DE102021110801A1 - Method, system and computer program for executing, suppressing or varying at least one measure in a vehicle - Google Patents

Abstract

The invention relates to a method, a neural network (40) and a vehicle (10, 11, 12) for executing, suppressing or varying at least one measure in a vehicle (10, 11, 12). In the method, at least one first measure is executed, suppressed or varied by means of a neural network (40), the neural network depending on data sections associated with the execution of a second measure in one or more second vehicles (20, 21, 22) correspond, set up, especially trained.

Description

The invention relates to a method, a system and a corresponding computer program for executing, suppressing or varying at least one measure, in particular a driving maneuver, in a vehicle.

It is the object of the present invention to provide an improved method, a system and a computer program for executing, suppressing or varying at least one measure, in particular a driving maneuver, in a vehicle. In particular, it is also the object of the invention to increase certain desired effects of a learning functionality of the vehicle and to avoid possible negative effects.

The object is solved by the features of each of the independent patent claims. Advantageous configurations are characterized in the dependent claims. It is pointed out that additional features of a patent claim dependent on an independent patent claim without the features of the independent patent claim or only in combination with a subset of the features of the independent patent claim can form a separate invention independent of the combination of all features of the independent patent claim, which can be made the subject of an independent claim, a divisional application or a subsequent application. This applies equally to the technical teachings described in the description, which can form an invention independent of the features of the independent patent claims.

According to a first aspect, the invention is characterized by a method for executing, suppressing or varying at least one measure, in particular a driving maneuver in a vehicle, wherein at least one first measure is executed, suppressed or varied by means of a neural network, the neural network is set up, in particular trained, depending on data sections that correspond to the execution of a respective second measure in one or more second vehicles.

In particular, a measure is to be understood as an operator action, reaction and/or action. In particular, a measure can be a reaction and/or action in connection with a specific object, event, situation or an object, event or situation of a specific type. In particular, it can be a measure in connection with a specific driving situation or a driving situation corresponding to a specific type (pattern).

The neural network is to be understood in particular as a so-called DNN (=deep neural network). Alternatively or additionally, the term "neural network" in the context of the present document is also to be understood as (any) other form of, in particular non-rule-based, machine learning and/or AI (=artificial intelligence).

In particular, the neural network is set up and/or trained to carry out, suppress or vary the measure in the first vehicle. The at least one measure in a first vehicle can be controlled or influenced depending on the output layer of the neural network. This is also to be understood in particular in such a way that a control signal for carrying out the at least one first measure is determined on the basis of the output layer of the neural network. For example, an actuator of the first vehicle can be controlled or influenced (directly or indirectly) depending on the output layer of the neural network.

Setting up the neural network is to be understood in particular as setting up an architecture, a structure of the neural network and/or the provision of data, e.g. checkpoint (weight information and/or bias terms) of the neural network. For example, the neural network can be set up in that a checkpoint of a neural network that has been trained (e.g. removed and/or based on the aggregated second data sections from one or more second vehicles) is loaded or updated on it.

For example, the neural network can be set up to update its (previous) weight information and/or bias terms depending on a correspondingly trained neural network. In this case, within the scope of the present document, this is to be understood as setting up, in particular as training, the neural network.

Alternatively or additionally, the neural network can be set up to influence, control and/or interact with another device in the vehicle, e.g. to control an actuator and/or an output device in the vehicle. For example, a control signal, a target value and/or parameters of a regulation are controlled by the neural network.

• - eine Unterstützung einer fahrerischen Handlung, und/oder
• - eine Ausführung einer Beeinflussung der Bewegung des ersten Fahrzeugs, und/oder
• - ein Generieren einer Information für den Fahrer des ersten Fahrzeugs, und/oder
• - ein Ausführen eines zumindest teilweise automatisierten Fahrmanövers, Einparkmanövers und/oder Ausparkmanövers, und/oder
• - eine Auswahl eines Parameters oder eines Modus des Fahrzeuglichts.

For example, the at least one first measure relates to the first vehicle one or more entries from the following list:

• - support for a driving action, and/or
• - executing an influencing of the movement of the first vehicle, and/or
• - generating information for the driver of the first vehicle, and/or
• - executing an at least partially automated driving maneuver, parking maneuver and/or parking maneuver, and/or
• - a selection of a parameter or mode of the vehicle light.

The first measure can be at least part of the driving maneuver and/or an action accompanying a driving maneuver, in particular a signaling of the vehicle (corresponding to the driving maneuver or accompanying the driving maneuver) to the outside (flashing, light signals, projections, generating a display, e.g. image display outwards).

• - eine Einfahrt in eine Schnellstraße,
• - eine Ausfahrt aus einer Schnellstraße,
• - einen Spurwechselvorgang,
• - einen Einfädelvorgang,
• - einen Überholvorgang,
• - einen Wendevorgang.

For example, the first measure can include a driving maneuver or part of a (multi-part) driving maneuver from the following list:

• - an entrance to an expressway,
• - an exit from an expressway,
• - a lane change operation,
• - a threading process,
• - an overtaking manoeuvre,
• - a turning operation.

Carrying out the at least one first measure can preferably relate to one or more specific driving situations and/or road sections.

For example, the neural network can be trained to carry out the at least one first measure in a first vehicle at a specific position within a section of road and/or when recognizing a specific driving situation as similarly as possible and/or to achieve a result that is as similar as possible to such a measure has already been carried out, in particular frequently, by one or more drivers and/or during one or more trips with a high performance, e.g. a performance that exceeds at least a first threshold value.

For example, the neural network can be set up, in particular trained, to cause (e.g. to control or influence and/or generate a control signal) that the at least one first measure is not carried out in a first vehicle (e.g. avoided and/or suppressed) or is replaced by an alternative first measure or is carried out with a modified first measure parameter. This can be done in particular if the neural network has been trained in such a way that this measure is never or very rarely used in a defined (e.g. high) number of the second vehicles in otherwise very successful, high-performance trips and/or very successful, high-performance drivers had happened.

