DE102020005615A1 - Method for operating a crosswind assistant in a vehicle - Google Patents

Abstract

The invention relates to a method for operating a crosswind assistant (2) of a vehicle (1). According to the invention, it is provided that in order to predict a crosswind event (S) that will affect the vehicle (1) in the future, at least one model on the basis of recorded and stored data of previous crosswind events (S) that have occurred controlled by a system for driving dynamics control, on the basis of recorded signals from vehicle-side sensors in relation to driving behavior vehicles ahead and / or based on fleet data is trained so that a driving style of the vehicle (1) is adapted by interventions in a steering, a drive train and / or a braking system of the vehicle (1) and a planned vehicle trajectory (T) remains unchanged.

Description

The invention relates to a method for operating a crosswind assistant of a vehicle.

From the DE 10 2017 009 407 A1 a method for operating a vehicle with a vehicle dynamics control is known. A behavior of the vehicle is detected by means of sensors and for this purpose the data measured by the sensors are passed on as input signals to a control unit for driving dynamics control, with output signals for automatic control intervention being output to a controller by the control unit if necessary. The input signals are also recorded by a classifier, and the classifier determines a probability for a control intervention by the controller before the controller carries out an automatic control intervention. If a certain probability is exceeded, a preparatory measure is automatically initiated. The classifier uses a classification model of machine learning based on an artificial neural network, with model training determining and storing at each point in time and therefore for each sequence of sets of input signals whether a control intervention was carried out by the controller and, if so, optionally which control intervention was carried out the controller took place, whereby these recorded control interventions are summarized in classes.

The invention is based on the object of specifying a method for operating a crosswind assistant of a vehicle.

The object is achieved according to the invention by a method which has the features specified in claim 1.

Advantageous embodiments of the invention are the subject of the subclaims.
A method for operating a crosswind assistant of a vehicle provides, according to the invention, that in order to predict a crosswind event that will affect the vehicle in the future, at least one model based on recorded and stored data from previous crosswind events that have occurred using a system for driving dynamics control, based on recorded signals from vehicle-side sensors with regard to driving behavior Vehicles and / or is trained on the basis of fleet data, so that a driving style of the vehicle is adapted by interventions in a steering, a drive train and / or a braking system of the vehicle and a planned vehicle trajectory remains unchanged.

By using the method, it is possible to increase safety and driving comfort in the vehicle, since the vehicle can react preventively to future crosswind events by means of the method and thus the planned vehicle trajectory can be maintained.

Embodiments of the invention are explained in more detail below with reference to drawings.

• 1 schematisch ein Fahrzeug ohne Seitenwindassistent bei einem Seitenwindereignis,
• 2 schematisch ein Fahrzeug mit einem reaktiven Seitenwindassistenten bei einem Seitenwindereignis,
• 3 schematisch ein Fahrzeug mit einem auf einem Modell basierenden Seitenwindassistenten bei einem Seitenwindereignis,
• 4 schematisch einen Verfahrensablauf eines Trainingsmodus des Seitenwindassistenten und
• 5 schematisch einen weiteren Verfahrensablauf eines Vorhersagemodus des Seitenwindassistenten.

Show:

• 1 schematically a vehicle without crosswind assistant in a crosswind event,
• 2 schematically a vehicle with a reactive crosswind assistant in the event of a crosswind event,
• 3 schematically a vehicle with a model-based crosswind assistant in the event of a crosswind event,
• 4th schematically a process sequence of a training mode of the crosswind assistant and
• 5 schematically, a further process sequence of a forecast mode of the crosswind assistant.

Corresponding parts are provided with the same reference symbols in all figures.

The 1 to 3 each show a vehicle 1 in a crosswind event S. , the vehicle 1 in 1 not via a crosswind assistant 2 disposes.

This in 2 shown vehicle 1 has a reactive crosswind assistant 2 on, whereas the vehicle 1 in 3 a predictive crosswind assistant 2 includes.

With an in 1 case shown, one may be on the vehicle 1 effective crosswind event S. can only be corrected by manual intervention by a vehicle user. The vehicle 1 shows one of a driveway F. different vehicle trajectory T , ie a different lane on.

Owns the vehicle 1 via a reactive crosswind assistant 2 becomes when a crosswind event is active S. by means of a control unit of the crosswind assistant, not shown in detail 2 , especially in a steering of the vehicle 1 , intervened to avoid a deviation from the vehicle trajectory T of the vehicle 1 to minimize.

