DE102018125712A1 - Driving support method for a vehicle - Google Patents

Abstract

The present invention relates to a driving support method for a vehicle, comprising the method steps: a) creating a first data record which has a first functional logic, the first functional logic in particular continuously generating a first control command for a reaction-demanding first driving situation of the vehicle; b) executing the first Function logic using the first control command; andc) using data relating to the first functional logic carried out in method step b) to carry out a teaching step in the context of a machine-learning driving support method while generating a second functional logic, the second functional logic in particular continuously generating a second control command for the reaction-demanding first driving situation of the vehicle.

Description

The present invention relates to a driving support method for a vehicle. The present invention relates in particular to a method with which a high safety standard can be achieved with reference to the issue of control commands based on driving situations of the vehicle. The present invention further relates to a driving support system that is set up to carry out such a method.

A variety of driving support systems are known per se. It is often necessary or a requirement to carry out an environmental survey. This enables a comprehensive picture of the vehicle environment and can provide reliable information on selecting and executing driving strategies. The driving strategies can be based on the determined vehicle environment or based on specific driving situations.

In the case of driving support systems, it is necessary not only to comply with functional safety requirements, such as with electronic errors, but also with regard to functional errors (SOTIF, safety often he intended functionality). In other words, driving support systems should work permanently and without failures and, furthermore, avoid errors with regard to their function, that is to say the output of data or the execution of driving assistance.

This problem is often solved by identifying dangerous situations, defining maximum control steps with regard to acceleration forces and also defining scenarios for the interaction between system-related driving interventions and driver interventions. It is also common to have the same driver assistance function developed by two completely independent development teams.

Such solutions known from the prior art can offer still further potential for improvement, in particular with regard to the effective issuing of control commands based on existing driving situations, a high safety standard being achievable.

It is the object of the present invention to at least partially overcome the disadvantages known from the prior art. It is in particular the object of the present invention to provide a solution by means of which it is possible to issue control commands based on existing driving situations in an effective and safe manner.

The task is solved according to the invention by a method with the features of claim 1. The task is further solved according to the invention by a computer program product with the features of claim 9. The task is further solved according to the invention by a driving support system with the features of claim 10. Preferred Embodiments of the invention are described in the subclaims, in the description or in the figures, wherein further features described or shown in the subclaims or in the description or in the figures, individually or in any combination, can constitute an object of the invention, if derived from the context does not clearly show the opposite.

1. a) Erstellen eines ersten Datensatzes, der eine erste Funktionslogik aufweist, wobei die erste Funktionslogik insbesondere kontinuierlich einen ersten Steuerbefehl auf eine reaktionsfordernde erste Fahrsituation des Fahrzeugs erzeugt;
2. b) Ausführen der ersten Funktionslogik unter Anwendung des ersten Steuerbefehls; und
3. c) Verwenden von Daten betreffend die in Verfahrensschritt b) ausgeführte erste Funktionslogik zum Durchführen eines Anlernschrittes im Rahmen eines maschinenlernenden Fahrunterstützungsverfahrens unter Erzeugung einer zweiten Funktionslogik, wobei die zweite Funktionslogik insbesondere kontinuierlich einen zweiten Steuerbefehl auf die reaktionsfordernde erste Fahrsituation des Fahrzeugs erzeugt.

A driving assistance method for a vehicle is proposed, comprising the method steps:

1. a) creating a first data record which has a first functional logic, the first functional logic in particular continuously generating a first control command for a reaction-demanding first driving situation of the vehicle;
2. b) executing the first functional logic using the first control command; and
3. c) Use of data relating to the first functional logic carried out in method step b) to carry out a teaching step in the context of a machine-learning driving support method while generating a second functional logic, the second functional logic in particular continuously generating a second control command for the reaction-demanding first driving situation of the vehicle.

Such a method can have significant advantages over the solutions from the prior art, in particular with regard to the effective and safe issuing of control commands based on an existing driving situation.

The method described here is thus a driving support method for a vehicle. The method can thus be carried out, for example, at least partially or even completely by a driving support system which is part of the vehicle. In particular, the vehicle can be a motor vehicle. The method can relate to fully autonomous driving, so the vehicle can be carried out without driver intervention. Alternatively, it is possible that the method merely supports the driver, that is, outputs driving information, for example, the driver having to carry out certain driving interventions himself.