The invention opens up the possibility of achieving a behavior of a first vehicle, e.g. an automated vehicle, which comes close to the behavior of a driver with above-average success and/or one or more trips with above-average success. At the same time, a correspondingly trained neural network can reduce or prevent a repetition of errors (caused by drivers and/or vehicles) (also to be understood as inconsistencies). Furthermore, it is also possible to avoid errors that have never been made or made by a second driver.

This can be achieved, for example, by at least largely not using the data from the use of one or more second vehicles by low-performing drivers or in the case of low-performing trips when training the neural network (e.g. filtered out, excluded) and/or in the sense of a negative Feedbacks (e.g. to punish the neural network) can be used.

The invention also opens up the possibility of achieving a human-like design, suppression or variation of measures or the resulting effect. The result can have an anticipatory effect (e.g. on the user of the first vehicle and/or the environment, e.g. other road users) or be anticipatory or "human", or have the effect of the best drivers or journeys. The training can (also explicitly) be carried out with the aim of achieving a "more human" behavior of the first vehicles when executing the first measure.

In this way (in addition to objectively measurable performance), the acceptance of the at least partially automated (first) vehicle for the user of the first vehicle and/or for the environment, e.g. other road users and/or for potential users (e.g. buyers), can be significantly increased .

The at least one first measure (in a first vehicle) and/or a second measure (in one or more second vehicles) can preferably be (entirely or predominantly) manually executed control, in particular a continuous change in a variable relevant for driving the vehicle, for example a vehicle steering system.

In a simplified case, the first measure and/or a second measure includes an operating process, a switching on, switching off and/or switching process. In particular, the at least one first measure can relate to a user functionality of the vehicle, e.g. a so-called customer function.

In a (simplified) example, the measure can include or be a gear shift, e.g. support for the driver with a gear shift, e.g. by means of haptic information.

The at least one measure parameter of the at least one first and/or a second measure preferably includes at least one (temporal and/or spatial) change function of a variable relevant for driving the vehicle.

For example, the first measure parameter can include a function indicating a change in the steering angle, wheelbase angle, pedal angle, etc., depending on a time-related variable (e.g. running seconds) and/or depending on a space-related variable (e.g. running meters). The time-related variable or space-related variable can be in relation to specific values, e.g. at the start of a maneuver and/or to a road user with whom the first vehicle is interacting.

Alternatively or additionally, a first and/or second measure can include activating a specific system or a specific functionality of the vehicle, setting it, activating it in a specific mode and/or bringing about a specific state transition of the vehicle system.

Particularly preferably, the at least one first measure and/or the at least one second measure and/or optionally the at least one alternative first alternative measure (third measure) relates to a process that is extended or continuous in terms of time and/or space (in a specific way). For example, the at least one first measure parameter and/or the at least one second measure parameter can characterize a (specific) temporal and/or spatial change or a course of a physical variable, e.g. in relation to the movement (so-called path guidance) of the vehicle.

Executing, suppressing and/or varying the at least one first measure in a first vehicle can be controlling, regulating or influencing the first measure in any way. However, the method is not limited to controlling or regulating a variable, for example in the sense of control technology or regulation technology.

Executing, suppressing and/or varying the at least one first measure can be or include (deterministic or non-deterministic) influencing the selection of the type of the first measure, the variant of the first measure and/or the determination or adjustment of a first measure parameter .

A variation of the first measure by the neural network can also be a determination, preference and/or a limitation of a solution space or a tendency to carry out, suppress and/or vary the at least one first measure.

The suppression of the at least one first measure can include or be a suppression or variation of one or more parameter values, parameter limit values and/or patterns of the respective measure (which could in principle be used in the first vehicle, for example). For example, the suppression of the at least one measure can include or be a limitation of the effect, the performance, of performance features, etc. of the measure. In other words, the at least one first measure can also be suppressed only to a certain extent or to a specific extent (e.g. determined by the neural network).

The at least one first measure can include a specific action, in particular an operator action, preferably a driver action, a combination or a specific sequence of a plurality of (different) operator actions or driver actions. For example, a measure parameter of a first measure can be a specific order or sequence of specific actions, in particular operator actions or driver actions. The driving action is in particular an action that is more or less directly related to driving the respective vehicle. The driving action preferably relates to a so-called trajectory guidance of the vehicle.

A first measure parameter and/or a second measure parameter of the first or second measure can be a qualitative and/or quantitative parameter.

A quantitative measure parameter can, for example, identify a selection from several alternative variants of a measure and/or a mode of a specific vehicle system (e.g. related to a possible second measure).

A quantitative measure parameter can, for example, characterize a quantitative, e.g. temporal and/or spatial, dimensioning of the first measure or second measure.

A first measure parameter and/or a second measure parameter of the first or second measure can, for example, be a time parameter indicative of a specific time interval, a point in time or a specific time dependency. For example, the at least one measure parameter can identify a suitable, unsuitable or particularly suitable time interval or a suitable, unsuitable or particularly suitable point in time for executing and/or ending a specific phase of the at least one second measure.

In one example of the method, the controller can propose, cause, change, prevent and/or carry out a signaling of the at least one first vehicle to the outside, in particular with a first action parameter, in particular depending on one or more second action parameters of a signaling of one or several second vehicles to the outside, which in the past (also according to other objective criteria) completed particularly good journeys and/or were served by particularly good drivers.

