on a variety of specifications, for example destination and compliance with common traffic rules can. Since the reality of traffic is highly complex due to the unpredictability of the behavior of other road users, in particular human road users, it is almost impossible to program corresponding control units of motor vehicles with conventional methods and on the basis of man-made rules.

In order to cope with complex problems by means of computers, it is also known to develop algorithms with methods of machine learning or artificial intelligence or to allow them to be developed by 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, with the help of artificial intelligence, it is in principle possible to further develop and continuously improve the algorithms during the development process and in everyday life through constant learning. Alternatively, the state of the algorithm can be frozen after the end of a training phase in the development process and a validation by the manufacturer.

From the DE 10 2017 007 136 A1 A method for training self-learning algorithms for an automatically drivable vehicle is known with a predetermined automation module by generating learning situations, the learning situations being generated as follows: Carrying out a traffic simulation in which a virtual ego vehicle is connected to the automation module of the real vehicle is set in a virtual scenario, the scenario comprising a route structure with a given route, furthermore, further, automatically generated further virtual moving objects with individually predeterminable object properties and behavior models, the objects in the course of the ongoing simulation with each other independently and adaptively on the basis of the respective object properties and behavior models interact, - Carrying out a driving dynamics simulation on the basis of the automation module and virtual sensor signals of the moving objects of a vehicle assigned to the ego vehicle th virtual sensor system, which corresponds to a sensor system of the actually existing vehicle, in which reactions of the ego vehicle are generated, - identification of a relevant learning situation based on selection criteria that are determined on the basis of predeterminable metrics.

The disadvantage here is that the simulation is carried out with road users who conform to the rules. 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 on the right, etc. An algorithm trained only with other road users who behave in accordance with the rules 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 reacts less flexibly.

Furthermore, there are situations in which the flow of traffic can be improved and, if necessary, the risk of an accident can be reduced if the ego vehicle does not behave one hundred percent according to the rules, e.g. if it drives over a solid line to avoid an obstacle, as long as this is the case is possible without risk, e.g. if there is no oncoming traffic. Braking could have the consequence that unprepared human drivers who were driving afterwards could have a rear-end collision due to the sudden interruption of the flow of traffic.

Thus, the task arises of developing methods, computer program products and motor vehicles of the type mentioned at the outset in such a way that a trained algorithm can be better adapted to real traffic situations.

The object is achieved by a method for training at least one algorithm for a control unit of a motor vehicle according to claim 1, a computer program product according to the independent claim 12 and a motor vehicle according to the independent claim 13.

1. a) Bereitstellen eines Computerprogrammproduktmoduls für die automatisierte bzw. autonome Fahrfunktion, wobei das Computerprogrammproduktmodul den zu trainierenden Algorithmus und das selbstlernende neuronale Netz enthält,
2. b) Bereitstellen einer Simulationsumgebung mit Simulationsparametern, wobei die Simulationsumgebung Kartendaten eines real existierenden Einsatzgebiets, das Kraftfahrzeug sowie wenigstens einen weiteren simulierten Verkehrsteilnehmer enthält, wobei ein Verhalten des Kraftfahrzeugs sowie des wenigstens einen weiteren Verkehrsteilnehmers durch einen Regelsatz bestimmt ist, wobei der Regelsatz zulässige Grenzen bestimmende Verhaltensparameter enthält,
3. c) Bereitstellen einer Mission für das Kraftfahrzeug,
4. d) Modifizieren wenigstens eines Verhaltensparameters des Kraftfahrzeugs, sodass der wenigstens eine Verhaltensparameter jenseits der zulässigen Grenzen liegt,
5. e) Durchführen einer Simulation der Mission.

A method for training at least one algorithm for a control unit of a motor vehicle is described below, the control unit for implementing an automated or autonomous driving function with intervention in units of the motor vehicle on the basis of input data using the at least one algorithm is provided, the algorithm being trained by a self-learning neural network, comprising the following steps:

1. 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,
2. b) Providing a simulation environment with simulation parameters, the simulation environment containing map data of a real existing area of use, the motor vehicle and at least one other simulated road user, the behavior of the motor vehicle and the at least one other road user being determined by a rule set, the rule set determining permissible limits Contains behavioral parameters,
3. c) Provision of a mission for the motor vehicle,
4. d) Modifying at least one behavior parameter of the motor vehicle so that the at least one behavior parameter lies beyond the permissible limits,
5. e) Performing a simulation of the mission.