Alternatively, the method can be designed to be carried out separately from a driving support system. In particular, it can be used to create at least a first and a second functional logic on which the function of a driving support system can be based and which can be implemented in a driving support system after the method has been carried out.

In order to achieve this, the process comprises the following process steps.

First, according to method step a), a first data record is created which has a first functional logic, the first functional logic generating a first control command for a reaction-demanding first driving situation of the vehicle.

According to this method step, a data record can thus be created which has a first functional logic, for example can consist of this. A functional logic is to be understood as a logic or algorithm of this type, which, based on a defined driving situation, can issue a control command corresponding to the driving situation. Such functional logic can be created, for example, in the conventional way by programming the logic, or also by real driving operation. It can also be a component of this procedural step that the functional logic is validated, for which the corresponding data can be inserted, both virtually in a simulation as pure computer technology, as well as in test operation and in real road traffic.

For an application in a driving support system, the creation of the functional logic can therefore mean that sensor data and other vehicle data at an appropriate level, such as at the least preprocessed level, are used as input and the control variables, such as steering wheel movement, acceleration, braking, as control commands to learn accordingly. In principle, any functionality of the driving support system can be learned if the amount of data used is sufficient and validated accordingly.

The function logic also serves in particular to preferably continuously generate and output a control command for a reaction-demanding driving situation. In the context of the present invention, a reaction-demanding driving situation can be understood to mean, in particular, a driving situation in which intervention, for example by a brake intervention or a steering intervention, is necessary or preferred to avoid or improve a critical situation, or whether another operation is preferred to put a turn signal. In principle, a driving situation that requires reaction can be any driving situation that requires a reaction or an action. A continuous output of control commands can furthermore include, in particular, an uninterrupted output of successive identical or different control commands or can comprise output of successive control commands, the output of the control commands being interrupted.

Furthermore, the term control command in the sense of the present invention is to be understood in particular that the control command need not be a simple digital or analog control command, such as an acceleration. In the sense of the invention, the control command rather comprises control profiles, that is to say acceleration profiles comprising a lateral and a longitudinal component, forming a trajectory over time, that is to say with a suitable speed profile. These complex control commands or control command profiles or control profiles are compared with one another as described.

According to the further method step b), the method further comprises executing the first functional logic using the first control command. Thus, be it computer-simulated or under real conditions, the functional logic created in method step a) can be used by executing the first control command based on the first functional logic when defined driving situations occur.

Based on this step, further process step c) uses data relating to the first functional logic carried out in method step b) to carry out a teaching step in the context of a machine-learning driving support method with the generation of a second functional logic, the second functional logic in turn preferably continuously issuing a second control command generates the reaction-demanding first driving situation of the vehicle.

This method step thus comprises using the first functional logic that has been created and, in particular, already validated, in order to obtain a second functional logic as part of a machine-learning driving support method. The second function logic in turn comprises or consists of creating a second control command, the second control command being based on the driving situation for which the second function logic was taught. In particular, for the same driving situation, a first control command is created in accordance with method step a) and one on one in accordance with method step c) machine-learning method based second control command created.

At the end of the development or the method, two functions or two functional logics are thus obtained with little additional effort, which reflect the same behavior of the driving support system or can provide the control commands for the same or comparable driving situations. A function or a functional logic is thus developed and this is used to train a learning network in such a way that the network or the machine-learning method imitates the same function. This makes it possible in a simple manner to achieve a high security standard by means of additional security mechanisms, for example by checking both control commands against one another and taking appropriate measures if the deviation is too great, as will be described in greater detail below.

It can be particularly advantageous that method step c) is carried out using a machine-learning method or using artificial intelligence. This is because, in particular using an algorithm based on artificial intelligence or a functional logic based on artificial intelligence, a particularly reliable result of a control command or a particularly reliable functional logic can be obtained. Using learning methods, it is even possible to derive driving strategies only on the basis of collected and interpreted sensor data or raw driving state data.

