Classification of AI modules

Abstract

The invention relates to a method for providing a classifier for an AI module for processing input data provided by a sensor system of a motor vehicle, a computer program having instructions and a device, and a classifier provided using such a method. Furthermore, the invention relates to a method for configuring a control system of a motor vehicle using a library of AI modules for processing input data provided by a sensor system of a motor vehicle, a computer program having instructions and a device, and a motor vehicle using such a method or such a device. In a first step, an AI module to be classified is selected (10). Furthermore, a suitable test data set is selected (11). Then, the AI module is applied (12) to data points of the test data set. Here, for these data points, the associated ground truth and context parameters are known. Subsequently, a functional quality is determined (13) for each of the data points based on the output of the AI module. Finally, a classifier is created (14) for the AI module, which outputs the functional quality for a given context parameter.

Description

Technical Field

[0001] The present invention relates to a method for providing a classifier for an AI module for processing input data provided by a sensor system of a motor vehicle, a computer program with instructions, and a device, as well as a classifier provided by such a method. Furthermore, the present invention relates to a method for configuring a control system of a motor vehicle using a program library of an AI module for processing input data provided by a sensor system of a motor vehicle, a computer program with instructions, and a device, as well as a motor vehicle using such a method or such a device. Background Art

[0002] Currently, (highly) automated driving cannot be achieved without the use of methods based on artificial intelligence (AI), in particular without image data processing based on deep neural networks. However, even if AI models are designed to solve the same task, these AI models are very diverse and also extremely different in terms of their functional quality. Here, the term "functional quality" describes the quality or quality of an AI module in terms of the function or task it provides. In current solutions, AI modules are mainly distinguished based on the architecture of the AI ​​model and the data used in training. Here, attempts are usually made to compensate for the lack of functional quality by increasing the training data or increasing the complexity of the architecture.

[0003] Simple solutions for using artificial intelligence in motor vehicles are limited to using a single AI module or a group of AI modules and their optimization. Here, an AI module should be understood as a software module used to implement an AI model.

[0004] Against this background, DE 10 2017 006 599 A1 describes a method for operating a motor vehicle which is driven at least in a partially automated manner. In this method, at least three artificial neural networks are trained in mutually disjoint training drives of the motor vehicle independently of one another, respectively by means of an end-to-end approach, on the basis of training actuator data of the vehicle actuators recorded during training drives, and on the basis of training sensor data of the vehicle sensors which are relevant to the training actuator data. In the at least partially automated operation of the motor vehicle, actual sensor data are detected as input data for the neural networks, and the actual sensor data are associated with the training actuator data as output data for the neural networks by means of a comparison with the training sensor data. The training actuator data of all neural networks are fed to a merging module which merges the training actuator data of all neural networks in accordance with pre-given rules and determines actual actuator data on the basis of the result of the merging. The motor vehicle is controlled laterally and / or longitudinally at least in a partially automated manner on the basis of a control of the vehicle actuators with the determined actual actuator data.

[0005] Furthermore, DE 10 2017 107 837 A1 describes an adjustable sensor device. This sensor device comprises at least one sensor element with a control and analysis unit. The sensor data are analyzed with a classifier. The classifier has a neural network. The sensor device can be connected via an interface to a superior computer network. The control and analysis unit is configured for an extension function. To this end, the sensor device transmits selected sensor data which are to be additionally processed by the classifier in the extension function to the computer network. There, the classifier is trained on the basis of the sensor data, and after the end of the training, the sensor device retrieves classifier data for a correction of the classifier, for example parameters, program segments or even the entire trained classifier. The sensor device is thus also equipped for a classification of the selected sensor data.

[0006] In contrast, in another approach, the focus is not on developing a function which works well and reliably in all cases. Rather, a combination of good functions is combined, that is to say, different AI modules are dynamically combined with one another on the vehicle side. SUMMARY

[0007] The technical problem addressed by the present application is to provide a solution which supports a dynamic combination of AI modules which can be used.