In particular, the at least one first vehicle is a different vehicle than the one or more second vehicles, at least with regard to an embodiment of the method for using the data provided. The method is preferably applied to a specific number of first vehicles and a specific number (specific or selected according to specific criteria) of second vehicles, with the number of first vehicles at least for one execution (one iteration) of the method or the method steps not having the Number of second vehicles matches. In particular, the method can be applied such that a second vehicle is a vehicle that has already been a first vehicle in the past. For example, a first time interval at which a vehicle is a first vehicle may differ from a second time interval at which a vehicle is a second vehicle. Thus, a step-by-step, iterative consideration of the information dependent on the selected data sections can be carried out. For example, learning based on the selected data sections can (also) be advantageously implemented. Only in exceptional cases can the second vehicle also be a first vehicle in relation to a time interval.

The first vehicle is preferably an at least predominantly automated vehicle or the first measure is a measure that is carried out at least predominantly automatically. Preferably, the one or more second vehicles are at least predominantly manual vehicles or the one or more second measures are at least predominantly manually executed measures. The method can also be applied to conventional vehicle technology or a manual driver.

According to a further refinement, the neural network is a recurrent neural network which is set up in such a way that execution, suppression or variation of the at least one first measure in the first vehicle is also dependent on the result of the execution, suppression or variation of a first measure carried out in the first vehicle in the past takes place.

According to a further embodiment, the neural network is set up, in particular trained, to carry out, suppress or vary at least one first measure in a vehicle, the neural network being trained on a measure of success and/or performance of the carrying out, suppressing or varying the first measure.

The result, measure of success and/or the performance of the first measure can (in each case) be assessable according to a criterion. For example, the neural network (depending on the second data sections) is set up, in particular trained, to strive for a quality of the result of the at least one first measure that corresponds to a target value. For example, the neural network can also be trained to achieve an increased target value or to achieve a measure of success corresponding to a target value (e.g. one or more specific measures of success), and/or to achieve an increased performance or a performance corresponding to a target value during execution, suppression or varying the (first) measure.

For example, the neural network can be a neural network that trai is intended to execute, suppress or vary at least one first measure in a vehicle, with reaching a target value of one or more measures of success and/or a performance (also to be understood as one or more measures of performance) in relation to a result, in particular a Result is sought in connection with the execution, suppression or variation of at least one first measure in a vehicle. The target value can, for example, correspond to or be based on a local or global maximum. In particular, the target value can be understood as exceeding a predetermined threshold and/or increasing or maximizing a driving performance indicator.

According to a further refinement, the neural network is trained as a function of one or more second driving performance indicators of one or more second vehicles when driving on a journey section and/or a section of actions by a user, in particular the driver of the second vehicle.

In particular, the one or more second drivers (e.g. selected according to the criterion presented herein) may be second drivers of one or more second vehicles. The term driver means in particular a user of the vehicle (e.g. passenger or user of an automated vehicle).

According to a further refinement, the neural network is trained with data sections which are determined given a predefined number of users and/or journey sections of the one or more second vehicles.

The data sections can be selected as a function of results, measures of success and/or performances, in particular driving performance indicators, corresponding to the data sections.

Alternatively or additionally, those data sections can be selected which correspond to driving on a journey section and/or to a section of the driving activity of a user of the second vehicle.

According to a further embodiment, varying the first measure in the at least one first vehicle includes determining or adjusting a first measure parameter of a first measure and/or selecting an alternative first alternative measure to a first measure selected in particular by the user of the vehicle.

According to a further embodiment, a first measure parameter of a first measure is determined or adjusted, in particular a coefficient, a function or a gradient for changing the measure parameter is determined or adjusted, and/or an alternative first alternative measure is selected for the first measure, in particular a Selection of a criterion for selecting the alternative first alternative measure is controlled depending on the output layer of the neural network.

• - eine Unterstützung einer fahrerischen Handlung des Nutzers des ersten Fahrzeugs;
• - eine Ausführung einer Beeinflussung der Bewegung des ersten Fahrzeugs;
• - ein Generieren einer Information für den Fahrer des ersten Fahrzeugs;
• - ein Ausführen eines zumindest teilweise automatisierten Fahrmanövers, Einparkmanövers, Ausparkmanövers und/oder Rangiermanövers;
• - eine Auswahl eines Parameters oder Modus des Fahrzeuglichts des ersten Fahrzeugs;
• - eine Signalisierung von dem ersten Fahrzeug nach außen.

According to a further embodiment, the at least one first measure in the first vehicle relates to one or more entries from the following list (the term “refer” is also to be understood in the sense of “being included” or “forming part of”):

• - A support of a driving action of the user of the first vehicle;
• - an execution of influencing the movement of the first vehicle;
• - generating information for the driver of the first vehicle;
• - Executing an at least partially automated driving maneuver, parking maneuver, parking maneuver and/or maneuvering;
• - a selection of a parameter or mode of the vehicle light of the first vehicle;
• - signaling from the first vehicle to the outside.

• - ein Lichtsignal in eine bestimmte Raumrichtung,
• - ein Tonsignal in einen bestimmte Raumrichtung,
• - ein Anzeigen einer zweidimensionalen oder dreidimensionalen Grafik und/oder einer fotorealistischen oder eines fotorealistisch verwandelten Bildes,
• - ein Anzeigen einer beweglichen Darstellung, beispielsweise einer Animation oder einer Bildsequenz und/oder
• - ein Veranlassen einer Darstellung eines Abschnitts einer virtuellen Realität für zumindest einen Verkehrsteilnehmer.

For example, a first measure can thus include support for signaling the vehicle to the outside. For example, said signaling to the outside can include the following signaling forms and/or signaling types:

• - a light signal in a specific spatial direction,
• - a sound signal in a specific spatial direction,
• - a display of a two-dimensional or three-dimensional graphic and/or a photorealistic or a photorealistic transformed image,
• - displaying a moving representation, for example an animation or an image sequence and/or
• - causing a representation of a section of a virtual reality for at least one road user.