Corresponding behavior parameters can, for example, be the permissible speed, a distance to be observed, thresholds for exceeding the prohibition (e.g. a period of time within which a traffic light that changes to red may be driven over, risk parameters in which a solid line may be driven over, and / or when it is allowed to continue driving despite the lack of right of way) ), a permitted variance of the position of the motor vehicle in the lane, a permitted overtaking side (in right-hand traffic only on the left or on both sides) and the like.

A corresponding mission can be, for example, to get from a starting point to a specified destination in the shortest possible time or in an energy-saving manner.

It was found that a suitably trained algorithm has a different driving behavior than a conventionally trained algorithm, even within narrower parameter limits, such as would then be used for use in a real motor vehicle. The driving behavior of such an algorithm is more natural, i.e. it corresponds more closely to the driving behavior of a person, which on the one hand benefits the occupants and on the other hand has a more natural effect on other road users. An example of this is passing a delivery vehicle parked in the second row and having to cross a solid line. An algorithm that was absolutely compliant with the rules would stop the motor vehicle and wait for the delivery vehicle to continue. An algorithm trained in accordance with the method described here, which may exceed applicable rules within narrow limits, e.g. if this can be done safely due to the lack of oncoming traffic, continues at such a point.

It can thus be achieved that a motor vehicle equipped with a corresponding algorithm can react more flexibly to traffic situations than an algorithm trained with conventional methods.

In a first further refinement, it can be provided that the neural network learns through reinforcement learning processes (also known as the RFL algorithm. RFL stands for: "reinforcement learning"), with at least one of the time to fulfill the mission and / or number of accidents in which the motor vehicle is involved during the mission, the simulation being repeated until a minimum metric is reached.

In particular, it can be provided that in order to successfully complete a mission it is necessary not to cause an accident. A further metric can be that it is not an indirect trigger of accidents with other road users, e.g. through sudden, unexpected hard braking.

By using reinforcement learning methods, the neural network learns better and better strategies for completing a given mission in successive simulations.

In a further refinement, it can be provided that the at least one abnormal road user is a motor vehicle, motorcyclist or pedestrian.

The vehicles are driven by people, some of whom do not comply with the rules. A simulation with such abnormally behaving road users is therefore particularly realistic.

In a further refinement, it can be provided that the computer program product module has an algorithm that has already been pre-trained with traffic participants that conform to the rules.

In this way, the already learned behaviors are refined and the training becomes more efficient and faster.

In a further refinement, it can be provided that the at least one behavior parameter is exceeded or undercut by a predetermined percentage.

This can be particularly useful for behavior parameters that can be expressed in numbers, for example a speed, a distance, a deviation from a specified driving line, etc.

In a further refinement, it can be provided that the simulation is repeated several times, with at least one simulation parameter being changed in each case.

Such a simulation parameter can be a behavior parameter, for example. By varying the corresponding parameters, over-specialization of the algorithm for a certain situation can be avoided.

In a further refinement, it can be provided that the simulation environment is varied.

This can also prevent the algorithm from being trained too much on the existing simulation environment. The variation can take place, for example, through modifications within the same traffic area (e.g. by changing road widths, right-of-way rules, traffic lights, road blocks, etc.) or by changing the traffic area as a whole.

In a further refinement, it can be provided that at least one behavior parameter is varied.

Such behavior parameters cover the usual driving behavior of different types of drivers, for example drivers who tend to drive too fast, drivers whose driving precision is less, etc.

In a further refinement, provision can be made for the number, positioning and / or missions of the other road users to be varied.

This creates new situations with which the algorithm can be trained further.

1. a) Bereitstellen des Computerprogrammproduktmoduls für die automatisierte bzw. autonome Fahrfunktion, wobei das Computerprogrammproduktmodul den zu trainierenden Algorithmus und das selbstlernende neuronale Netz enthält,
2. b) Bereitstellen einer Simulationsumgebung mit Simulationsparametern, wobei die Simulationsumgebung Kartendaten eines real existierenden Einsatzgebiets, das Kraftfahrzeug sowie wenigstens einen weiteren simulierten Verkehrsteilnehmer enthält, wobei der weitere simulierte Verkehrsteilnehmer durch einen Algorithmus simuliert wird, der nach dem zuvor beschriebenen Verfahren trainiert wurde,
3. c) Bereitstellen einer Mission für das Kraftfahrzeug,
4. d) Durchführen einer Simulation der Mission.

