DE102024201254B3 - Method and device for detecting a hands-off state on a steering wheel of a vehicle - Google Patents

Abstract

The invention relates to a method for detecting a hands-off state (6) on a steering wheel (51) of a vehicle (50), wherein at least one steering variable (4) is detected on the steering wheel (51), wherein the detected at least one steering variable (4) is fed to a trained machine learning model (5) as input data, wherein the machine learning model (5) is trained to detect a hands-off state (6) based on at least the detected at least one steering variable (4) and to output an associated state variable as output data (20), wherein the trained machine learning model (5) is provided in such a way that the trained machine learning model (5) has at least three component sections (5-x) that follow one another with respect to a signal flow, wherein a trained component (7-1) is provided in a first component section (5-1), wherein at least one trained component (5-2-x) in at least one middle component section (5-2) corresponding to a vehicle configuration of the vehicle (50) is provided, and wherein in a last component section (5-3) a trained component (7-3-x) corresponding to the vehicle configuration of the vehicle (50) is provided. Furthermore, the invention relates to a device (1) for detecting a hands-off state (6) on a steering wheel (51).

Description

The invention relates to a method and a device for detecting a hands-off state on a steering wheel of a vehicle. The invention also relates to a method and a device for providing a trained machine learning model for detecting a hands-off state on a steering wheel of a vehicle. The invention also relates to a method and a device for maintaining a totality of machine learning models provided in vehicles that have been trained to detect a hands-off state.

Sensors such as a capacitive steering wheel are used in vehicles to monitor driver activity. Such a steering wheel detects whether the driver has touched or not touched the steering wheel (“hands-off”) using a capacitive sensor. A result is transmitted to the functions used, such as a longitudinal and/or lateral guidance assistance system. Driver activity and driver attention can be determined from the touch of the hands on the steering wheel. For example, it can be intended to advise the driver to put their hands on the steering wheel if it is detected that the hands were not on the steering wheel for a specified time during lateral guidance.

In order to save additional costs for a capacitive sensor in the steering wheel, it is known to monitor driver activity using machine learning models, in particular using artificial neural networks, based on a torque (hand torque) recorded on the steering wheel. Such a method is known, for example, from DE 10 2019 211 016 A1 known.

A major challenge of steering torque-based detection is to identify the steering torque induced by the driver in the measured (noisy) steering torque. Many factors can lead to a noisy steering torque, in particular a position of the sensor (this usually forms part of the steering gear or steering assistance, which creates a system capable of torsional vibration via the elasticity of the steering column in conjunction with the steering wheel, the inherent dynamics of which make it difficult to precisely measure the torque induced by the driver); a level of friction in the steering system; a re-excitation from the road due to unevenness; a dead weight of the steering wheel / steering system; a vibration of the steering wheel due to an assistance function (e.g. due to haptic feedback when leaving the lane).

In addition, the characteristics (used for hands-off detection) of the measured steering torque can change due to external influences, e.g. temperature, vehicle loading, the presence of a trailer, tire type and/or tire condition, steering system changes over the service life, road gradient/inclination/slant, etc.

In order to achieve the most robust possible recognition of the hands-off state, it is particularly necessary that all of the influences and scenarios mentioned are present in the training data. In practice, however, this is difficult to achieve, since the input data already spans a very large dimensional space.

The invention is based on the object of improving the recognition of a hands-off state by means of a trained machine learning model.

The object is achieved according to the invention by a method having the features of patent claim 1 and a device having the features of patent claim 3. Furthermore, the object is achieved according to the invention by a method having the features of patent claim 6 and a device having the features of patent claim 11. Furthermore, the object is achieved according to the invention by a method having the features of patent claim 12 and a device having the features of patent claim 13. Advantageous embodiments of the invention emerge from the subclaims.

In particular, according to a first aspect, a method for detecting a hands-off state on a steering wheel of a vehicle is provided, wherein at least one steering variable is detected on the steering wheel, wherein the detected at least one steering variable is fed to a trained machine learning model as input data, wherein the machine learning model is trained to detect a hands-off state based on at least the detected at least one steering variable and to output an associated state variable as output data, wherein the trained machine learning model is provided in such a way that the trained machine learning model has at least three component sections that follow one another with respect to a signal flow, wherein a trained component is provided in a first component section, wherein at least one trained component is provided in at least one middle component section corresponding to a vehicle configuration of the vehicle, and wherein in a last component section a trained component corresponding to the vehicle configuration of the vehicle is provided.