[0008] The above-mentioned technical problem is solved by a method having the features of the present application, by a computer-readable storage medium according to the present application, by a device having the features of the present application and by a classifier according to the present application. Preferred design solutions of the present application are the subject of the following description.

[0009] According to a first aspect of the present application, a method for providing a classifier for an AI module, the AI module being used for processing input data provided by a sensor system of a motor vehicle, comprises the following steps:

[0010] applying the AI module to two or more data points in a test data set, wherein for the two or more data points the associated ground truth and context parameters are known;

[0011] determining a functional quality for each of the two or more data points; and

[0012] creating a classifier for the AI module, the classifier outputting the functional quality for a given context parameter.

[0013] According to a further aspect of the present application, a computer program comprises instructions which, when executed by a computer, cause the computer to perform the following steps for providing a classifier for an AI module, the AI module being used for processing input data provided by a sensor system of a motor vehicle:

[0014] applying the AI module to two or more data points in a test data set, wherein for the two or more data points the associated ground truth and context parameters are known;

[0015] determining a functional quality for each of the two or more data points; and

[0016] creating a classifier for the AI module, the classifier outputting the functional quality for a given context parameter.

[0017] The term "computer" is to be understood broadly herein. In particular, a computer also comprises a workstation, a distributed system, and other processor-based data processing devices.

[0018] The computer program may, for example, be provided for electronic download, or the computer program may be stored on a computer-readable storage medium.

[0019] According to a further aspect of the present application, an apparatus for providing a classifier for an AI module, the AI module being used for processing input data provided by a sensor system of a motor vehicle, has the following modules:

[0020] a test module for applying the AI module to two or more data points in a test data set, wherein for the two or more data points the associated ground truth and context parameters are known; and

[0021] - an analysis module for determining the functional quality for each of the two or more data points and creating a classifier for the AI module, the classifier outputting the functional quality for a given context parameter.

[0022] When using AI modules or AI models implemented by these AI modules, it should be noted that there is a correlation between the functional quality of the AI model and the data processed by the AI model. This correlation ensures that the AI model itself is not good or bad, but has a context-dependent functional quality. By means of the solution according to the application, a convincing description of the AI model and its capabilities can be created in the test phase. Here, a list of the most convincing properties can be determined as a measure of the performance of the AI module. Thus, for example, the AI model can not only be better understood compared to a group of expert models, but also be used more diversely.

[0023] In the course of the test phase, the classification system evaluates the AI modules with respect to the expected functional quality of the AI modules with respect to the respective context parameters on the basis of a set of context dimensions, a set of AI modules and a set of test data. The context parameters can comprise, for example, properties in the context of the data points or properties of the architecture of the AI modules. Here, the term "data point" is to be understood as input data for a given case. For the test data, the ground truth is known, that is to say, there is a correct result for the respective input data. Furthermore, for the test data, the context parameters for evaluating the data in terms of the context dimensions are known. By means of the context association, the resulting classification becomes traceable, testable and secure. The resulting classifier is configured to classify the AI modules in terms of the expected functional quality on the basis of the context parameters of the input data.

[0024] According to one aspect of the application, the AI module implements one AI model or a family of AI models in the sense of a group. Usually, one AI model, for example a neural network, is implemented in one AI module, which is classified in the course of the test phase. However, the solution according to the application can also be used to determine an experience-based weighting function for joint inference of a family of AI models in the sense of a group, that is to say, of a collective of AI models that process the same input data.

[0025] According to an aspect of the application, for determining the functional quality for a data point, the output of the AI module for the data point is compared with the associated ground truth. For this, for example, the IoU measure (IoU: Intersection over Union; ratio between the intersection and the union, also called Jaccard coefficient) can be used. By comparison with the ground truth, the functional quality can be determined in a simple manner. Here, it has proven to be reliable to use the IoU measure, in particular in AI modules for object recognition.