In a further refinement, it can be provided that the algorithm is further trained by a self-learning neural network, comprising the following steps:

1. a) providing the 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,
2. b) Providing a simulation environment with simulation parameters, the simulation environment containing map data of a real existing area of use, the motor vehicle and at least one further simulated road user, the further simulated road user being simulated by an algorithm that was trained according to the method described above,
3. c) Provision of a mission for the motor vehicle,
4. d) Performing a simulation of the mission.

In this way it is possible not only to design your own motor vehicle in a manner that does not conform to behavior, but also other motor vehicles, which increases the robustness of the algorithm.

The algorithm can also be applied to other road users, for example pedestrians or cyclists, in which case no computer program product module for an automated or autonomous driving function, but a computer program product module for movement behavior simulation is used. As a result, these agents can be made more realistic and can then be used in future training missions of the type described above, which increases the quality of the simulation.

In a further refinement, provision can be made for the computer program product module to be integrated in a control device of a motor vehicle and for the algorithm to be tested and / or trained in the real-world application.

In this way, influences of the real motor vehicle, which possibly cannot be completely simulated, can be taken into account. A motor vehicle that is actually moving can react differently than it is simulated.

1. a) Mittel zum Bereitstellen eines Computerprogrammproduktmoduls für die automatisierte bzw. autonome Fahrfunktion, wobei das Computerprogrammproduktmodul den zu trainierenden Algorithmus und das selbstlernende neuronale Netz enthält,
2. b) Mittel zum Bereitstellen einer Simulationsumgebung mit Simulationsparametern, wobei die Simulationsumgebung Kartendaten eines real existierenden Einsatzgebiets, das Kraftfahrzeug sowie wenigstens einen weiteren simulierten Verkehrsteilnehmer enthält, wobei ein Verhalten des Kraftfahrzeugs sowie des wenigstens einen weiteren Verkehrsteilnehmers durch einen Regelsatz bestimmt ist, wobei der Regelsatz zulässige Grenzen bestimmende Verhaltensparameter enthält,
3. c) Mittel zum Bereitstellen einer Mission für das Kraftfahrzeug,
4. d) Mittel zum Modifizieren wenigstens eines Verhaltensparameters des Kraftfahrzeugs, sodass der wenigstens eine Verhaltensparameter jenseits der zulässigen Grenzen liegt,
5. e) Mittel zum Durchführen einer Simulation der Mission.

A first independent subject relates to a device for training at least one algorithm for a control unit of a motor vehicle, wherein the control unit is used to implement an automated or autonomous driving function by intervening in units of the motor vehicle on the basis of input data at least one algorithm is provided, the algorithm being trained by a self-learning neural network, the following being provided:

1. a) Means for 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,
2. b) Means for providing a simulation environment with simulation parameters, the simulation environment containing map data of a real existing area of use, the motor vehicle and at least one other simulated road user, the behavior of the motor vehicle and the at least one other road user being determined by a rule set, the rule set being permissible Contains behavioral parameters that determine boundaries,
3. c) means for providing a mission for the motor vehicle,
4. d) means for modifying at least one behavior parameter of the motor vehicle so that the at least one behavior parameter is beyond the permissible limits,
5. e) means for performing a simulation of the mission.

In a first further refinement, it can be provided that the neural network has means for learning by reinforcing learning methods, with at least one of the time to fulfill the mission and / or the number of accidents in which the motor vehicle is involved as the reward metric during the Mission is provided, means are provided for repeating the simulation, so that the simulation is repeated until a minimum metric is reached.

In a further refinement, it can be provided that the at least one abnormal road user is a motor vehicle, motorcycle or pedestrian.

In a further refinement, it can be provided that the computer program product module has an algorithm that has already been pre-trained with traffic participants that conform to the rules.

In a further refinement, provision can be made for means to be provided for exceeding or falling below the at least one behavior parameter by a predetermined percentage.

In a further refinement, provision can be made for means to be provided for repeating the simulation several times, in which at least one simulation parameter is changed in each case.

In a further refinement, provision can be made for means for varying the simulation environment to be provided.

In a further refinement, provision can be made for means for varying at least one behavior parameter to be provided.

In a further refinement, provision can be made for means for varying the number, positioning and / or missions of the other road users to be provided.