Furthermore, according to the first aspect, a device for detecting a hands-off state on a steering wheel of a vehicle is also provided, comprising at least one steering variable sensor which is designed to detect at least one steering variable on the steering wheel, and a data processing device, wherein the data processing device is designed to receive the detected at least one steering variable, to provide a trained machine learning model and to supply the detected at least one steering variable to the trained machine learning model as input data, wherein the machine learning model is trained to detect a hands-off state based on at least the detected at least one steering variable and to output associated state information as output data, wherein the trained machine learning model is provided in such a way that the trained machine learning model has at least three component sections which follow one another with respect to a signal flow, wherein the first component section provides a trained component, wherein at least one trained component is provided in at least one middle component section corresponding to a vehicle configuration of the vehicle, and wherein in a last component section a trained component is provided corresponding to the vehicle configuration of the vehicle.

Furthermore, in particular in a second aspect, a method is provided for providing a trained machine learning model for recognizing a hands-off state on a steering wheel of a vehicle, wherein the trained machine learning model is provided for a plurality of vehicle configurations, wherein the trained machine learning model is provided such that the trained machine learning model has at least three component sections that follow one another with respect to a signal flow, wherein a trained component is provided in a first component section for all vehicle configurations, wherein at least one trained component is provided in at least one middle component section for at least a subset of all vehicle configurations, and wherein in a last component section a trained component is provided exclusively for a vehicle configuration corresponding thereto, and wherein to provide the trained machine learning model for a given vehicle configuration, the respective trained components are selected and provided taking the vehicle configuration into account.

According to the second aspect, in particular, a device is also created for providing a trained machine learning model for recognizing a hands-off state on a steering wheel of a vehicle, comprising a central server, wherein the central server is configured to provide the trained machine learning model for a plurality of vehicle configurations, and to provide the trained machine learning model in such a way that the trained machine learning model has at least three component sections that follow one another with respect to a signal flow, wherein a trained component is provided in a first component section for all vehicle configurations, wherein at least one trained component is provided in at least one middle component section for at least a subset of all vehicle configurations, and wherein in a last component section a trained component is provided exclusively for a vehicle configuration corresponding thereto, and wherein in order to provide the trained machine learning model for a given vehicle configuration, the respective trained components are selected and provided taking the vehicle configuration into account.

According to a third aspect, in particular a method is provided for maintaining a totality of machine learning models provided in vehicles and trained to recognize a hands-off state, wherein recorded and/or collected maintenance data relating to respective components of at least three component sections of the respective trained machine learning model are received from the vehicles, wherein the received maintenance data are clustered according to components, and wherein the machine learning model is retrained according to the clustered components, taking into account the received maintenance data, and the retrained machine learning model is provided.

According to the third aspect, in particular, a device is also provided for maintaining a totality of machine learning models provided in vehicles and trained to recognize a hands-off state, comprising a central server, wherein the central server is configured to receive maintenance data recorded and/or collected by the vehicles relating to respective components of at least three component sections of the respective trained machine learning model from the vehicles, the received maintenance data according to components ten, and to retrain the machine learning model according to the clustered components, taking into account the received maintenance data, and to provide the retrained machine learning model.

The methods and devices according to the individual aspects make it possible to provide a machine learning model that can be generalized across vehicle variants or vehicle configurations and at the same time to provide efficient and comprehensible life cycle management for recognizing the hands-off state. Characteristics from individual vehicle configurations can play an important role in recognizing the hands-off state. Using the proposed methods and devices, these characteristics can be taken into account individually. In principle, these could also be mapped in a large machine learning model with a lot of data. A disadvantage here, however, is that a lot of training data is required and, as a result, a clear separation according to individual vehicle configurations is no longer possible. If complaints arise during use, the complete machine learning model must also be retrained. This is computing time and resource intensive and subsequently requires complete validation in particular in order to be able to be used in the vehicle. If different, smaller machine learning models are trained depending on the vehicle configuration, they can be specifically adapted, but versioning and handling the machine learning models is more complex than with a global machine learning model. In addition, shared characteristics that are similar for each vehicle configuration are not efficiently taken into account in such an approach.

It is a basic idea of the invention to provide the trained machine learning model in such a way that the trained machine learning model has at least three component sections that follow one another with respect to a signal flow. In a first component section, a (single) trained component is provided. This trained component is used by all vehicles or all vehicle configurations. In at least one middle component section, at least one trained component is provided corresponding to a vehicle configuration of the vehicle. These components can be shared by several vehicles or vehicle configurations. In a last component section, a (single) trained component is provided corresponding to the vehicle configuration of the vehicle. This trained component is only used by one vehicle configuration. In particular, the invention makes it possible to provide different variants of the machine learning model and yet structurally exploit characteristics that are shared by several vehicles or vehicle configurations when recognizing the hands-off state. Furthermore, the proposed structure of the trained machine learning model makes it possible to simplify maintenance. In particular, only the component or components that have a problem and/or a complaint when recognizing the hands-off state need to be maintained, in particular retrained. Unaffected components do not need to be maintained, in particular retrained. This is especially true if a component is used in multiple vehicle configurations. This can be the case in particular for the trained component of the first component section, since this is used by all vehicle configurations, and for one or more components of the at least one middle component section, which can be used by multiple vehicle configurations. If problems and/or complaints arise with multiple vehicle configurations that all include this component or these components, this component or these components will be retrained, while the other components are not affected by the retraining.