[0026] According to an aspect of the application, the classifier is formed by a neural network. This has the advantage that the classifier can be trained in the test phase without prior knowledge of the importance of the context parameters. However, the classifier can also be implemented by other functions.

[0027] Preferably, the classifier for the AI module is provided by means of the method according to the application. By implementing such a classifier for given input data, a situation-dependent evaluation of AI modules can be carried out, which can be used for processing the input data.

[0028] According to a further aspect of the application, a method for configuring a control system of a motor vehicle using a library of AI modules for processing input data provided by a sensor system of the motor vehicle comprises the following steps:

[0029] - detecting input data to be processed by the AI modules;

[0030] - evaluating the AI modules on the basis of context parameters; and

[0031] - determining the AI module or the combination of AI modules to be used for the input data and the associated weights.

[0032] According to a further aspect of the application, a computer program comprises instructions which, when executed by a computer, cause the computer to carry out the following steps for configuring a control system of a motor vehicle using a library of AI modules for processing input data provided by a sensor system of the motor vehicle:

[0033] - detecting input data to be processed by the AI modules;

[0034] - evaluating the AI modules on the basis of context parameters; and

[0035] - determining the AI module or the combination of AI modules to be used for the input data and the associated weights.

[0036] The term "computer" is to be understood broadly here. In particular, a computer also comprises a control device and other processor-based data processing devices.

[0037] The computer program can be provided for example for electronic downloading, or the computer program can be stored on a storage medium.

[0038] The device for configuring a control system of a motor vehicle with a library of AI modules for processing input data provided by sensors of the motor vehicle has the following modules:

[0039] a data module for detecting input data to be processed by the AI modules;

[0040] a classifier for evaluating the AI modules on the basis of context parameters; and

[0041] an analysis module for determining an AI module or a combination of AI modules to be used for the input data and associated weights.

[0042] By means of the solution according to the application, conditions or context parameters are determined in a test phase which have the most significant influence on the functional properties of the AI modules, for example deep neural networks, which can be used for use in automated driving. The classifier now uses these previously determined conditions to determine a combination of a plurality of AI modules which is particularly effective for a given input data in order to increase the efficiency of the overall system.

[0043] It is particularly advantageous to use the method according to the application or the device according to the application in a motor vehicle. The use of the described solution is particularly meaningful when a level of autonomy (Autonomiestufe) or automated driving of level 4 or 5 is to be implemented. In this context, the AI modules can in particular be configured for environmental recognition for the automated driving function of the motor vehicle.

[0044] Different AI modules can for example be matched to different light conditions, different vehicle speeds, different vehicle environments, different driving situations, different environmental conditions, different driving conditions or different target settings.

[0045] Matching to different light conditions is particularly advantageous for environmental recognition for the automated driving function in the event of a change in the light conditions. For example, the light conditions can change due to a change in the weather, driving into or out of a tunnel or very short twilight near the equator, for example. For different light conditions, expert systems can accordingly be provided as AI modules.

[0046] Matching to different speeds, for example, is of course used for 3D object recognition. Here, it can make sense to provide the AI module as an expert system for different speeds of the vehicle, for example, for the case that the vehicle drives into a street with a speed limit that is significantly different from the speed limit of the street previously driven.

[0047] In matching to different vehicle environments, for example, a distinction can be made between urban and rural environments, but also between the vehicle being located in the city center or near a school or hospital.

[0048] In matching to different driving situations, the AI module can be set up, for example, for driving on a motorway, in a parking lot, in the case of traffic congestion or for complex crossroads with the aid of a specialized Car2X infrastructure.

[0049] With regard to matching to different environmental situations, the AI module can be set up, for example, for special weather conditions, light conditions, traffic density, road type, time of day or geographical location. In this context, the pedestrian density can also be noted. Thus, in particular, an expert system for streets with high pedestrian density and an expert system for recognizing pedestrians at different distances can be provided. In the case of high pedestrian density, the expert system must be able to detect the intentions of pedestrians in the vicinity of the vehicle. In the case of low pedestrian density, it is usually possible to drive faster. Here, it is also important to recognize pedestrians at a distance as early as possible. In contrast, the intentions of these pedestrians are less important.