1. a) Mittel zum Bereitstellen des Computerprogrammproduktmoduls für die automatisierte bzw. autonome Fahrfunktion, wobei das Computerprogrammproduktmodul den zu trainierenden Algorithmus und das selbstlernende neuronale Netz enthält,
2. b) Mittel zum Bereitstellen einer Simulationsumgebung mit Simulationsparametern, wobei die Simulationsumgebung Kartendaten eines real existierenden Einsatzgebiets, das Kraftfahrzeug sowie wenigstens einen weiteren simulierten Verkehrsteilnehmer enthält, wobei der weitere simulierte Verkehrsteilnehmer durch einen Algorithmus simuliert wird, der nach dem zuvor beschriebenen Verfahren trainiert wurde,
3. c) Mittel zum Bereitstellen einer Mission für das Kraftfahrzeug,
4. d) Mittel zum Durchführen einer Simulation der Mission.

In a further refinement, provision can be made for means for training the algorithm to be provided by a self-learning neural network, the following being provided:

1. a) Means for providing the 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,
2. b) Means for providing a simulation environment with simulation parameters, the simulation environment containing map data of a real existing area of use, the motor vehicle and at least one further simulated road user, the further simulated road user being simulated by an algorithm that was trained according to the method described above,
3. c) means for providing a mission for the motor vehicle,
4. d) means for performing a simulation of the mission.

In a further refinement, provision can be made for the computer program product module to be integrated in a control device of a motor vehicle, and means for testing and / or training the algorithm in the actual area of use being provided.

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, cause the at least one processing unit to be set up for this purpose is to carry out the procedure of the type described above.

The method can be carried out on one or more processing units, so that certain method steps are carried out on one processing unit and other method steps are carried out on at least one further processing unit, with calculated data being 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.

• 1 ein Kraftfahrzeug, das zum automatisierten bzw. autonomen Fahren eingerichtet ist;
• 2 ein Computerprogrammprodukt für das Kraftfahrzeug aus 1;
• 3 eine Simulationsumgebung dem Kraftfahrzeug aus 1, sowie
• 4 ein Ablaufdiagramm des Verfahrens.

Further features and details emerge from the following description, in which at least one exemplary embodiment is described in detail - possibly with reference to the drawing. Described and / or graphically represented features form the subject on their own or in any meaningful combination, possibly also independently of the claims, and in particular can additionally also be the subject of one or more separate applications. Identical, similar and / or functionally identical parts are provided with the same reference symbols. They show schematically:

• 1 a motor vehicle that is set up for automated or autonomous driving;
• 2 a computer program product for the motor vehicle 1 ;
• 3 a simulation environment from the motor vehicle 1 , as
• 4th a flow chart of the process.

1 shows a motor vehicle 2 that is set up for automated or autonomous driving.

The car 2 has a control unit 4th with one arithmetic unit 6th and a memory 8th on. In the storage room 8th a computer program product is stored which will be used in the following in connection with the 2 to 4th will be described in more detail.

The control unit 4th is connected on the one hand to a number of environmental sensors that detect the current position of the motor vehicle 2 as well as the respective traffic situation. These include environmental sensors 10 , 11 at the front of the motor vehicle 2 , Environmental sensors 12th , 13th at the rear of the motor vehicle 2 , a camera 14th as well as a GPS module 16 . The environmental sensors 10 to 13th can include, for example, radar, lidar and / or ultrasonic sensors.

In addition, there are sensors for detecting the state of the motor vehicle 2 provided, including wheel speed sensors 16 , Accelerometers 18th and pedal sensors 20th that came with the control unit 4th are connected. With the help of this motor vehicle sensor system, the current state of the motor vehicle 2 can be reliably recorded.

During the operation of the motor vehicle 2 has the arithmetic unit 6th that in memory 8th The stored computer program product is loaded and executed. The processing unit makes a decision based on an algorithm and the input signals 6th via the control of the motor vehicle 2 who have the arithmetic unit 6th through intervention in the steering 22nd , Engine control 24 as well as brakes 26th would achieve each with the control unit 4th are connected.

Data from the sensors 10 to 20th are continuously in memory 8th temporarily stored and discarded after a predetermined period of time so that these environmental data can be available for further evaluation.

The algorithm was trained according to the method described below.

2 shows a computer program product 28 with a computer program product module 30th .