A steering variable is in particular a variable that represents and/or describes a current state of the steering wheel. A steering variable is in particular a torque that is detected in particular by means of a torque sensor on the steering wheel. In principle, however, a steering variable can also be another variable that is detected directly or indirectly on the steering wheel. For example, it can be provided to detect a current on an electrical machine on the steering wheel and use it as a steering variable. The hands-off state can be detected exclusively on the basis of the steering variable detected on the steering wheel, in particular a detected torque. However, it is also possible in particular for the machine learning model to be fed with further (steering) variables that are detected on the steering wheel (e.g. a steering wheel angle and/or a steering wheel angular velocity, etc.) and for the trained machine learning model to detect the hands-off state taking this additional variable(s) into account. Variables that are not recorded on the steering wheel, such as vehicle speed, lateral acceleration, yaw rate, wheel ticks, damper information and/or other driving dynamics variables, as well as weather and/or situation data, etc., can be taken into account in particular as context information and also fed to the trained machine learning model as input data. In particular, however, No capacitive sensor is provided on the steering wheel.

A vehicle configuration is in particular a description of properties of a vehicle. These properties include in particular a vehicle class, a vehicle model and/or individual equipment of the vehicle. The vehicle configuration is used in particular to select the components of the at least one middle component section and the last component section that are relevant for recognizing the hands-off state.

The first component section comprises in particular a (single) trained component that is used by all vehicles or all vehicle configurations. All vehicles or vehicle configurations in particular use the same weights and/or parameters of this one component.

For a given vehicle configuration, the at least one middle component section includes at least one trained component corresponding to the vehicle configuration. For example, individual components can be provided for the following areas: e.g. in the drive area: electric drive, combustion engine or hybrid; in the vehicle class area: compact class, sports utility vehicle (SUV), minibus, etc.

For a given vehicle configuration, the last component section includes a trained component that corresponds to the respective vehicle configuration. For example, it can be provided to provide an individual component for each vehicle model, e.g. vehicle model 1, vehicle model 2, vehicle model 3, etc. In particular, exactly one component is then active in the last component section for each vehicle model.

Components are in particular components defined virtually in the memory. A component is in particular an area of the signal processing of the machine learning model that can be defined with regard to a signal flow. In principle, however, it can also be provided that the components are designed in physical form.

It may be provided that when recognizing the hands-off state, a hands-on state is also recognized.

A hands-off state is in particular a state in which the driver does not touch the steering wheel. In particular, none of the driver's fingers are in contact with the steering wheel. Detecting the hands-off state can in particular comprise providing a hands-off state signal. This comprises, for example, a hands-off probability or coded signals for the states "hands-off detected" and "hands-off not detected". A hands-on state is in particular a state in which the driver touches the steering wheel. A hands-on/hands-off state can also be provided, for example as a hands-on/hands-off state signal with in particular at least two signal states (e.g. "hands-on detected" or "hands-off detected").

The machine learning model and in particular the components can in particular comprise one or more neural networks. The neural network(s) can in particular comprise several inner layers. The machine learning model in particular comprises one or more artificial recurrent neural networks that process the input data X _t at each time t and output a hands-off probability y _t in [0,1]: y _t = p(x _t | x _0:t-1 ). In this case, a recurrent neural network has in particular a so-called memory h in which information from previous time steps is stored and which can be used for the output at the current time step.

The output data of the trained machine learning model are further processed, for example, by filtering before decreasing functions (e.g. a lateral guidance assistant) process them. In particular, it can be provided that, based on a comparison of the hands-off probability with a predetermined threshold value, a binary hands-off signal is provided (with the two states "hands-off detected" and "hands-off not detected").

During a training phase, the machine learning model is trained in particular for different vehicle configurations. This is done using training data that includes pairs in which data of the at least one steering variable, in particular torque data, are each paired with a hands-off state (as ground truth). The data of the at least one steering variable, in particular the torque data, are in particular time series of the at least one steering variable recorded on the steering wheel, in particular time series of torques recorded on the steering wheel. The training data are obtained in particular with the help of test drives and/or in simulators for the individual vehicle configurations. In principle, the provision of training data can in particular be carried out in accordance with the DE 10 2019 211 016 A1 Depending on the vehicle configuration, it is intended that during training or a training step only those components are changed those that correspond to the vehicle configuration to which the training data belongs. In other words, the component of the first component section is always trained, at least during initial training. Components of the at least one middle component section and the last component section, on the other hand, are only changed during training if they belong to the vehicle configuration that corresponds to the current training data. Training is otherwise carried out in a manner known per se, in particular by means of supervised learning. In particular, during training by means of back projection, the parameters of the machine learning model (in particular the weights) are adjusted until a calculated error of the output to the ground truth is below a predetermined threshold.