[0050] The driving behavior of the autonomous driving function can be adjusted to different driving conditions, for example, speed, vehicle type, presence of a trailer, journey parameters or preferences of the vehicle occupants, by using matching AI modules.

[0051] Different target settings can result from legal or environment-related boundary conditions. Thus, for example, AI modules for low-noise or low-emission driving can be provided. Other artificial intelligence modules can be matched to special behavior rules.

[0052] In all these examples, it makes sense to determine the AI module to be used and the associated weights for processing the input data on the basis of context parameters. BRIEF DESCRIPTION OF DRAWINGS

[0053] Further features of the present invention will become clear from the following description.

[0054] Figure 1 A method for providing a classifier for AI modules for processing input data provided by a sensor system of a motor vehicle is schematically shown;

[0055] Figure 2A first embodiment of a device for providing a classifier for an AI module for processing input data provided by a sensor system of a motor vehicle is shown;

[0056] Figure 3 A second embodiment of a device for providing a classifier for an AI module for processing input data provided by a sensor system of a motor vehicle is shown;

[0057] Figure 4 A method for configuring a control system of a motor vehicle with a library of AI modules for processing input data provided by a sensor system of the motor vehicle is schematically shown;

[0058] Figure 5 A first embodiment of a device for configuring a control system of a motor vehicle with a library of AI modules for processing input data provided by a sensor system of the motor vehicle is shown;

[0059] Figure 6 A second embodiment of a device for configuring a control system of a motor vehicle with a library of AI modules for processing input data provided by a sensor system of the motor vehicle is shown;

[0060] Figure 7 A motor vehicle implementing the solution according to the present application is schematically shown;

[0061] Figure 8 A system diagram of the solution according to the present application for providing a classifier for an AI module for processing input data provided by a sensor system of a motor vehicle is schematically shown; and

[0062] Figure 9 A system diagram of the solution according to the present application for configuring a control system with a library of AI modules is schematically shown. DETAILED DESCRIPTION

[0063] For a better understanding of the principles of the present application, embodiments of the present application are explained in detail below with reference to the drawings. It is to be understood that the present application is not limited to these embodiments, and that features described can be combined or modified without departing from the scope of the present application.

[0064] Figure 1A method for providing a classifier for an AI module for processing input data provided by a sensor system of a motor vehicle is schematically shown. In a first step, 10, the AI module to be classified is selected. The AI module for example implements one AI model or an ensemble of AI models. Further, a suitable test data set is selected, 11. Then, the AI module is applied, 12, to data points of the test data set. Here, for these data points, the associated ground truth and context parameters are known. The context parameters for example can include properties of the context of the data points or properties of the architecture of the AI module. Subsequently, a functional quality is determined, 13, for each of the data points based on the output of the AI module. For this, for example, the IoU metric can be used, compared to the respective associated ground truth. Finally, a classifier is created, 14, for the AI module, which outputs the functional quality for a given context parameter. The classifier for example can be formed by a neural network.

[0065] Figure 2 A simplified schematic diagram of a device 20 for providing a classifier for an AI module for processing input data provided by a sensor system of a motor vehicle is shown for a first embodiment. The AI module for example implements one AI model or an ensemble of AI models. The device 20 has an input 21 via which data of a test data set for example can be received. However, such a test data set also can be saved in a database 22 of the device 20. A test module 23 causes the selected AI module to be applied to data points of the selected test data set. Here, for these data points, the associated ground truth and context parameters are known. The context parameters for example can include properties of the context of the data points or properties of the architecture of the AI module. Subsequently, an analysis module 24 determines a functional quality for each of the data points based on the output of the AI module. For this, the analysis module 24 for example can use the IoU metric, compared to the respective associated ground truth. Further, the analysis module 24 creates a classifier for the AI module, which outputs the functional quality for a given context parameter. For this, the analysis module 24 for example can access the classifier K via an output 27 of the device 20. Alternatively, the creation of the classifier K also can be performed by another independent module. The classifier for example can be formed by a neural network.