The computer program product module 30th has a self-learning neural network 32 on that an algorithm 34 trained. The self-learning neural network 32 learns according to methods of reinforcement learning, ie the neural network 32 tried by varying the algorithm 34 , Rewards for improved behavior according to one or more metrics or measures, i.e. for improvements to the algorithm 34 to obtain. Alternatively, known learning methods of supervised and unsupervised learning, as well as combinations of these learning methods, can also be used.

The algorithm 34 can essentially consist of a complex filter with a matrix of values, usually called weights by experts, that define a filter function that determines the behavior of the algorithm 34 depending on input variables, which are present via the environmental sensors 10 to 20th are recorded, determined and control signals for controlling the motor vehicle 2 generated.

The computer program product module 30th can be used both in motor vehicles 2 as well as outside of the motor vehicle 2 can be used. It is thus possible to use the computer program product module 30th train in a real environment as well as in a simulation environment. In particular, according to the teaching described here, the training begins in a simulation environment, since this is safer than training in a real environment.

The computer program product module 30th is set up to set up a metric that should be improved. Such a metric can be, for example, a time until a predetermined mission, for example a destination, is reached. If the metric has exceeded a certain threshold, for example a time less than a limit time, the metric can be considered to be fulfilled and the algorithm can be frozen in this regard. It can then either be optimized with regard to another metric and trained further, or the algorithm can be tested in a real environment.

3 shows a simulation environment 36 the motor vehicle 2 out 1 .

In the simulation environment 36 is a street intersection 38 provided on which is a street 40 with a road 42 cross. The intersection 38 is based on real existing map data, so that the behavior of the algorithm is concrete 34 at this intersection 38 is simulated.

On the side of the road 40 parks a motor vehicle 44 such that a drive past cannot without crossing a solid line 46 is possible. At the same time, according to the simulation, a motorcyclist would like to 48 from the street 42 in the street 40 turn in. In addition, a pedestrian is moving 50 without paying attention to the traffic at high speed in the direction of movement 52 on the street 40 to apparently to cross this.

In the situation in question there is for the algorithm 34 make a variety of complex decisions. The first decision to be made is whether to see the solid line 46 may be run over at all. There is a passing of the parked motor vehicle 44 without crossing the solid line 46 is not possible, the decision is made by the algorithm 34 the answer is yes, but the question arises as to which driving parameters are used. The algorithm 34 must make a prediction about how the motorcyclist will feel 48 is behaved, which may be relatively close to the motor vehicle on its normal trajectory 2 would come up. In everyday life, however, it is often the case that corresponding motorcyclists can easily avoid or drive further to the right in their lane due to the small width of the motorcycle and the low speed in the intersection area.

Furthermore, there is the speed of the motor vehicle 2 to consider. When the motor vehicle 2 to overtake the parked vehicle 44 slightly accelerated, the probability that the motor vehicle 2 the planned trajectory of the motorcyclist 48 impaired. However, this could lead to the motor vehicle 2 the trajectory of the possibly inattentive pedestrian 50 just about to hit the road 40 to cross, crosses, which could result in an accident.

In successive iterations, the algorithm could 34 so first try the motor vehicle 44 without stopping to happen. The motor vehicle could do this 2 first to increase its speed to the motor vehicle beyond the permitted maximum speed 44 to happen. However, this could result in the pedestrians falling below a minimum distance 50 and the motor vehicle 2 to lead.

In a subsequent interaction, the algorithm 34 the car 2 move more slowly, which is dangerous for the motorcyclist 48 could cause.

The algorithm could then 34 the car 2 initially to pass the parked vehicle 44 accelerate and then brake again. This solution is to be preferred, since on the one hand it enables the parked motor vehicle to be passed 44 and enables the present mission to be completed quickly and, on the other hand, the metrics of the risk to other road users 48 , 50 optimized.

Optimization with regard to other criteria and metrics can then take place in order to increase the degree of maturity of the algorithm 34 to increase further.

4th shows a flow chart of the method.

First, after the start, the computer program product module is made available. The computer program product module contains the algorithm to be trained and a self-learning neural network.

A simulation environment is then provided on the basis of real map data. In addition to roads and certain rules, the simulation environment can also contain other road users and their missions.

On the basis of a basic algorithm, a rule set of the ego vehicle can be varied, which includes rules of behavior, for example Compliance with speeds, driving over solid lines, position in the lane, etc.