Maintenance data can, for example, include information on customer complaints, for example information on frequently occurring false warnings. The false warnings can, for example, be identified by vehicle users and transmitted to the central server as maintenance data. The false warnings are in particular assigned to the respective vehicle configurations or linked to them so that they can be clustered according to components and the machine learning model can be retrained according to the clustered components. Furthermore, maintenance data can also be generated by recording fleet data and a statistical analysis. For example, a confidence of outputs of the trained machine learning model can be evaluated as part of the statistical analysis. For example, the confidence can be compared with a specified threshold value, whereby maintenance data, in particular for the respective vehicle configuration, is generated and transmitted if the confidence of an output is below the specified threshold value. The maintenance data can be clustered according to components based on the respective associated vehicle configuration.

Parts of the devices, in particular the data processing device and the central server, can be designed individually or in combination as a combination of hardware and software, for example as program code that is executed on a microcontroller or microprocessor. However, it can also be provided that parts are designed individually or in combination as an application-specific integrated circuit (ASIC) and/or field-programmable gate array (FPGA). The data processing device and the central server each comprise in particular at least one computing device and at least one memory. Furthermore, the data processing device and the central server can each have at least one communication interface for communication.

In particular, it can be provided that the methods according to the second aspect and the third aspect are each carried out or designed as computer-implemented methods.

In one embodiment, the vehicle configuration of the last component section relates to a vehicle model. This allows a trained component to be provided for each vehicle model. Characteristics when recognizing the hands-off state can then be taken into account individually for each vehicle model and learned during training.

In one embodiment, it is provided that the provision includes loading the trained machine learning model into a memory of a control unit of the respective vehicle. The provided trained machine learning model can then be used in the vehicle to recognize the hands-off state.

In one embodiment, it is provided that trained components of the trained machine learning model that do not correspond to the vehicle configuration of the respective vehicle are deactivated or will be deactivated. This can save memory and/or computing resources when applying the trained machine learning model, since only the necessary components are used. The deactivation can in particular be carried out in such a way that memory areas that contain the unused components are released. Furthermore, it can be provided that when loading the trained machine learning model (or a description and parameters thereof) into a memory of a data processing device and/or a device (e.g. control unit) of a vehicle, only the non-deactivated components are loaded into the memory, but not the deactivated components.

In one embodiment, it is provided that, in order to provide the trained machine learning model, information relating to the vehicle configuration is recorded and/or obtained, wherein the trained machine learning model is provided on the basis of the recorded and/or obtained information. This allows the trained machine learning model to be individually configured for a vehicle. This can be done, for example, at a suitable point during the manufacture of the vehicle or at a location provided for the vehicle. The information can be transmitted to a specific device (e.g. control unit) that is intended to detect the hands-off state. The information can be, for example, a serial number, a unique identification number and/or a barcode and/or a QR code. Based on this, the required components of the trained machine learning model can be determined. The trained machine learning model is then provided by providing the specific components, linked to one another using signal technology.

In one embodiment, it is provided that, in the case of a vehicle configuration that is not yet known, at least one component for the at least one middle component section and/or one component for the last component section is generated and trained and provided. As a result, a trained machine learning model for a vehicle configuration that is not yet known can be provided based on the remaining components that have already been trained.

Further features for the design of the respective devices emerge from the description of the designs of the respective methods. The advantages of the respective devices are the same as those of the designs of the respective methods.

Further features for the design of the respective aspects emerge from the description of the designs of the other aspects. The advantages of the respective aspects are the same as for the designs of the other aspects.

Furthermore, a steering system is also provided, comprising a device for detecting a hands-off state on a steering wheel of a vehicle according to one of the described embodiments.

Furthermore, a vehicle is also provided, comprising a steering system according to one of the described embodiments and/or a device for detecting a hands-off state on a steering wheel of a vehicle according to one of the described embodiments.