With regard to the artificial intelligence, a neural network can be integrated in a manner known per se into the vehicle or into the driving support system or also into a processor at least partially used by the vehicle or the driving support system. This can be designed and set up as is known in principle to the person skilled in the art. It can then be possible to upload sensor data or driving situations to a server, such as a central or back-end server, or to transfer them to the memory of the driving support system or, in principle, of the processor. The relevant data can be evaluated there, in particular by machine or human, and added to the training database. The neural networks in the driving support systems or in other processors can then be updated accordingly, so they learn what can significantly increase the safety of the functional logic.

For example, with regard to uploading to a server, comparisons can be made with comparable driving situations that have already been classified by other vehicles or their drivers, so that learning by adding information from a wide variety of vehicles and thus based on an extraordinarily large number of classified objects can be done. This can be possible using a conventional data connection and can significantly increase the classification accuracy.

The functional logics created in this way can then be used in a driving support method in order to enable assisted driving, up to autonomous driving.

For this purpose, sensor data are generated by at least one environment detection sensor, which reflect the surroundings of the vehicle and which can thus identify specific driving situations. In addition to environment detection sensors, it is also possible to use sensors that detect, for example, the driving behavior of the vehicle. In principle, these steps can be carried out, for example, as is known for driving support systems, for example in order to observe the surroundings of the vehicle and to detect objects and to detect the state of the vehicle. The raw data of the sensor or sensors can be used for the further method, for example, or the data used can also be preprocessed in a suitable manner.

Accordingly, a driving situation can be recorded and, based on the functional logics, i.e. on the basis of the first functional logic as well as the second functional logic, a control command for each of the functional logics can be generated and output according to the driving situation that occurs so that the vehicle can react correctly.

It is therefore possible to receive two control commands for the same or for comparable situations. In this way, a verification of the first control command or a combination of both control commands can result in a significantly higher security level than the creation of only one control command.

Instead of using a conventional "end-to-end learning" approach to learn a function that can then be used in a driving support system, the method described here instead uses a method in which a function or a first functional logic is used, for example, in the conventional way can be created and this created and in particular validated first functional logic, for example during the Development and / or testing can teach a machine-learning functionality or a learning network. In doing so, a second function logic is created which functionally affects the same behavior as the first function or the first function logic itself.

The method described here makes it possible to achieve a safety standard in accordance with ASIL (Automotive Safety Integrity Level) or, if necessary, also in accordance with SOTIF (Safety Of The Intended Functionality, safety of the target function), which are generally recognized safety standards in accordance with ISO 26262.

In addition, the method described here also enables such driving situations to be countered safely or with a reliable control signal which are difficult to grasp using conventional methods. This is because such situations, which can also be referred to as borderline situations, can be detected or created by teaching using the first functional logic. Even if these were not recorded at the time of training. This can preferably apply to online systems.

Furthermore, the implementation of an improved security standard can be made possible by the method described here with reduced expenditure compared to methods known in the prior art.

It may be preferred that the machine-learning driving support method is based on a neural network. In particular, using a neural network, a large amount of data can be created and processed, so that the security of the second functional logic can be particularly high. In addition, neural networks, such as the networks known as DNN or CNN, are well known, so that they can be easily implemented or the method described here can be easily implemented by neural networks.

Additionally or alternatively, it can be provided that the first functional logic is based on a neural network. In this embodiment, the features or advantages of the neural network described above can also be used to create the first functional logic, which can also significantly improve the first functional logic and make driving support based thereon particularly safe.

• d) Verwenden von Daten betreffend die zweite Funktionslogik zum Durchführen eines Anlernschrittes im Rahmen eines maschinenlernenden Fahrunterstützungsverfahrens unter Erzeugung einer dritten Funktionslogik, wobei die dritte Funktionslogik einen dritten Steuerbefehl auf die reaktionsfordernde erste Fahrsituation des Fahrzeugs erzeugt.

It may further be preferred that the method has the further method step:

• d) using data relating to the second functional logic to carry out a learning step in the context of a machine-learning driving support method, generating a third functional logic, the third functional logic generating a third control command for the reaction-demanding first driving situation of the vehicle.

In this embodiment, not only is second function logic provided based on a first function logic, but a third function logic is additionally provided by a machine-learning method, reference being made to the second function logic with regard to the respective features of the provision and configuration of the third function logic.