• - einen Inhalt, bzw. ein inhaltliches Merkmal der Signalisierung,
• - einen Zeitpunkt bzw. ein Zeitintervall, eine Intensität, einen optischen und/oder akustischen Parameter der Signalisierung,
• - einen bestimmten Raumbereich und/oder zu einem bestimmten Verkehrsteilnehmer.

At least one second measure parameter can be determined or adjusted by the neural network. B. is characteristic of:

• - a content or a content-related feature of the signaling,
• - a point in time or a time interval, an intensity, an optical and/or acoustic parameter of the signaling,
• - a specific spatial area and/or to a specific road user.

For example, the neural network in a first vehicle can also display specific shapes, images, projections and/or output tone or frequency sequences with specific parameters to the outside.

For example, the first measure parameter can also identify for which specific, exclusive spatial angle range a signaling is to be output to the outside, in particular which signaling is to be output to the outside.

The signaling from the vehicle to the outside can in particular represent a specific, e.g. planned, action by the first vehicle or an occupant of the first vehicle and/or an action recommended or to be avoided for a road user in the vicinity of the first vehicle. This can be the signaling for a specific road user within the first driving situation, e.g. for a crossing vehicle, a vehicle to be overtaken, a pedestrian, a cyclist, a road user who plays a specific role within the first traffic situation.

• - ein Merkmal einer Spurführung, insbesondere einer Abweichung von einer Spurmitte, von einer Ideallinie und/oder in Bezug auf ein Einhalten des Abstands zu seitlichen Objekten,
• - ein Merkmal einer Längsführung, insbesondere in Bezug auf ein Einhalten einer Distanz, einer Relativgeschwindigkeit und/oder einer Relativbeschleunigung zu einem weiteren Verkehrsteilnehmer,
• - ein Risiko, insbesondere ein Kollisionsrisiko, dessen Konsequenz eingetreten und/oder nicht eingetreten ist,
• - einen Verlauf einer einwirkenden Beschleunigungen auf das Fahrzeug, insbesondere auf die Fahrgastzelle,
• - ein Signalisieren anderer Verkehrsteilnehmer, insbesondere ein Signalisieren mittels akustischer Warnsignale oder Lichtwarnsignale,
• - ein Maß der Einhaltung und/oder Abweichung vom Einhalten von, insbesondere im jeweiligen Fahrbahnabschnitt, geltenden Verkehrsregeln,
• - ein Maß des Erfolgs eines oder mehrerer ausgeführter Maßnahmen oder Fahrmanöver des Fahrzeugs und/oder
• - ein Maß des Energieverbrauchs und/oder ein Maß für eine Energieeffizienz des Fahrzeugs.

According to a further embodiment, the (first or second) measure of success and/or the performance of a first or second measure is characteristic of:

• - a feature of lane guidance, in particular a deviation from a lane center, from an ideal line and/or in relation to maintaining a distance from objects on the side,
• - a feature of a longitudinal guidance, in particular with regard to maintaining a distance, a relative speed and/or a relative acceleration to another road user,
• - a risk, in particular a risk of collision, the consequence of which has occurred and/or has not occurred,
• - a course of accelerations acting on the vehicle, in particular on the passenger cell,
• - signaling other road users, in particular signaling by means of acoustic warning signals or light warning signals,
• - a degree of compliance with and/or deviation from compliance with traffic regulations applicable, in particular in the respective section of the roadway,
• - a measure of the success of one or more actions or maneuvers carried out by the vehicle and/or
• - a measure of the energy consumption and/or a measure of an energy efficiency of the vehicle.

The at least one performance can preferably be determined from at least two different types of variables, in particular from at least two of the variables mentioned. For example, the at least one performance can be determined from a time for a roadway section in relation to the corresponding compliance with the traffic rules for the same roadway section. There can be any desired (many selectable), appropriate, mathematical dependency between the variables on which the (respective) performance indicator or the performance is based.

Furthermore, different measures of performance, e.g. those intended or optimized for different purposes, can be determined and taken into account. These can relate, for example, to “sportiness”, “safety”, “driving comfort” and/or “energy efficiency”.

At least two different measures of performance are particularly preferably determined and/or taken into account.

Alternatively or additionally, the at least one measure of performance can be a two-dimensional variable. In particular, the at least one performance includes and/or is based on at least two values that are essentially independent of one another—at least within a certain value range—in particular on mathematically orthogonal values. For example, at least two measures of performance can be determined (approximately for the same journey segment and/or user). These can each characterize different aspects of performance.

For example, the performance or an associated performance indicator includes one or more variables that are characteristic of a measure of a (scientifically, mathematically objectified) driving quality or driving skill of the at least one (second) driver of the one or more second vehicles and/or or for a measure of ride quality at a or multiple trips with the one or more second vehicles.

In particular, one or more variables and/or criteria can be taken into account in a performance, which, in their interaction or in their relationships to one another, give people, e.g. statistically selected people, experts, test subjects and/or certain so-called customer groups, an impression of a produce good drivers or good driving skills. A corresponding measure of performance can be determined or taken into account as a function of the variables and/or criteria according to a predetermined mathematical dependency.

The at least one measure of performance may relate to a specific driving style and/or use case of the vehicle, or comprise an identifier for one or more different driving styles or use cases to which it refers.

The consideration, in particular a weighting, of specific underlying variables in a performance indicator can preferably be different for different driving styles. For example, a certain driver performance indicator with a focus on "EFFICIENCY" shows a significantly higher dependency on energy consumption or wear and tear with energy consumption or with a parameter that correlates with wear and tear than another measure of performance with a focus on "SPORT". For example, another specific driver performance indicator with a focus on "SAFETY" shows a significantly higher dependency on risks or certain disadvantages that occur in connection with driving or with a parameter that correlates with risks or disadvantages than a measure of performance for "SPORT".