[0066] The test module 23 and the analysis module 24 can be controlled by a control unit 25. If necessary, the settings of the test module 23, the analysis module 24 or the control unit 25 can be changed via a user interface 28. Data occurring in the device 20 can be stored in a memory 26 if necessary, for example for later analysis or for use by components of the device 20. The test module 23, the analysis module 24 and the control unit 25 can be implemented as dedicated hardware, for example as an integrated circuit. Of course, they can also be partially or entirely combined or implemented as software running on a suitable processor, for example on a GPU or a CPU. The input 21 and the output 27 can be implemented as separate interfaces or as a combined bidirectional interface.

[0067] Figure 3 A simplified schematic diagram of a device 30 for providing a classifier for an AI module for processing input data provided by a sensor system of a motor vehicle is shown for a second embodiment. The device 30 has a processor 32 and a memory 31. The device 30 is for example a computer or a control device. In the memory 31 instructions are stored which, when executed by the processor 32, cause the device 30 to perform the steps according to one of the described methods. Thus, the instructions stored in the memory 31 represent a program executable by the processor 32 implementing the method according to the present application. The device 30 has an input 33 for receiving information, for example data of a test data set. Data generated by the processor 32 are provided via an output 34. Furthermore, data can be stored in the memory 31. The input 33 and the output 34 can be combined into one bidirectional interface.

[0068] The processor 32 can comprise one or more processor units, for example microprocessors, digital signal processors or combinations thereof.

[0069] The memories 26, 31 of the described embodiments can have not only volatile storage areas but also non-volatile storage areas and can comprise different storage devices and storage media, for example hard disks, optical storage media or semiconductor memories.

[0070] Figure 4A method for configuring a control system of a motor vehicle with a library of AI modules for processing input data provided by a sensor system of the motor vehicle is schematically shown. In a first step, the input data to be processed by an AI module is detected 40. Subsequently, the AI module is evaluated 41 based on context parameters. For this, a classifier, which can for example be formed by a neural network, can be used, which has been created beforehand in a manner as described above. The context parameters can for example include properties in the context of the input data or properties of the architecture of the AI module. Subsequently, based on the evaluation, the AI module or a combination of AI modules to be used for the input data and associated weights is determined 42.

[0071] Figure 5 A simplified schematic of a device 50 for configuring a control system of a motor vehicle with a library of AI modules for processing input data provided by a sensor system of the motor vehicle is shown for a first embodiment. The device 50 has an input 51 via which the input data to be processed by an AI module can be received and detected by a data module 52. Then, a classifier K, which can for example be formed by a neural network, evaluates the available AI modules based on context parameters. The context parameters can for example include properties in the context of the input data or properties of the architecture of the AI module. Finally, an analysis module 53 determines based on the evaluation the AI module or a combination of AI modules to be used for the input data and associated weights. Information about the AI module to be used and about the weights to be used can be transmitted via an output 56 of the device 50 to a merging module 80.

[0072] The data module 52 and the analysis module 53 can be controlled by a control unit 54. If necessary, the settings of the data module 52, the analysis module 53 or the control unit 54 can be changed via a user interface 57. Data occurring in the device 50 can be stored in a memory 55 when required, for example for later analysis or for use by components of the device 50. The data module 52, the analysis module 53 and the control unit 54 can be implemented as dedicated hardware, for example as an integrated circuit. Of course, they can also be partially or entirely combined or implemented as software running on a suitable processor, for example on a GPU or a CPU. The input 51 and the output 56 can be implemented as separate interfaces or as a combined bidirectional interface.