In 3 includes the vehicle 1 the predictive crosswind assistant 2 , by means of which a crosswind event S. is predictable and by means of which the vehicle trajectory T essentially the planned vehicle trajectory T corresponds.

In particular with the reactive crosswind assistant known from the prior art 2 the system reacts to a dangerous situation based on cross winds, since the vehicle control system is only intervened when the vehicle is in motion 1 has already been detected by a gust of wind, in particular if an impulse and the resulting transverse moment are already acting on a vehicle wheel and a deviating travel movement has been caused.

With a conventional crosswind assistant 2 a system for driving dynamics control measures, among other things, control movements of a vehicle user and rotational movements of the vehicle 1 around a vehicle vertical axis and lateral accelerations of the vehicle 1 to avoid sudden movements of the vehicle 1 recognizable due to the effect of crosswinds.

A change in the planned vehicle trajectory can be achieved through the interaction of an adaptive braking system, in particular for braking vehicle wheels facing the wind, a mechanical-electric power steering and, if necessary, electronic damping T , as in 2 is shown minimized and the vehicle 1 be kept as far as possible in its lane.

A method of operating a crosswind assistant 2 , in which a crosswind event S. is predicted, so that a planned vehicle trajectory T remains largely unchanged, is described below.

The method is based on a model that is trained to anticipate a future crosswind event S. to forecast.

The method provides that earlier interventions carried out by means of the system for driving dynamics control to stabilize the vehicle 1 in a crosswind event S. be taken into account when training the model. In particular, machine learning methods are used for this purpose.

In addition, signals from sensors such. B. camera data, a driving speed of the vehicle 1 as well as location and time information, recorded during and before a detected crosswind event time and transferred as property values to a method of machine learning for training the model.

In particular, recorded signals from native, vehicle-bound sensors are evaluated directly, with sensors from assistance systems being used, for example, to detect a vehicle driving ahead and its change in position relative to the center of the lane and / or lane.

Furthermore, historical data from a server-based service are taken into account, these data being e.g. B. to locations with increased susceptibility to crosswinds and / or to local and / or temporal accumulations of crosswind events S. acts.

Furthermore, semantic information is extracted from external visual sensor information with the aid of machine learning methods, in particular with regard to object recognition, in order to learn abstract relationships.

In addition, the method provides that a general and / or region-specific model is created by distributed learning from information from a vehicle fleet via the cloud and the model or models are transferred to the vehicles 1 the vehicle fleet is or will be distributed.
The vehicle is based on the model available on the vehicle side or the models available on the vehicle side 1 regulated in such a way that as soon as the model has a comparatively high probability of a crosswind event S. predicts, automatically vehicle properties, in particular vehicle dynamics, is adapted in such a way that the planned vehicle trajectory T when a crosswind event occurs S. remains unchanged. For this purpose, a current driving speed can be reduced and / or a braking system is prepared according to a crosswind direction.

A training mode of the model or models is shown in 4th described in more detail, with a yes with a y and a no with an n in the case of decisions, with an end of the training mode being indicated by an E.

In a first process step V1 a query is made as to whether a control unit of the crosswind assistant is at the moment 2 a crosswind event S. is detected, wherein the control unit can also be part of the system for driving dynamics control. Will not be a crosswind event S. is detected, the training mode is ended.

Becomes a crosswind event S. are recorded in a second process step V2 Signals from on-board sensors from the current point in time to points in time immediately before the crosswind event S. recorded and saved. In particular, the signals are transmitted by means of a control unit, e.g. B. the system for driving dynamics control, detected and recorded, as signals z. B. a current driving speed of the vehicle 1 , Location and time information, a temperature and / or an air pressure can be recorded at the corresponding times.

In a third process step V3 it is queried whether a global model is being trained using the recorded signals. If so, it will be done in a fourth step V4 a crosswind event S. aggregated in a cloud using triggered signals.

In a fifth process step V5 the global model is trained using machine learning methods, with the global model in a sixth process step V6 on vehicles 1 be transmitted to a vehicle fleet and the training mode is then ended.

Used in the third step V3 it is determined that the global model is not being trained in a seventh method step V7 asks whether a local model is being trained, which in an eighth process step V8 trained or continued training using methods of machine learning and the training mode is ended.

Using machine learning methods, e.g. B. based on neural networks, a model or several models are generated, which or which based, in particular, detected optical signals, a crosswind event S. can or can predict.

This means that the respective trained model is able to indicate a probability that a crosswind event will occur at the current point in time for given recorded signals from vehicle-side sensors at a point in time or recorded signals of a temporal sequence from a moment in time S. entry.