• 1 eine schematische Darstellung zur Verdeutlichung von Ausführungsformen der Vorrichtung zum Erkennen eines Hands-Off-Zustands an einem Lenkrad;
• 2 eine schematische Darstellung zur Verdeutlichung einer Struktur des trainierten Maschinenlernmodells;
• 3 ein schematisches Ablaufdiagramm zur Verdeutlichung von Ausführungsformen des Verfahrens zum Bereitstellen eines trainierten Maschinenlernmodells zum Erkennen eines Hands-Off-Zustands an einem Lenkrad eines Fahrzeugs;
• 4 ein schematisches Ablaufdiagramm zur Verdeutlichung von Ausführungsformen des Verfahren zum Warten einer Gesamtheit aus in Fahrzeugen bereitgestellten zum Erkennen eines Hands-Off-Zustands trainierten Maschinenlernmodellen;
• 5a-5c schematische Darstellungen des Maschinenlernmodells zur Verdeutlichung des Nachtrainierens in unterschiedlichen Konstellationen.

The invention is explained in more detail below using preferred embodiments with reference to the figures.

• 1 a schematic representation to illustrate embodiments of the device for detecting a hands-off state on a steering wheel;
• 2 a schematic representation to illustrate a structure of the trained machine learning model;
• 3 a schematic flow diagram to illustrate embodiments of the method for providing a trained machine learning model for detecting a hands-off state on a steering wheel of a vehicle;
• 4 a schematic flow diagram to illustrate embodiments of the method for maintaining a set of machine learning models provided in vehicles and trained to recognize a hands-off state;
• 5a-5c Schematic representations of the machine learning model to illustrate retraining in different constellations.

The 1 shows a schematic representation of an embodiment of the device 1 for detecting a hands-off state 6 on a steering wheel 51. The device 1 is arranged in particular in a vehicle 50 and is part of a steering system 60 there. The method described in this disclosure according to the first aspect is illustrated and explained in more detail below using the device 1.

The device 1 comprises a steering variable sensor 2 and a data processing device 3. The steering variable sensor 2 is designed to detect a steering variable 4 on the steering wheel 51 of the vehicle 50. The steering variable sensor 2 is, for example, a torque sensor and the steering variable 4 is a torque. In principle, additional steering variable sensors can be provided as an alternative or in addition, which detect additional steering variables. The device 1 can comprise a communication interface (not shown) for communication.

The data processing device 3 comprises a computing device 3-1 and a memory 3-2. The computing device 3-1 is set up to carry out computing operations necessary for carrying out measures of the method and can access data stored in the memory 3-2 for this purpose.

The data processing device 3 is configured to receive the recorded at least one steering variable 4, a trained machine learning model 5 (cf. 2 ) and to feed the recorded at least one steering variable 4 to the trained machine learning model 5 as input data.

The machine learning model 5 ( 2 ) is trained to recognize the hands-off state 6 based on at least the recorded at least one steering variable 4 and to use associated state information as output data 20 ( 2 ). The trained machine learning model 5 is provided in such a way that the trained machine learning model 5 has at least three component sections 5-1, 5-2, 5-3 that follow one another with respect to a signal flow. The first component section 5-1 provides a trained component 7-1. At least one middle component section 5-2 provides at least one trained component 7-2-x corresponding to a vehicle configuration of the vehicle 50. A last component section 5-3 provides a trained component 7-3-x corresponding to the vehicle configuration of the vehicle 50.

The hands-off state 6 is fed, for example, as a state signal or state information to a control unit 52 of the vehicle 50 for further processing. The control unit 52 can be, for example, a lateral guidance assistant or another assistance system. The state signal or the state information can, for example, contain a hands-off probability or a binary state value with the two states "hands-off recognized" and "hands-off not recognized".

The 2 shows a schematic representation to illustrate a structure of the trained machine learning model 5. The component sections 5-1, 5-2, 5-3 are shown. More than just one middle component section 5-2 can also be provided. The trained machine learning model 5 is supplied with the recorded at least one steering variable 4 as input data. The signal flow runs from the first component section 5-1 via the middle component section 5-2 to the last component section 5-3. Here, the (only) component 7-1 of the first component section 5-1 is always active (ie in all vehicle configurations). The components 7-2-x, 7-3-x of the middle component section 7-2 and the last component section 7-3, on the other hand, are based on an existing vehicle configuration of the vehicle 50 ( 1 ), in which the trained machine learning model 5 is used, is selected and used. In the last component section 5-3, only one of the components 7-3-x is active when used and outputs the hands-off state 6.

• Komponente 7-2-1: elektrischer Antrieb
• Komponente 7-2-2: Verbrennungsmotor
• Komponente 7-2-3: Kompaktklasse
• Komponente 7-2-4: Sports Utility Vehicle (SUV)
• Komponente 7-2-5: Kleinbus

usw.The components 7-2-x of the middle component section 5-2 can, for example, correspond to the following properties of the vehicle:

• Component 7-2-1: electric drive
• Component 7-2-2: Internal Combustion Engine
• Component 7-2-3: Compact class
• Component 7-2-4: Sports Utility Vehicle (SUV)
• Component 7-2-5: Minibus

etc.