In this embodiment, for example, a second neural network or, depending on the configuration of the first functional logic, a third neural network can be taught in by changing the input side, for example by using partially pre-filtered data as an input. This teaches in another network in the same way and even gives you the option of making a 1 out of 3 decision when comparing the control command based on the function logic. This is possible with a development effort that is approximately 1.5 times normal.

In this embodiment, the provision of three functional logics based on the same driving situation can therefore be used to issue a particularly safe control command which is based on three control commands, which likewise makes the corresponding driving support particularly safe and robust.

Furthermore, it may be preferred that the first control command and the second control command and possibly the third control command are compared with one another when the reaction-demanding driving situation is present, when at least two control commands of the first control command, the second control command and possibly the third control command differ from one another by a more than a predefinable factor, a fourth control command is issued, which is different from the first control command, the second control command and possibly the third control command.

In this embodiment, the corresponding control commands can thus be compared with one another. If there is an excessive deviation of at least two control commands from one another, it can be assumed that at least one control command is incorrect, which can lead to those based on the function logics Control commands are not executed. The predefinable range can be selected based on the driving situation, for example, and can generally be selected based on the respective conditions. Instead of the control commands based on the function logic, a fourth control command can then be carried out, which is different from the control commands based on the function logic, in order to avoid a critical situation in a safe manner.

For example, it can be provided that the fourth control command includes transferring the vehicle to a safe state. Such a maneuver can also be referred to as a minimal risk maneuver and is generally known for driving support systems by using the minimally riskiest control command according to the situation.

Furthermore, it can be preferred that when the reaction-demanding driving situation is present, again computer-based or under real conditions, the first control command and the second control command and possibly the third control command are compared with one another, when all of the first control command, the second control command and if necessary, the third control command differ from one another less than a predefinable factor, a fifth control command is output, which is a combination of the first control command, the second control command and, if appropriate, the third control command.

Alternatively, in the above-described conditions, i.e. if all of the first control command, the second control command and possibly the third control command differ from one another less than a predefinable factor, it can be provided that one of the determined control commands is output which corresponds to suitably defined criteria, such as the safest trajectory, the most comfortable trajectory or the like exhibits the best behavior. In other words, the control commands can be compared under these conditions and the suitable control command can be used or output according to predefinable criteria.

In this embodiment, the corresponding control commands can in turn be compared with one another. If there is a slight deviation of all control commands from one another, it can be assumed that the control commands are correct, which is why they can be executed. The predefinable range can in turn be selected based on the driving situation and can in principle be selected on the basis of the respective conditions. Instead of the control commands based on the function logics, a fifth control command can then be carried out, which may be different from the control commands based on the function logics. For example, the fifth control command can be an average of the first to third control commands issued. Alternatively, the first control command, the second control command and possibly the third control command can be subjected to a suitable weighting.

Furthermore, it can be preferred that at least one of steps c) and d) takes place at least partially using data stored in a memory that is separate from the vehicle. In this embodiment, steps c) and / or d) can thus take place online, and in particular using wireless communication, such as an Internet connection or another connection. This makes it possible to teach-in not only on the basis of data from one's own vehicle, but also to use data from a large number of vehicles. This allows an incomparably larger number of driving situations and a corresponding reaction to be available so that teaching can be based on a particularly large number of data. This makes it possible to base the second or third function logic on a particularly large amount of data, which makes driving support based on this particularly reliable and robust.

With regard to further advantages and features of the method, reference is made to the description of the computer program product, the driving support system and also to the figure and the description of the figure, and vice versa.

Also described is a computer program product for issuing control commands for a driving support system of a vehicle, the computer program product being designed to carry out a method at least in part, for example completely, as described above.

In particular, it is provided in the computer program product described here that it comprises program parts which are loaded in a processor of a computer-based evaluation device, such as a driving support system, for carrying out the aforementioned method.

Thus, the computer program product has programs, or at least program parts, which can be loaded in a processor of a computer-based evaluation device and can be stored, for example, in a memory, which can be part of a driving support system. Based on these programs or program parts, it can enable the above-described method for issuing control commands for a vehicle to be carried out.