According to a further embodiment, the measure of success and/or the performance, in particular the driving performance, relates to at least partially automated driving and/or to at least partially manual driving.

According to a further refinement, the neural network is trained to carry out, suppress or vary at least one first measure within a specific route, on specific roadway and/or lane arrangements and/or under specific boundary conditions when driving on a route.

According to a further embodiment, the first measure includes determining or adjusting a degree of automation of a driving function, the degree of automation relating to the journey of the first vehicle in the current road section or in a road section ahead and/or to the execution of the first measure in the first vehicle in the current road section or in a road section ahead section of the road relates.

According to a further embodiment, the measure of success and/or the performance and/or the corresponding data sections relate to a specified number of trips by the one or more second vehicles, in particular in a specific application of the one or more second vehicles, and/or on a driver of the second vehicle, in particular on a specific driver or on a driver from a group of drivers or on a driver with a measure of success that exceeds a certain level or with a performance that exceeds a certain level, in particular driving performance.

According to a further refinement, the data sections are determined from a large number of data sections from the second vehicles, which in particular are part of a customer fleet.

According to a further refinement, the at least one first measure is an at least partially new measure and/or a measure which has never or seldom been carried out in a large number of the second vehicles.

The invention can also expediently succeed in executing, suppressing or varying such measures which, for example, would not be possible in an expedient manner with a conceivable rule-based approach.

• Das trainierte neuronale Netz kann so eingerichtet, insbesondere trainiert werden, dass ein Erfolgsmaß beim Ausführen, Unterdrücken und/oder Variieren einer ersten Maßnahme in einem ersten Fahrzeug erhöht, insbesondere zu einem Zielwert hin verändert werden kann. Beispielsweise kann somit eine resultierende Performance, mit der die erste Maßnahme in dem ersten Fahrzeug ausgeführt wird, bzw. eine resultierende Performance des ersten Fahrzeugs viel besser als etwa eine Durchschnitts-Performance aus der Fahrweise verschiedener Fahrer und/oder auf unterschiedlichen Fahrten sein.

Based on the features described in this document, the following advantageous advantages or developments are possible:

• The trained neural network can be set up, in particular trained, in such a way that a measure of success when executing, suppressing and/or varying a first measure in a first vehicle can be increased, in particular changed towards a target value. For example, a resulting performance with which the first measure is carried out in the first vehicle, or a resulting performance of the first vehicle, can be much better than an average performance from the driving style of different drivers and/or on different journeys.

The result, in particular a resulting performance of the neural network or an off controlled by the neural network Guidance, suppression and/or a variation of at least one measure in a first vehicle can be better predicted or can correspond to the expectation or the goal of the training with an increased degree of probability.

The amount of data that is or must be used to train the neural network can be drastically reduced, for example by a factor of 10 ² , 10 ³ , 10 ⁴ , 10 ⁵ , 10 ⁶ , . . .

The data for training the neural network can be determined from a fleet, in particular a customer fleet, of (second) vehicles. These vehicles can be vehicles that have already been delivered and are in regular driving operation. This results in the advantage that the data sections correspond to the use cases, roads, concrete actions and/or journeys that have occurred in practice with the second vehicles (and are therefore relevant with a high probability for the at least one first vehicle). In addition, there is a further advantage in that the data sections can be generated very easily from data that is generated anyway (almost without costs). As a result, data for training the neural network and consequently also the trained neural network can be set up, in particular trained, in a very economical manner. The invention can be used to control an - otherwise enormous - data flood that is determined in the second vehicles, stored, transmitted between vehicles and/or to the backend, processed, and/or transmitted to at least a first vehicle, and thereby a saving hardware costs, transmission costs, risks of delayed processing.

At least partially automated first measures can be carried out in a much more "human-like" or "human" manner, which means that the user's acceptance of the measures can be increased.

The invention can (also freely and universally) be applied to a large number of diverse functions and services.

Also, incidentally, very valuable data can be obtained in the method based on the data sections (which in themselves have a high value). These can also be used for other purposes.

The invention also includes all functionalities that can be executed in a vehicle, which are operated or made operable (also to be understood as executable) on the basis of the trained neural network - for example for a corresponding (operated by the neural network) at least partially automated driving or parking .

In the context of the present document, the term "functionality" can also be understood as a performance characteristic (also to be understood as a feature) of a (larger) function, e.g. a user function or equipment of the vehicle. Functionality can also be, for example, an automatic activation, deactivation and/or change of a (further) functionality or driver information that suggests or announces an activation, deactivation or change of a functionality. For example, an (automatically generated) suggestion to the driver of the vehicle to switch on or off a certain functionality or to change it in a certain way can also be regarded as a functionality. A functionality is to be understood in particular as a functionality that is (initially at least potentially) useful, usable and/or executable for the user of the vehicle.

The driver of an at least partially automated vehicle is to be understood in particular as the user (or multiple users) of the (in particular automated) vehicle.

• - ein erstes Fahrzeug umfassend das neuronale Netz, und
• - ein oder mehrere zweite Fahrzeuge, insbesondere eine Vielzahl an zweiten Fahrzeuge, und wobei das System ausgebildet ist das neuronale Netz abhängig von Datenabschnitten, die mit dem Ausführen von jeweils einer zweiten Maßnahme in einem oder mehreren zweiten Fahrzeugen korrespondieren, zu trainieren.

According to a further aspect, the invention is characterized by a system for executing, suppressing or varying at least one first measure in a first vehicle using a neural network, the system comprising:

• - a first vehicle comprising the neural network, and
• - One or more second vehicles, in particular a large number of second vehicles, and the system is designed to train the neural network as a function of data sections which correspond to the execution of a second measure in one or more second vehicles.