• Komponente 7-3-1: Fahrzeugmodell 1
• Komponente 7-3-2: Fahrzeugmodell 2
• Komponente 7-3-3: Fahrzeugmodell 3
• Komponente 7-3-4: Fahrzeugmodell 4

usw.The components 7-3-x of the last component section 5-3 can, for example, correspond to the following properties of the vehicle:

• Component 7-3-1: Vehicle Model 1
• Component 7-3-2: Vehicle Model 2
• Component 7-3-3: Vehicle Model 3
• Component 7-3-4: Vehicle Model 4

etc.

If a vehicle configuration includes, for example, that the vehicle of vehicle model 4 has an electric drive and is a minibus, the components 7-1, 7-2-1, 7-2-5 and 7-3-4 are active. In particular, a signal flow only takes place between these active components 7-1, 7-2-1, 7-2-5 and 7-3-2. The component 7-3-2 outputs the hands-off state 6 as output data 20.

The remaining components 7-2-x and 7-3-x are not used and can be deactivated in particular to save memory and computing resources.

The 3 shows a schematic flow chart to illustrate embodiments of the method for providing a trained machine learning model for detecting a hands-off state on a steering wheel of a vehicle. The method is carried out by means of a device 40 ( 1 ) according to the second aspect. The device 40 comprises a central server 41. The central server 41 comprises a computing device 41-1 and a memory 41-2. The computing device 41-1 is set up to carry out computing operations necessary for carrying out measures of the method and can access data stored in the memory 41-2 for this purpose. The central server 41 can comprise a communication interface (not shown) for communication.

In a measure 100, the trained machine learning model is provided for several vehicle configurations, wherein the trained machine learning model is provided such that the trained machine learning model has at least three component sections that follow one another with respect to a signal flow, wherein a trained component is provided in a first component section for all vehicle configurations, wherein at least one trained component is provided in at least one middle component section for at least a subset of all vehicle configurations, and wherein in a last component section a trained component is provided exclusively for a vehicle configuration corresponding thereto.

In a measure 101, in order to provide the trained machine learning model for a given vehicle configuration, the respective trained components are selected and provided taking into account the vehicle configuration. According to the example already used above, the given vehicle configuration can provide that the vehicle of the vehicle model 4 has an electric drive and is a minibus. Therefore, the following trained components 7-1, 7-2-1, 7-2-5 and 7-3-4 ( 2 ) in measure 101 and the trained machine learning model configured in this way is provided, in particular in the form of a structural description which describes the trained components 7-1, 7-2-1, 7-2-5 and 7-3-4 and respective associated links between the trained components 7-1, 7-2-1, 7-2-5 and 7-3-4 as well as associated parameters (weights, hyperparameters, etc.).

In a measure 102, it can be provided that the provision includes loading the trained machine learning model into a memory of a device (e.g. control unit) of the respective vehicle. This can be done, for example, when manufacturing the vehicle and/or the device.

Measure 101 may provide that trained components 7-2-x, 7-3-x ( 2 ) of the trained machine learning model that do not correspond to the vehicle configuration of the respective vehicle are deactivated. The structure description then no longer contains any information about these trained components 7-2-x, 7-3-x.

It may be provided in a measure 99 that, in order to provide the trained machine learning model 5, information 42 ( 1 ) relating to the vehicle configuration is recorded and/or obtained, wherein the trained machine learning model 5 is provided on the basis of the recorded and/or obtained information 42. The information 42 contains in particular details on the properties of the respective vehicle 50 corresponding to the individual components 5-2-x, 5-3-x. The central server 40 can, for example, look up in a database on the basis of the information 42 which components 5-2-x, 5-3-x must be used or activated for the trained machine learning model 5.

It can be provided that, in the case of a vehicle configuration that is not yet known, at least one component 5-2-x for the at least one middle component section 5-2 and/or a component 5-3-x for the last component section 5-3 is generated, trained and provided. This is also done using the central server 40. For this purpose, the central server 40 receives training data that corresponds in particular to the vehicle configuration that is not yet known in order to be able to train the newly generated component 5-2-x, 5-3-x. The training takes place in the same way as already explained in the general description.

The 4 shows a schematic flow chart to illustrate embodiments of the method for maintaining a set of machine learning models provided in vehicles and trained to recognize a hands-off state. The method is carried out in particular by means of a central server 40 ( 1 ) is carried out.