Such a method creates a simple and effective solution for the safe and robust issuance of control commands. As a result, a statement regarding the selection and / or implementation of driving strategies of a vehicle can be made safely and efficiently.

With regard to further advantages and features of the computer program product, reference is made to the description of the method, the driving support system and also to the figure and the description of the figure, and vice versa.

Also described is a driving support system for a vehicle for issuing a control command to a reaction-demanding driving situation, the driving support system having at least one environment detection sensor for environment detection and further comprising a control unit that can be fed with data from the at least one environment detection sensor, the control unit being designed to based on the data of the at least one environment detection sensor, to recognize whether the detected driving situation is a reaction-demanding driving situation, characterized in that the control unit is designed to carry out a method as described above.

The driving support system may be part of a vehicle, such as a car. The driving support system can be provided, in particular, for generating environmental data as the basis for driving support, such as for generating a driving trajectory. The driving support system is used in particular to issue a control command to a reaction-demanding driving situation.

In order to detect this, the driving support system has at least one environment detection sensor. The environment detection sensor can preferably be part of the driving support system that also executes the method. The at least one environment detection sensor can be, for example, any suitable sensor known in the art. Furthermore, only one or more different or identical sensors can be used to monitor the environment, which form a sensor arrangement from one or more environment sensors or environment detection sensors. The at least one environment detection sensor can, for example, have or consist of one or more identical or different sensors that are selected from the group consisting of LIDARs, for example comprising laser scanners, ultrasound sensors, cameras or radar-based sensors or other sensors that are used in the field for monitoring the environment are known.

A control unit, such as a processor, is also provided. The control unit is designed to evaluate the sensor data or measurement data supplied by the sensor or sensors. The control unit can thus be fed with data from the at least one environment detection sensor. It is further provided that the control unit is designed to recognize, based on the data of the at least one environment detection sensor, whether the detected driving situation is a reaction-demanding driving situation. This can be carried out by means of appropriate algorithms in a manner known per se, in that the sensor data are evaluated to determine whether a driver intervention is necessary in order to improve or avoid a critical situation or whether another action is required. For this purpose, a computer program product, for example, can be loaded into the control unit, as described above.

The driving support system described here can thus make it possible to provide a simple and effective solution for issuing control commands, which allows driving support in a particularly safe and robust manner.

With regard to further advantages and features of the driving support system, reference is made to the description of the method, the computer program product and also to the figure and the description of the figure, and vice versa.

Further advantages and advantageous configurations of the objects according to the invention are illustrated by the drawing and explained in the following description. It should be noted that the drawing is only descriptive and is not intended to limit the invention in any way. It shows 1 is a schematic representation of a diagram illustrating an embodiment of a method according to the invention for issuing control commands.

In the 1 is a schematic representation of a diagram showing an embodiment of a method according to the invention for issuing control commands. In particular, the method described here is thus a driving support method for a vehicle. This has the following steps.

First, represented by the block 10th , creating a first record, the one has first function logic, the first function logic generating a first control command to a reaction-demanding first driving situation of the vehicle. This can take place, for example, in a conventional manner and / or, for example, based on a neural network. This functional logic can already flow into a control unit of a driving support system or can be used there, which control unit is implemented by the block 12th is represented like this by the arrow 22 is shown. A first control command can be generated in this way.

Furthermore, this is done by the block 14 characterized by executing the first function logic using the first control command for the first driving situation, using data relating to the executed first function logic to carry out a teaching step in the context of a machine-learning driving support method while generating a second function logic, the second function logic executing a second control command generates the reaction-demanding first driving situation of the vehicle. This step can thus be used to generate a further functional logic, which in turn relates to a control command for the previously existing driving situation, as indicated by the arrow 24th is shown.

Again, this step can be used using a neural network. The functional logic generated in this way can in turn be used in the control unit by issuing a second control command, as indicated by the arrow 26 is shown.

Optionally, it is possible that the procedure is as described by the block 16 The further method step comprises the use of data relating to the second functional logic for carrying out a teaching step in the context of a machine learning driving support method while generating a third functional logic, the third functional logic generating a third control command for the reaction-demanding first driving situation of the vehicle. This is through the arrow 28 shown. The third functional logic generating the third control command can in turn be carried out in the control unit, as indicated by the arrow 30th is shown.