[0073] Figure 6A simplified schematic diagram of a device 60 for configuring a control system of a motor vehicle with a library of AI modules for processing input data provided by a sensor system of the motor vehicle is shown for a second embodiment. The device 60 has a processor 62 and a memory 61. The device 60 is, for example, a computer or a control device. In the memory 61, instructions are stored which, when executed by the processor 62, cause the device 60 to perform the steps according to one of the described methods. The instructions stored in the memory 61 thus represent a program executable by the processor 62 which implements the method according to the application. The device 60 has an input 63 for receiving information, for example input data to be processed by the AI module. Data generated by the processor 62 is provided via an output 64. Furthermore, data can be stored in the memory 61. The input 66 and the output 64 can be combined into one bidirectional interface.

[0074] The processor 62 can comprise one or more processor units, for example microprocessors, digital signal processors or combinations thereof.

[0075] The memories 55, 61 of the described embodiments can have not only volatile storage areas, but also non-volatile storage areas, and can comprise different storage devices and storage media, for example hard disks, optical storage media or semiconductor memories.

[0076] Figure 7 A motor vehicle 70 implementing the solution according to the application is shown schematically. The motor vehicle 70 has a control system 71 for automated or highly automated driving operation, which is configured by the device 50. In Figure 7 The device 50 is a separate component in the described embodiment, but the device 50 can also be integrated in the control system 71. In order to select an AI module from the library of AI modules, the device 50 uses a series of input data. This can be, for example, environmental data from an environmental sensor system 72 installed in the motor vehicle 70 or operating parameters of the motor vehicle 70 provided by a control device 73. A further component of the motor vehicle 70 is a data transmission unit 74 via which, inter alia, a connection to a backend can be established, for example for obtaining additional or modified AI modules. In order to store the library of AI modules or other data, there is a memory 75. Data exchange takes place between the different components of the motor vehicle 70 via a network 76.

[0077] In the following, preferred embodiments of the application are explained by way of example with Figure 8 and Figure 9

[0078] ​The unique characteristics in the context of the input data to be processed influence the functional quality of the AI module performing the processing to a particular extent. These characteristics can be very diverse and are not necessarily intuitive to humans as, for example, a special color value, right or wrong or a distribution of specific frequencies. The architectural characteristics of the AI module additionally play a certain role. Thus, for example, specific features in the combination of a neural network have an influence on the performance of the neural network. For example, if in addition to a learned neural network there is also a rule-based knowledge base for the allowed street signs, the resulting AI model will have a better performance when performing sign recognition than a similar AI model without such a knowledge base. All influences on the data that influence or at least can influence the functional behavior of the AI module belong to the characteristics in the context of the input data to be processed. These influences can be not only semantic, for example, weather, traffic or environment, but also not intuitive as described above.

[0079] Figure 8 A system diagram is schematically shown according to the present application for providing a solution for a classifier K for AI modules NN i for processing input data provided by a sensor system of a motor vehicle. The classification system uses a set of AI modules NN i , for example trained neural networks, as candidates for later implementation in a specific environment. The AI modules NN i are set up for the same task, for example object recognition or semantic segmentation, but differ in architecture, training data and training parameters.

[0080] Now, during the test phase, all given AI modules NN i are used to infer for all data points d n in the test data set D. Here, the term "inference" denotes the process of using a trained model to draw a conclusion. By comparing the output of the AI module NN i with the respective ground truth G n using the given context parameters P n , a functional quality FG i_n is determined for each data point d n . Now, for example, this information can be used to train a neural network which learns during training the importance of different context parameters P n and the functional quality FG i of different AI modules NN i with context parameters P nthe relevance of the data points. In this way a classifier K is created, which outputs for a given context parameter P, for all AI modules NN i the functional quality FG in relation to the data points i or the functional quality FG i corresponding weights W i Alternatively, the classifier K can output an empirically based weighting function for the joint inference of the AI modules NN i family in the sense of a group.