In a measure 200, recorded and/or collected maintenance data 43 ( 1 ) concerning respective components 7-1, 7-2-x, 7-3-x of the at least three component sections 5-1, 5-2, 5-3 of the respective trained machine learning model 5 are received from the vehicles 50 by means of the central server 40. It is assumed here that a large number of vehicles 50 are present in the same or different vehicle configurations. These vehicles 50 collect and transmit the maintenance data 43.

In a measure 201, the received maintenance data 43 are sorted by components 5-1, 5-2-x, 5-3-x ( 2 ) clustered (ie, in particular, summarized or grouped).

In a measure 202, the machine learning model is retrained taking into account the maintenance data 43 received according to the clustered components 5-1, 5-2-x, 5-3-x and the retrained machine learning model 5 is made available. The retraining only affects those components 5-1, 5-2-x, 5-3-x that have been identified as problematic and/or for which there is a complaint from a user of the vehicles. The complaints, which may include, for example, that the recognition of the hands-off state does not function reliably, are used to identify the problem. nated, those components 7-1, 7-2-x, 7-3-x that are affected by the complaint can be identified, particularly through clustering or the formation of intersections.

In particular, a measure 203 provides that the machine learning model 5 ( 1 ) is transmitted to the respective vehicle 50 and is activated there, for example by loading the retrained machine learning model 5 into the memory of a device (e.g. a control unit).

The 5a , 5b and 5c show schematic representations of the machine learning model 5 to illustrate the retraining in different constellations.

In the 5a For example, only a complaint about component 7-3-1 (e.g. "vehicle model 1") of the last component section 5-3 can be derived from the maintenance data. Accordingly, only this component 7-3-1 is retrained with training data. In this case, the training data corresponds to an associated vehicle configuration. The parameters (weights and hyperparameters, etc.) of the other components are not changed during retraining. The remaining active components of an associated vehicle configuration are only used for inference.

In the 5b For example, a complaint about components 7-2-1 and 7-2-4 (e.g. "electric drive" and "SUV") of the middle component section 5-2 and components 7-3-2 (e.g. "vehicle model 2") and 7-3-3 (e.g. "vehicle model 3") can be derived from the maintenance data clustered by components. Accordingly, only these components 7-2-1, 7-2-4, 7-3-2, 7-3-3 are retrained with training data. In this case, the training data corresponds to the associated vehicle configurations. The parameters (weights and hyperparameters, etc.) of the other components are not changed during retraining. The remaining active components of an associated vehicle configuration are only used for inference.

In the 5c For example, a complaint can be derived from the maintenance data clustered by component, which affects all vehicle configurations. Accordingly, all components 7-1, 7-2-x, 7-3-x of all component sections 5-1, 5-2, 5-3 are then retrained with training data. In this case, the training data corresponds to all vehicle configurations.

After retraining, in particular, optimization, testing and release of the retrained machine learning model 5 are planned. The retrained machine learning model 5 is then transmitted to the respective vehicles 50 and used there.

list of reference symbols

1

device

2

steering size sensor

3

data processing facility

3-1

computing device

3-2

memory

4

steering size

5

trained machine learning model

5-1

first component section

5-2

middle component section

5-3

last component section

6

hands-off state

7-1

trained component

7-2-x

trained component

7-3-x

trained component

20

initial data

40

device

41

central server

41-1

computing device

41-2

memory

42

information

43

maintenance data

50

vehicle

51

steering wheel

60

steering system

99-102

measures of the procedure

200-202

measures of the procedure

Claims (13)