According to a further aspect, the invention is characterized by a neural network for use in a first vehicle, the neural network being selected and/or trained and/or updated according to a method according to any one of claims 1-14. The neural network can be designed and/or trained to carry out, suppress or vary a first measure when it is executed on a computing unit of the first vehicle.

The neural network can be designed as an update of a (previous) neural network, which, for example, as part of a radio tion extension, for example as part of a feature update - is updated - preferably wirelessly.

The neural network can be trained (specifically, e.g. through a correspondingly selected training goal) to carry out, suppress or vary the first measures so that the first measures are carried out as much as possible as by the selected drivers and/or on the best trips and/or or that the result of the first measures is as close as possible to that of the best drivers.

In a further example, also (or only) data sections can be selected, made available and/or used for training the neural network, which do not or do not directly correspond to the or the same measure or action that is later to be carried out by the trained neural network .

For example, the at least one performance that was determined from a number of successes, risks, consequences, positive results and/or negative results that occurred when using the vehicle by a specific driver or during the said journeys, particularly preferably retrospectively (actually retrospectively).

In particular, a measure of performance in relation to a driver and/or a trip can be determined from a large number of measures of success within one or more longer driving routes and/or time intervals. Alternatively or additionally, a measure of performance can also relate to a large number of route sections, driving situations, boundary conditions and/or to significantly longer time intervals.

For example, the at least one performance can be determined as a function of a frequency of one or more second driving sections and/or drivers in one or more second vehicles with a measure of success that exceeds a first limit value and/or a measure of success that falls below a second limit value. The second measures can be the same, related or not the same, not related, or different measures, in particular related to different functionalities and/or systems.

According to a further aspect, the invention is distinguished by a vehicle, preferably a motor vehicle, with which the neural network is operated.

The one or more first vehicles and/or one or more second vehicles are preferably a motor vehicle or motorcycle. The invention includes one or more first vehicles and/or second vehicles, the neural network, and/or a correspondingly spaced computing unit, e.g. a so-called backend, in which, for example, the neural network is trained, in particular by determining checkpoints for operating the neural network in the first vehicle will. Furthermore, the invention includes a system comprising a number of first vehicles and a number of second vehicles and/or the remote computing unit for operating the vehicles. Furthermore, the invention includes all functionalities of the first vehicle that are operated dependently or on the basis of the trained neural network or can be operated expediently.

• 1 eine Ausgestaltung eines erfindungsgemäßen Systems zum Ausführen, Unterdrücken oder Variieren zumindest einer ersten Maßnahme in einem ersten Fahrzeug. , und
• 2 ein beispielhaftes Ablaufdiagramm zur Veranschaulichung des erfindungsgemäßen Verfahrens bzw. des korrespondierenden Computerprogramms.

Exemplary embodiments of the invention are explained in more detail below with reference to the schematic drawings, without restricting the generality. These can be freely combined with the features described above. Show it:

• 1 an embodiment of a system according to the invention for executing, suppressing or varying at least one first measure in a first vehicle. , and
• 2 an exemplary flowchart to illustrate the method according to the invention and the corresponding computer program.

the 1 shows a system for executing, suppressing or varying at least one first measure in a first vehicle 10, 11 and 12, in which the at least one first measure is executed, suppressed or varied by means of a neural network 40, in particular a recurrent neural network, the neural network 40 receiving relevant information or data from a backend 30 via an interface. The backend in turn receives information from a number of second vehicles 20, 21 and 22. Further details of the system, the neural network 40 and the backend 30 are explained in more detail in the context of the following description.

This in 2 The exemplary method shown or a corresponding program or a sequence of different programs (eg corresponding to certain steps of the method) can be executed by the system, in particular on one or more computing units of the system.

A variant of the method is explained in detail below. The features of the present example description can be combined in many ways with all the features described in this document.

The terms “journey” and/or “driver” or the “number of drivers” and/or “number of journeys” always refer to specific journeys and/or drivers and/or journeys and/or drivers selected according to specific criteria.

This in 2 Method 100 illustrated serves to execute, suppress or vary at least one first measure in a first vehicle, wherein the execution, suppression or variation of the at least one first measure takes place by means of a neural network.

The first measure can include determining or adapting a degree of automation of a driving function, with the degree of automation relating to the journey of the first vehicle 10, 11, 12 in the current road section or a road section ahead and/or to the execution of the first measure in the first vehicle 10, 11, 12 in the current road section or in a road section ahead.

• - eine Unterstützung einer fahrerischen Handlung des Nutzers des ersten Fahrzeugs;
• - eine Ausführung einer Beeinflussung der Bewegung des ersten Fahrzeugs;
• - ein Generieren einer Information für den Fahrer des ersten Fahrzeugs;
• - ein Ausführen eines zumindest teilweise automatisierten Fahrmanövers, Einparkmanövers, Ausparkmanövers und/oder Rangiermanövers;
• - eine Auswahl eines Parameters oder Modus des Fahrzeuglichts des ersten Fahrzeugs;
• - eine Signalisierung von dem ersten Fahrzeug nach außen.

Alternatively or additionally, the at least one first measure in the first vehicle 10, 11, 12 can relate to one or more entries from the following list:

• - A support of a driving action of the user of the first vehicle;
• - an execution of influencing the movement of the first vehicle;
• - generating information for the driver of the first vehicle;
• - Executing an at least partially automated driving maneuver, parking maneuver, parking maneuver and/or maneuvering;
• - a selection of a parameter or mode of the vehicle light of the first vehicle;
• - signaling from the first vehicle to the outside.

A decision can be made in the method about carrying out, suppressing or varying the at least one measure (in a specific case).

Alternatively or additionally, depending on the output layer of the neural network 40, one or more parameter values, parameter limit values, patterns of the respective measure in a first vehicle 10, 11, 12 can be used, suppressed or varied. For example, the suppression of the at least one measure can include or be a limitation of the effect, performance, etc. of the measure.