[0081] Figure 9 A system diagram is shown schematically, which illustrates a solution according to the application for configuring a control system 71 with a library of AI modules NN i During the operation of the control system 71, for example in a vehicle, context parameters P are extracted from the input data E obtained by means of the sensor system 81 by means of the classifier K. On the basis of these context parameters P, the optimal combination of AI modules NN i and the associated weights W i is determined. This is preferably carried out synchronously with the data processing in the AI modules NN i , because only for the merging of the outputs of the AI modules NN i must it be known how the respective outputs are to be evaluated. The merging module 80 combines the outputs of the selected AI modules NN i according to the weights W i provided by the classifier K. Furthermore, the context parameters P determined by the classifier K can be transmitted to the selection unit 82, which can start or stop individual AI modules NN i on the basis of the parameters P.

[0082] List of reference signs

[0083] 10 selection of AI modules

[0084] 11 selection of test data sets

[0085] 12 application of AI modules to data points of test data sets

[0086] 13 determination of functional quality for data points

[0087] 14 creation of classifier for AI modules

[0088] 20 device

[0089] 21 input

[0090] 22 database

[0091] 23 test module

[0092] 24 analysis module

[0093] 25 control unit

[0094] 26 memory

[0095] 27 output

[0096] 28 user interface

[0097] 30 device

[0098] 31 memory

[0099] 32 processor

[0100] 33 input

[0101] 34 output

[0102] 40 detecting input data to be processed by an AI module

[0103] 41 evaluating an AI module according to context parameters

[0104] 42 determining an AI module and weights to be used

[0105] 50 device

[0106] 51 input

[0107] 52 data module

[0108] 53 analysis module

[0109] 54 control unit

[0110] 55 memory

[0111] 56 output

[0112] 57 user interface

[0113] 60 device

[0114] 61 memory

[0115] 62 processor

[0116] 63 input

[0117] 64 output

[0118] 70 motor vehicle

[0119] 71 control system

[0120] 72 environment sensor

[0121] 73 control device

[0122] 74 data transmission unit

[0123] 75 memory

[0124] 76 network

[0125] 80 merging module

[0126] 81 sensing mechanism

[0127] d n data point

[0128] D test data set

[0129] E input data

[0130] FG i ,FG i_n functional quality

[0131] G n ground truth

[0132] K classifier

[0133] NN i AI module

[0134] P,P n context parameter

[0135] W i weight

Claims

1. A method for providing an AI module (NN i ), the AI ​​module being used to process input data (E) provided by a sensor system (72, 81) of a motor vehicle (70), the method comprising the following steps: -By applying the AI ​​module (NN i ) applies (12) to two or more data points (d n ) to determine the AI ​​module (NN i ) output, where For the two or more data points (d n ), the associated basic facts (G n ) and context parameters (P n ) is known; - By applying the n ) of the AI ​​module (NN i ) output and the associated ground truth (G n ) are compared, for the two or more data points (d n ) determines each of the (13) functional qualities (FG i_n );as well as - Using the context parameters (P n ) and for the two or more data points (d n ) Determined functional quality (FG i_n ), for the AI ​​module (NN i ) is trained (14) with a classifier (K) that outputs an expected functional quality (FG i ), The context parameters include characteristics of the context of the data point or characteristics of the architecture of the AI ​​module.

2. The computer-implemented method of claim 1 , wherein: The AI ​​module (NN i ) implements an AI model or a family of AI models in a collective sense.

3. The computer-implemented method of claim 1 , wherein: For the data point (d n )'s AI module (NN i ) output and the associated ground truth (G n ), using the IoU metric.

4. The computer-implemented method of claim 1 , wherein: The classifier (K) is formed by a neural network.

5. The computer-implemented method according to any one of claims 1 to 4, wherein: AI module (NN i ) is configured to perform environmental recognition for an automated driving function of a motor vehicle (70) based on environmental data from an environmental sensor system (72).

6. The computer-implemented method of any one of claims 1 to 4, wherein: Different AI modules (NN i ) to match different lighting conditions, different vehicle speeds, different vehicle environments, different driving situations, different environmental conditions, different driving conditions or different target settings.