Method for detecting a hands-off state (6) on a steering wheel (51) of a vehicle (50), wherein at least one steering variable (4) is detected on the steering wheel (51), wherein the detected at least one steering variable (4) is fed to a trained machine learning model (5) as input data, wherein the machine learning model (5) is trained to detect a hands-off state (6) based on at least the to recognize the at least one steering variable (4) detected and to output an associated state variable as output data (20), wherein the trained machine learning model (5) is provided in such a way that the trained machine learning model (5) has at least three component sections (5-x) which follow one another with respect to a signal flow, wherein a trained component (7-1) is provided in a first component section (5-1), wherein at least one trained component (5-2-x) is provided in at least one middle component section (5-2) corresponding to a vehicle configuration of the vehicle (50), and wherein a trained component (7-3-x) corresponding to the vehicle configuration of the vehicle (50) is provided in a last component section (5-3). procedure according to claim 1 , characterized in that the vehicle configuration of the last component section (5-3) relates to a vehicle model. Device (1) for detecting a hands-off state (6) on a steering wheel (51) of a vehicle (50), comprising: at least one steering variable sensor (2) which is designed to detect at least one steering variable (4) on the steering wheel (51), and a data processing device (3), wherein the data processing device (3) is designed to receive the detected at least one steering variable (4), to provide a trained machine learning model (5) and to supply the detected at least one steering variable (4) to the trained machine learning model (5) as input data, wherein the machine learning model (5) is trained to detect a hands-off state (6) based on at least the detected at least one steering variable (4) and to output associated state information as output data (20), wherein the trained machine learning model (5) is provided in such a way that the trained machine learning model (5) has at least three component sections which follow one another with respect to a signal flow (5-x), wherein the first component section (5-1) provides a trained component (7-1), wherein at least one trained component (7-2-x) is provided in at least one middle component section (7-2) corresponding to a vehicle configuration of the vehicle (50), and wherein in a last component section (5-3) a trained component (7-3-x) corresponding to the vehicle configuration of the vehicle (50) is provided. Steering system (60) comprising a device (1) according to claim 3 . Vehicle (50) comprising a device (1) according to claim 3 and/or a steering system (60) according to claim 4 . Method for providing a trained machine learning model (5) for recognizing a hands-off state (6) on a steering wheel (51) of a vehicle (50), wherein the trained machine learning model (5) is provided for a plurality of vehicle configurations, wherein the trained machine learning model (5) is provided in such a way that the trained machine learning model (5) has at least three component sections (5-x) that follow one another with respect to a signal flow, wherein a trained component (7-1) is provided in a first component section (5-1) for all vehicle configurations, wherein at least one trained component (7-2-x) is provided in at least one middle component section (5-2) for at least a subset of all vehicle configurations, and wherein in a last component section (5-3) a trained component (7-3-x) is provided exclusively for a vehicle configuration corresponding thereto, and wherein for providing the trained machine learning model (5) for a given vehicle configuration, the respective trained components (7-1,7-2-x,7-3-x) are selected and deployed taking into account the vehicle configuration. procedure according to claim 6 , characterized in that the provision comprises loading the trained machine learning model (5) into a memory (3-2) of a control unit of the respective vehicle. procedure according to claim 6 or 7 , characterized in that trained components (7-2-x,7-3-x) of the trained machine learning model (5) which do not correspond to the vehicle configuration of the respective vehicle (50) are deactivated. Method according to one of the Claims 6 until 8 , characterized in that , in order to provide the trained machine learning model (5), information (42) relating to the vehicle configuration is recorded and/or obtained, wherein the trained machine learning model (5) is provided on the basis of the recorded and/or obtained information (42). Method according to one of the Claims 6 until 9 , characterized in that , in the case of a vehicle configuration that is not yet known, at least one component (7-2-x) for the at least one middle component section (5-2) and/or one component (7-3-x) for the last component section (5-3) is generated and trained and provided. Device (40) for providing a trained machine learning model (5) for recognizing a hands-off state (6) on a steering wheel (51) of a vehicle (50), comprising: a central server (41), wherein the central server (41) is set up to provide the trained machine learning model (5) for a plurality of vehicle configurations, and to provide the trained machine learning model (5) in such a way that the trained machine learning model (5) has at least three component sections (5-x) that follow one another with respect to a signal flow, wherein a trained component (7-1) is provided in a first component section (5-1) for all vehicle configurations, wherein at least one trained component (7-2-x) is provided in at least one middle component section (5-2) for at least a subset of all vehicle configurations, and wherein in a last component section (5-3) a trained component (7-3-x) is provided exclusively for a vehicle configuration corresponding thereto, and wherein, in order to provide the trained machine learning model (5) for a given vehicle configuration, the respective trained components (7-1,7-2-x,7-3-x) are selected and provided taking into account the vehicle configuration. Method for maintaining a totality of machine learning models (5) provided in vehicles (50) and trained to recognize a hands-off state (6), wherein recorded and/or collected maintenance data (43) relating to respective components (7-1,7-2-x,7-3-x) of at least three component sections (5-x) of the respective trained machine learning model (5) are received from the vehicles (50), wherein the received maintenance data (43) are clustered according to components (7-1,7-2-x,7-3-x), and wherein the machine learning model (5) is retrained taking into account the received maintenance data (43) in each case according to the clustered components (7-1,7-2-x,7-3-x) and the respectively retrained machine learning model (5) is provided. Device (40) for maintaining a totality of machine learning models (5) provided in vehicles (50) and trained to recognize a hands-off state (6), comprising: a central server (41), wherein the central server (41) is set up to receive maintenance data (43) recorded and/or collected by the vehicles (50) relating to respective components (7-1,7-2-x,7-3-x) of at least three component sections (5-x) of the respective trained machine learning model (5) from the vehicles (50), cluster the received maintenance data (43) according to components (7-1,7-2-x,7-3-x), and retrain the machine learning model (5) taking into account the received maintenance data (43) in each case according to the clustered components (7-1,7-2-x,7-3-x), and retrain the respective retrained machine learning model (5) to provide.

Images