Alternatively or additionally, at least one first measure can be an at least partially new measure and/or a measure which has never or seldom been carried out in a large number of the second vehicles.

• Schritt 101: Ermitteln von Datenabschnitte mittels des einen oder der mehreren (bevorzugt Vielzahl) zweiten Fahrzeuge. Diese Datenabschnitte können z.B. vom Bordnetz, Speicher,
• Steuergerät und/oder Sensor eines oder mehrerer zweiter Fahrzeuge ermittelt (vereinfacht:
  • auswahlweise erfasst) werden. In diesem Beispiel handelt es sich um Daten im Zusammenhang mit einer Fahrt, insbesondere in einem bestimmten Anwendungsfall des Fahrzeugs und/oder im Zusammenhang mit der Nutzung des einen oder der mehreren zweiten Fahrzeuge durch einen bestimmten (anhand seiner Performances ausgewählten) Fahrer.

This method 100 may include one or more of the following described steps or features in any combination or order:

• Step 101: determining data sections by means of the one or more (preferably a plurality) of second vehicles. These data sections can, for example, come from the vehicle electrical system, memory,
• Control unit and/or sensor of one or more second vehicles determined (simplified:
  • selectively recorded). In this example, it is data in connection with a journey, in particular in a specific use case of the vehicle and/or in connection with the use of the one or more second vehicles by a specific driver (selected on the basis of his performances).

The data sets can be data sets that correspond to the execution of a second measure in one or more second vehicles. Alternatively or additionally, data records can be determined for a predetermined number of users and/or journey sections of the one or more second vehicles. In particular, the data sections can be determined from a large number of data sections of the second vehicles, which are part of a customer fleet, for example.

Step 102: Training the neural network depending on data sections that correspond to the execution of a second action in one or more second vehicles.

Step 102a (alternatively or in addition to step 102): training the neural network depending on one or more second driving performance indicators of one or more second vehicles when driving on a journey section and/or a section of actions by a user of the one or more second vehicles.

Step 102b (alternatively or in addition to steps 102 and 102a): training the neural network with data sections that were determined for a predetermined number of users and/or journey sections of the one or more second vehicles

Step 102c (alternatively or in addition to step 102 - 102b) training the neural network for a measure of success and/or for a performance of executing, suppressing or varying the first measure with data sections that occur for a predetermined number of users and/or journey sections of the one or the plurality of second vehicles have been determined. For example, the neural network is trained as a function of data sections that correspond to the execution of a second measure in one or more second vehicles.

The measure of success and/or the performance can relate to at least partially automated driving and/or to at least partially manual driving.

Alternatively or additionally, the measure of success and/or the performance and/or the corresponding data sections can relate to a specified number of trips by the one or more second vehicles, in particular in the case of a specific application of the one or more second vehicles, and/or to one driver of the second vehicle, in particular to a specific driver or to a driver from a group of drivers or a driver with a measure of success that exceeds a certain level or a performance that exceeds a certain level, in particular driving performance.

• - ein Merkmal einer Spurführung, insbesondere einer Abweichung von einer Spurmitte, von einer Ideallinie und/oder in Bezug auf ein Einhalten des Abstands zu seitlichen Objekten,
• - ein Merkmal einer Längsführung, insbesondere in Bezug auf ein Einhalten einer Distanz, einer Relativgeschwindigkeit und/oder einer Relativbeschleunigung zu einem weiteren Verkehrsteilnehmer,
• - ein Risiko, insbesondere ein Kollisionsrisiko, dessen Konsequenz eingetreten und/oder nicht eingetreten ist,
• - einen Verlauf einer einwirkenden Beschleunigungen auf das Fahrzeug, insbesondere auf die Fahrgastzelle,
• - ein Signalisieren anderer Verkehrsteilnehmer, insbesondere ein Signalisieren mittels akustischer Warnsignale oder Lichtwarnsignale,
• - ein Maß der Einhaltung und/oder Abweichung vom Einhalten von, insbesondere im jeweiligen Fahrbahnabschnitt, geltenden Verkehrsregeln,
• - ein Maß des Erfolgs eines oder mehrerer ausgeführter Maßnahmen oder Fahrmanöver des Fahrzeugs,
• - ein Maß des Energieverbrauchs und/oder ein Maß für eine Energieeffizienz des Fahrzeugs.

Alternatively or additionally, the measure of success and/or performance of a first or second measure may be indicative of:

• - a feature of lane guidance, in particular a deviation from a lane center, from an ideal line and/or in relation to maintaining a distance from objects on the side,
• - a feature of a longitudinal guidance, in particular with regard to maintaining a distance, a relative speed and/or a relative acceleration to another road user,
• - a risk, in particular a risk of collision, the consequence of which has occurred and/or has not occurred,
• - a course of accelerations acting on the vehicle, in particular on the passenger cell,
• - signaling other road users, in particular signaling by means of acoustic warning signals or light warning signals,
• - a degree of compliance with and/or deviation from compliance with traffic regulations applicable, in particular in the respective section of the roadway,
• - a measure of the success of one or more actions or driving maneuvers carried out by the vehicle,
• - a measure of the energy consumption and/or a measure of an energy efficiency of the vehicle.

Step 102d (as an alternative to or in addition to steps 102-102c): Training the neural network to execute, suppress or vary at least one first measure within a specific route, on specific roadway and/or lane arrangements and/or under specific boundary conditions when driving on a route .

Step 103: Executing, suppressing or varying at least one first measure in the first vehicle using the trained neural network (40). Advantageously, the at least one first measure can also be executed, suppressed or varied in the first vehicle depending on the result of the execution, suppressed or varied of a first measure executed in the first vehicle in the past.