7. A computer-readable storage medium having a computer program having instructions stored thereon, which, when executed by a computer, causes the computer to execute the method for providing a method for an AI module (NN) according to any one of claims 1 to 6. i ) of the method, wherein the AI ​​module is used to process input data (E) provided by the sensor system (72, 81) of the motor vehicle (70).

8. A method for providing an AI module (NN i ), the AI ​​module is used to process input data (E) provided by a sensor mechanism (72, 81) of a motor vehicle (70), the device having: - a testing module (23) for making the AI ​​module (NN i ) applies (12) to two or more data points (d n ) to determine the AI ​​module (NN i ) output, where For the two or more data points (d n ), the associated basic facts (G n ) and context parameters (P n ) is known; as well as - an analysis module (24) for analyzing the data points (d n ) of the AI ​​module (NN i ) output and the associated ground truth (G n ) are compared, for the two or more data points (d n ) determines each of the (13) functional qualities (FG i_n ), and using the context parameter (P n ) and for the two or more data points (d n ) Determined functional quality (FG i_n ), for the AI ​​module (NN i ) is trained (14) with a classifier (K) that outputs an expected functional quality (FG i ), The context parameters include characteristics of the context of the data point or characteristics of the architecture of the AI ​​module.

9. A method for AI module (NN i ) is configured to output the AI ​​module (NN) for a given context parameter (P). i ) of the expected functional quality (FG i ),in, The classifier (K) is provided by means of a method according to any one of claims 1 to 6.

10. A method for utilizing an AI module (NN i ) program library to configure the control system (71) of a motor vehicle (70) using a computer-implemented method, wherein the AI ​​module (NN i ) for processing input data (E) provided by a sensor system (72, 81) of the motor vehicle (70), the method comprising the following steps: - Detection (40) is to be performed by the AI ​​module (NN i ) input data (E) processed; - using a classifier (K) provided by the method according to any one of claims 1 to 6, classifying the AI ​​module (NN) based on context parameters (P) i ) is evaluated (41), and the classifier outputs the AI ​​module (NN) for the context parameter (P). i ) of the expected functional quality (FG i );as well as - Based on the result of said evaluation (41), determining (42) the AI ​​module (NN) to be used for the input data (E) i ) or the AI ​​module to be used (NN i ) and the associated weights (W i ).

11. The computer-implemented method of claim 10, wherein: AI module (NN i ) is configured to perform environmental recognition for an automated driving function of a motor vehicle (70) based on environmental data from an environmental sensor system (72).

12. The computer-implemented method of claim 10, wherein: Different AI modules (NN i ) to match different lighting conditions, different vehicle speeds, different vehicle environments, different driving situations, different environmental conditions, different driving conditions or different target settings.

13. A computer-readable storage medium having stored thereon a computer program having instructions, which, when executed by a computer, causes the computer to execute the method for utilizing an AI module (NN) according to any one of claims 10 to 12. i )'s program library (B) configures the control system (71) steps.

14. A method for utilizing an AI module (NN i ) program library to configure the control system (71) of the motor vehicle (70) device (50), the AI ​​module (NN i ) for processing input data (E) provided by a sensor system (72, 81) of the motor vehicle (70), the device comprising: - a data module (52) for detecting (40) the data to be transmitted by the AI ​​module (NN i ) input data (E) processed; - a classifier (K) provided by the method according to any one of claims 1 to 6 for classifying the AI ​​module (NN) based on context parameters (P) i ) is evaluated (41), where The classifier (K) is configured to output the AI ​​module (NN) for the context parameter (P). i ) of the expected functional quality (FG i ); as well as - an analysis module (53) for determining (42) the AI ​​module (NN) to be used for the input data (E) based on the output of the classifier (K) i ) or the AI ​​module to be used (NN i ) and the associated weights (W i ).

15. A motor vehicle (70), characterized in that The motor vehicle (70) has The apparatus according to claim 14 or configured to perform the method according to any one of claims 10 to 12.

Images