The simulation can then be carried out, using reinforcement learning methods to attempt to achieve individual metrics. As long as this is not the case, the strategy or the algorithm is varied and the simulation is carried out again until a certain individual metric is reached. This process is repeated for all metrics.

As soon as all metrics have been reached, the rule set of the ego vehicle is varied and the process is repeated until the algorithm has matured sufficiently. The algorithm can then be frozen.

The algorithm can be used, for example, in traffic simulations for simulated vehicles other than that of the motor vehicle to be trained. The method can also be applied to other road users.

The training can be continued in a real environment that is fully or mixed real.

Although the subject matter has been illustrated and explained in more detail by means of exemplary embodiments, the invention is not restricted by the disclosed examples 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 embodiments cited by way of example only represent examples which are not to be interpreted 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 implement the exemplary embodiments in concrete terms, whereby the person skilled in the art, with knowledge of the disclosed inventive concept, can make various changes, for example with regard to the function or the arrangement of individual elements mentioned in an exemplary embodiment, without the To leave the scope of protection that 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

4th

Control unit

6th

Arithmetic unit

8th

Storage

10

Environmental sensor

11

Environmental sensor

12th

Environmental sensor

13th

Environmental sensor

14th

camera

15th

GPS module

16

Wheel speed sensor

18th

Accelerometer

20th

Pedal sensor

22nd

steering

24

Engine control

26th

Brakes

28

Computer program product

30th

Computer program product module

32

neural network

34

algorithm

36

Simulation environment

38

Intersection

40, 42

road

44

parked motor vehicle

46

solid line

48

Motorcyclists

50

pedestrian

52

Direction of movement of the pedestrian 50

54

planned trajectory

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• DE 102017007136 A1 [0004]

Claims (13)

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 ) is provided on the basis of input data 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, the computer program product module (30) containing the algorithm (34) to be trained and the self-learning neural network (32), b) Providing a simulation environment (36) with simulation parameters, the simulation environment (36) containing map data (38) of a real existing area of use, the motor vehicle (2) and at least one further simulated road user (48, 50), with a behavior of the motor vehicle ( 2) and the at least one other road user (48, 50) is determined by a rule set, the rule set containing behavior parameters that determine permissible limits, c) providing a mission for the motor vehicle (2), d) Modifying at least one behavior parameter of the motor vehicle (2) so that the at least one behavior parameter is beyond the permissible limits, e) Performing a simulation of the mission. Procedure according to Claim 1 , wherein the neural network (32) learns by reinforcing learning method, wherein at least one of the time to fulfill the mission and / or the number of accidents in which the motor vehicle (2) is involved during the mission is used as the reward metric, the Simulation is repeated until a minimum metric is reached. Procedure according to Claim 1 or 2 wherein the at least one abnormal road user is a motor vehicle (2), motorcyclist (48) or pedestrian (50). Method according to one of the preceding claims, wherein the computer program product module (30) has an algorithm (34) which has already been pre-trained with traffic participants (48, 50) that conform to the rules. Method according to one of the preceding claims, wherein the at least one behavior parameter is exceeded or undercut by a predetermined percentage. Method according to one of the preceding claims, wherein the simulation is repeated several times, with at least one simulation parameter being changed in each case. Procedure according to Claim 6 , wherein the simulation environment (36) is varied. Procedure according to Claim 6 or 7th , wherein at least one behavior parameter is varied. Method according to one of the Claims 6 to 8th , the number, positioning and / or missions of the other road users (48, 50) being varied. Method according to one of the preceding claims, wherein the algorithm (34) is trained by a self-learning neural network (32), comprising the following steps: a) Providing the computer program product module (30) for the automated or autonomous driving function, 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, the simulation environment (36) being map data of a real application area (38), the motor vehicle (2) and at least contains a further simulated traffic participant (48, 50), the further simulated traffic participant (48, 50) being simulated by an algorithm which, according to one of the Claims 1 to 9 was trained, c) providing a mission for the motor vehicle (2), e) performing a simulation of the mission. Method according to one of the preceding claims, wherein the computer program product module (30) is integrated in a control unit (4) of a motor vehicle (2) and wherein the algorithm (34) is tested and / or trained in the actual application area (38). Computer program product with a computer-readable storage medium (8) on which instructions are embedded which, when executed by at least one computing unit (6), have the effect that the at least one computing unit (6) is set up to implement the method according to one of the preceding To carry out claims. Motor vehicle with a computer program product Claim 12 .

Images