When varying the first measure in the at least one first vehicle, a determination or adjustment of a first measure parameter of a first measure and/or a selection of an alternative first alternative measure to a first measure can be included. A first measure parameter (e.g. a coefficient, a function or a gradient for changing the measure parameter) can be determined or adjusted and/or an alternative first alternative measure to the first measure can be selected, in particular a selection of a criterion for selecting the alternative first alternative measure takes place or can be controlled depending on the output layer of the neural network.

Claims (16)

Method (100) for executing, suppressing or varying at least one first measure in a first vehicle (10, 11, 12), in which the at least one first measure is executed, suppressed or varied by means of a neural network (40), the neural network is set up, in particular trained, as a function of data sections which correspond to the execution of a second measure in each case in one or more second vehicles (20, 21, 22). procedure after claim 1 , wherein the neural network is a recurrent neural network (40) and the neural network (40) is set up in such a way that execution, suppression or variation of the at least one first measure in the first vehicle (10, 11, 12 ) also depends on the result of executing, suppressing or varying a first measure executed in the first vehicle (10, 11, 12) in the past. procedure after claim 1 or 2 , wherein the neural network (40) is trained on a measure of success and/or on a performance of the execution, suppression or variation of the first measure. Method according to one of the preceding claims, wherein the neural network (40) depends on one or more second driving performance indicators of one or more second vehicles (20, 21, 22) when driving on a journey section and/or a section of actions by a user of one or of the plurality of second vehicles, in particular a driver of the one or more second vehicles (20, 21, 22), is trained. Method according to one of the preceding claims, wherein the neural network (40) is trained with data sections which are determined for a predetermined number of users and/or journey sections of the one or more second vehicles (20, 21, 22). Method according to one or more claims, in which varying the first measure in the at least one first vehicle (10, 11, 12) involves determining or adapting a first measure parameter of a first measure and/or selecting an alternative first alternative measure for one, in particular first measure selected by the user of the vehicle (10, 11, 12). procedure after claim 6 , wherein a determination or adjustment of a first measure parameter of a first measure, in particular a determination or adjustment of a coefficient, a function or a gradient for changing the measure parameter, and/or a selection of an alternative first alternative measure to the first measure, in particular a selection of a criterion the selection of the alternative first alternative measure takes place or is controlled depending on the output layer of the neural network (40). Method according to one or more claims, in which the at least one first measure in the first vehicle (10, 11, 12) relates to one or more entries from the following list: - Supporting a driving action by the user of the first vehicle (10, 11, 12); - An execution of influencing the movement of the first vehicle (10, 11, 12); - Generating information for the driver of the first vehicle (10, 11, 12); - Executing an at least partially automated driving maneuver, parking maneuver, parking maneuver and/or maneuvering; - a selection of a parameter or mode of the vehicle light of the first vehicle (10, 11, 12); - A signaling from the first vehicle (10, 11, 12) to the outside. Procedure according to one of claims 3 - 8th , in which the degree of success and/or the performance of a first or second measure are characteristic of: - a feature of lane guidance, in particular a deviation from a lane center, from an ideal line and/or in relation to maintaining a distance to objects at the side, - a feature of a longitudinal guidance, in particular with regard to maintaining a distance, a relative speed and/or a relative acceleration to another road user, - a risk, in particular a risk of collision, the consequence of which has occurred and/or has not occurred, - a course of an acting acceleration on the vehicle (10, 11, 12, 20, 21, 22), in particular on the passenger compartment, - signaling to other road users, in particular signaling by means of acoustic warning signals or light warning signals, - a measure of compliance with and/or deviation from compliance with, in particular in the respective lane section, applicable traffic rules, - a Ma ß the success of one or more measures or driving maneuvers carried out by the vehicle (10, 11, 12, 20, 21, 22), - a measure of the energy consumption and/or a measure of the energy efficiency of the vehicle (10, 11, 12, 20, 21, 22). Procedure according to one of claims 3 - 9 , in which the measure of success and/or the performance, in particular the driving performance, relates to at least partially automated driving and/or to at least partially manual driving. Method according to one of the preceding claims, in which the neural network (40) for executing, suppressing or varying at least one first measure within a specific route, on specific lanes and/or Lane arrangements and / or is trained in certain boundary conditions when driving on a route. Method according to one of the preceding claims, wherein the first measure comprises determining or adapting a degree of automation of a driving function, the degree of automation relating to the journey of the first vehicle (10, 11, 12) in the current road section or a road section ahead and/or to the execution the first measure in the first vehicle (10, 11, 12) in the current road section or in a road section ahead. Procedure according to one of claims 3 - 12 , the measure of success and/or the performance and/or the corresponding data sections being based on a predetermined number of journeys by the one or more second vehicles (20, 21, 22), in particular in the case of a specific application of the one or more second vehicles ( 20, 21, 22), and/or a driver of the second vehicle (20, 21, 22), in particular a specific driver or a driver from a group of drivers or a driver with a measure of success that exceeds a certain level or with a performance that exceeds a certain level, in particular driving performance. Method according to one of the preceding claims, in which the at least one first measure is an at least partially new measure and/or a measure which has never or seldom been carried out in a large number of the second vehicles (20, 21, 22). System for executing, suppressing or varying at least one first measure in a first vehicle (10, 11, 12) by means of a neural network (40), the system comprising: - a first vehicle (10, 11, 12) comprising the neural network (40), and - One or more second vehicles (20, 21, 22), in particular a plurality of second vehicles (20, 21, 22), and wherein the system is formed the neural network (40) depending on data sections associated with the execution of each correspond to a second measure in one or more second vehicles (20, 21, 22). Neural network (40) for use in a first vehicle (10, 11, 12) according to a method according to one of Claims 1 - 14 is selected and/or trained and/or updated, the neural network (40) being designed and/or trained to carry out, suppress or close a first measure when it is executed on a computing unit of the first vehicle (10, 11, 12). vary.

Images