It is also possible, as an alternative or in addition to the signals recorded directly on the vehicle side, to process them before training a model. For example, semantic information can be obtained from external visual sensor information using machine learning methods, e.g. B. with regard to object recognition, extracted in order to learn the abstract relationships. The model or models, e.g. B. on the basis of objects and their relative speed to the vehicle 1 trained.

As described above, general and / or region-specific models can be trained by distributed learning, so-called federated learning, from the vehicle fleet via the cloud of a vehicle manufacturer and this model or these models is or are transmitted back to the vehicle fleet, as crosswind events S. occur relatively rarely. Such a model or several such models can already be used during the development of a vehicle 1 be pre-trained.

3 shows a further method sequence of a prediction mode of the crosswind assistant.

Used in a first step V1 a future crosswind event by means of the control unit S. is predicted by the trained model in a second process step V2 determines whether the model has a likely strength and direction of the crosswind event S. can predict.

If this is possible, a third process step takes place V3 a compensation of the predicted crosswind event S. , ie a cross wind impulse, to maintain a current vehicle trajectory T of the vehicle 1 . The prediction mode is then ended.
It is in the second step V2 not possible, the probable strength and direction of the crosswind event S. Predict are in a fourth step V4 z. B. prepared a braking and chassis system for a control intervention, possibly a current speed of the vehicle 1 is decreased. The forecast mode is then ended.

So it will be a crosswind event S. predicted with a comparatively high probability by means of the trained corresponding model, the vehicle can 1 be regulated accordingly preventively. In particular, vehicle properties and vehicle dynamics can be adapted in such a way that the vehicle trajectory T when a crosswind event occurs S. remains unchanged. The vehicle user of the vehicle 1 can also be informed visually and / or acoustically about a likely future crosswind event, so that manual preventive intervention by the vehicle user can also take place.

Region-specific models also enable crosswind centers to be identified. Such information can e.g. B. can be used to plan routes for vehicles at risk of crosswinds 1 to optimize, for example with a box structure, ie in particular to show danger spots preventively or to suggest structural changes to a route.

As described above, the method provides that the strength of a crosswind event S. as well as a direction, ie a direction of incidence of the crosswind event S. be predicted. It is thus possible according to the third method step V3 of the prediction mode to compensate for the predicted crosswind impulse through targeted interventions in a vehicle control system.

For example, next to a road there is a forest with a firebreak running perpendicular to the road through which a comparatively strong crosswind event occurs when the air pressure is comparatively low S. can occur.

Several of the vehicle 1 Vehicles driving ahead experience control interventions at this point in the firebreak, in the crosswind event S. to compensate. These crosswind events S. are sent to the cloud, i.e. a central computer unit of the vehicle manufacturer, together with signals from vehicle-side sensors recorded before the crosswind event.
In the cloud, one or more models is or are trained using machine learning methods. For example, by means of a model. B. Correlations between occurring crosswind events and perpendicular to the road structures as well as weather conditions and location dependencies can be recognized.

Such a model is capable of a crosswind event S. predict and prevent sudden movements of the vehicle due to cross winds by preventively, e.g. B. by reducing the current speed of the vehicle 1 as well as by preparing an adaptive braking system and an adaptive chassis one through the future crosswind event S. on the vehicle 1 acting lateral force to be compensated.

List of reference symbols

1

vehicle

2

Crosswind Assistant

F.

Track

S.

Crosswind event

T

Vehicle trajectory

V1 to V8

Process step

QUOTES INCLUDED IN THE DESCRIPTION

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

Patent literature cited

• DE 102017009407 A1 [0002]

Claims (3)

Method for operating a crosswind assistant (2) of a vehicle (1), characterized in that to predict a crosswind event (S) that will affect the vehicle (1) in the future, at least one model - based on recorded and stored data of previous crosswind events that have occurred using a system for driving dynamics control (S), - on the basis of detected signals from vehicle-side sensors in relation to the driving behavior of vehicles driving ahead and / or - on the basis of fleet data, training is carried out so that a driving style of the vehicle (1) is carried out by intervening in a steering, a drive train and / or a braking system of the Vehicle (1) is adapted and a planned vehicle trajectory (T) remains unchanged. Procedure according to Claim 1 , characterized in that a strength and a direction of the crosswind event (S) are predicted on the basis of the trained model. Procedure according to Claim 1 or 2 , characterized in that when a crosswind event (S) is predicted to occur, a current driving speed of the vehicle (1) is reduced.

Images