If so, like this through the block 18th a specific driving situation is to be detected using an environment detection sensor or this is recognized by the control system, as is indicated by the arrow 32 is shown, the control system can issue a final control command, as indicated by the block 20th or the arrow 34 is shown.

In this regard, it can be provided, for example, that the first control command and the second control command and, if appropriate, the third control command are compared with one another when the reaction-demanding driving situation is present, when at least two control commands of the first control command, the second control command and, if appropriate, the third control command are compared deviate by a predefinable factor, a fourth, by the block 20th shown, control command is issued, which is different from the first control command, the second control command and possibly the third control command. It can be provided that the fourth control command includes transferring the vehicle to a safe state.

Alternatively, when the reaction-demanding driving situation is present, again computer-based or under real conditions, the first control command and the second control command and possibly the third control command can be compared with one another, when all of the first control command, the second control command and possibly the third control command from one another deviate less than a predefinable factor, a fifth, again by the block 20th shown, control command are output, which is a combination of the first control command, the second control command and possibly the third control command.

Thus, the method described here in particular allows the behavior of a driving support system to be learned “end to end”, which can be implemented either offline or online, with development and testing being able to take place in the control unit, for example. The basic principle of the applied "end to end" learning is that the learning algorithms learn the behavior that you want to develop, which can later be used as a training data record with the learning algorithms e.g. neural networks such as DNN, CNN, and many others together produce the functional software and reproduce the learned behavior as best as possible.

Reference list

10th

block

12th

block

14

block

16

block

18th

block

20th

block

22

arrow

24th

arrow

26

arrow

28

arrow

30th

arrow

32

arrow

34

arrow

Claims (10)

Driving support method for a vehicle, comprising the procedural steps: a) creating a first data record which has a first functional logic, the first functional logic in particular continuously generating a first control command for a reaction-demanding first driving situation of the vehicle; b) executing the first functional logic using the first control command; and c) Use of data relating to the first functional logic carried out in method step b) to carry out a teaching step in the context of a machine-learning driving support method while generating a second functional logic, the second functional logic in particular continuously generating a second control command for the reaction-demanding first driving situation of the vehicle. Procedure according to Claim 1 , characterized in that the machine learning driving support method is based on a neural network. Procedure according to Claim 1 or 2nd , characterized in that the first functional logic is based on a neural network. Procedure according to one of the Claims 1 to 3rd , characterized in that the method has the further method step: d) using data relating to the second function logic to carry out a teach-in step as part of a machine-learning driving support method with the generation of a third function logic, the third function logic in particular continuously giving a third control command to the reaction-demanding first driving situation of the vehicle. Procedure according to one of the Claims 1 to 4th , characterized in that when the reaction-demanding driving situation is present, the first control command and the second control command and, if appropriate, the third control command are compared with one another, and if at least two control commands of the first control command, the second control command and, if appropriate, the third control command differ from one another by more as a predefinable factor, a fourth control command is issued, which is different from the first control command, the second control command and possibly the third control command. Procedure according to Claim 5 , characterized in that the fourth control command comprises bringing the vehicle into a safe state. Procedure according to one of the Claims 1 to 4th , characterized in that in the presence of the reaction-demanding driving situation, the first control command and the second control command and possibly the third control command are compared with one another, when all of the first control command, the second control command and possibly the third control command differ from one another by less than a predefinable factor, a fifth control command is output, which is a combination of the first control command, the second control command and possibly the third control command Procedure according to one of the Claims 1 to 7 , characterized in that at least one of steps c) and d) takes place at least partially using data stored in a memory separate from the vehicle. Computer program product for issuing control commands for a driving support system of a vehicle, the computer program product being configured, a method according to one of the Claims 1 to 8th to perform at least in part. Driving support system for a vehicle for issuing a control command to a reaction-demanding driving situation, the driving support system having at least one environment detection sensor for environment detection and further comprising a control unit that can be fed with data from the at least one environment detection sensor, the control unit being designed based on the data of the at least one environment detection sensor to recognize whether the detected driving situation is a reaction-demanding driving situation, characterized in that the control unit is designed to implement a method according to one of the Claims 1 to 8th to execute.

Images