DE102016207276A1 - Method for releasing a driving function in a vehicle - Google Patents

Abstract

A method for releasing a driving function of a vehicle for a preceding roadway section is described. The method comprises determining a first driving scenario data record that describes a first driving situation that is to be handled autonomously by the driving function in the preceding roadway section. In addition, the method includes determining, based on a classifier and based on the first driving scenario data set, whether the driving function will release for the preceding roadway section.

Description

The invention relates to a method and a corresponding control unit for releasing one or more driving functions in a vehicle.

A vehicle may include one or more different semi-automated, highly automated or fully automated driving functions that assist a driver of the vehicle in driving the vehicle. Exemplary driving functions are: a longitudinal guidance assistant, a lateral guidance assistant, a lane departure warning, a congestion assistant, autonomous driving, etc.

A driving function typically uses sensor data from one or more environmental sensors of the vehicle. The sensor data is processed by the logic of a driving function and, depending on the sensor data, one or more actuators of the vehicle are controlled to at least partially autonomously guide the vehicle.

The fact whether an at least partially automated or autonomous driving function can be used in a particular driving situation depends on the concrete implementation of the driving function in a vehicle and on current environmental conditions (for example, certain weather conditions, road conditions, traffic conditions, etc.).

The present document deals with the technical problem to provide a method and a corresponding control unit, with which it can reliably be determined in good time whether or not a driving function in a vehicle can be used in a certain roadway section.

The object is solved by the independent claims. Advantageous embodiments are described i.a. in the dependent claims.

According to one aspect, a method for determining a classifier for evaluating a driving function of a vehicle is described. The driving function can be designed to guide the vehicle autonomously along and / or transversely. Exemplary driving functions are: (possibly partially, highly, or fully) automated driving, a traffic jam assistant, a longitudinal guidance assistant, a lane departure warning system, etc.

The method may include providing training data with a plurality of training records. In this case, a training record comprises a driving scenario data record which describes a driving situation that is to be handled autonomously by the driving function. In this case, a driving scenario data record may comprise feature values for a plurality of features in order to describe a driving situation of a vehicle. The driving scenario data record can thus represent an expression of a feature vector, wherein the feature vector comprises the plurality of features as dimensions.

The plurality of features may include: one or more features related to a lane on which the vehicle is traveling (e.g., lane travel, number of lanes, a tunnel, a bridge, etc.); one or more features relating to an object in an environment of the vehicle (e.g., a position and / or a trajectory of an object, particularly another road user); one or more features related to a driving maneuver planned by the driving function of the vehicle (e.g., a planned trajectory of the vehicle); and / or one or more features describing weather conditions around the vehicle (e.g., precipitation).

Herein, the driving situation described by a driving scenario record may include a certain planning horizon (for example, one minute or less) from a current time. In other words, a driving scenario record may describe a driving situation that has to cope with a driving function in a certain preceding time period (i.e., the planning horizon).

A training record further includes a reliability record indicating how the driving situation (described by the corresponding driving scenario record) was handled by the driving function. In particular, the reliability record may include (or correspond to such a value of) a reliability value indicating whether or not a driving situation described by the corresponding driving scenario record has been handled by the driving function. In this case, the reliability value may assume values from a predetermined value interval (for example, between 0 and 1).

The method further comprises determining a classifier by teaching a model by means of the training data, such that the classifier is designed to provide reliability information for a specific driving situation. In particular, the classifier can be designed to indicate reliability information (for example a reliability data record) for a specific driving scenario data record (by means of which a specific driving situation is described). In this case, the model may in particular comprise a neural network. As discussed above, a driving scenario record may include or include a feature vector correspond. A feature vector has a certain dimension and, in connection with the value ranges of the individual features, defines a possible driving scenario space (with the specific dimension). The classifier (in particular, the model used) may be configured such that a reliability record can be assigned to each possible point in the driving scenario space.

The provision of a classifier trained with training data sets enables the efficient and reliable activation or possibly degradation of driving functions during the driving operation of a vehicle. The training data may include real training data sets that have been acquired from one or more sensors of a vehicle. Alternatively or additionally, the training data may include synthetic training data sets determined by simulating the behavior of the driving function. By providing synthetic training datasets, the number of training datasets can be efficiently increased and the coverage of the possible feature space expanded. Thus, the quality of the classifier can be further increased.

The training data may depend on a type of driving function and / or on a type of vehicle. Thus, driving function type-specific and / or vehicle-type-specific classifiers can be provided, whereby the reliability of the function release can be further increased. The training data can be determined as part of the development of a driving function and / or a vehicle.

According to a further aspect, a method for enabling a driving function of a vehicle for a preceding roadway section is described. The method may be performed by a control unit of a vehicle. The method comprises determining a first driving scenario data record that describes a first driving situation that is to be handled autonomously by the driving function in the preceding roadway section. The driving scenario record may include characteristic values for one or more of the above. Features include. In this case, the first driving scenario data record can be determined on the basis of environmental data from one or more environment sensors of the vehicle. Alternatively or additionally, the first driving scenario data record can be determined on the basis of external data from one or more vehicle-external sensors, wherein the external data may include information relating to the preceding roadway section that is not or insufficiently provided by the on or several environment sensors of the vehicle can be detected. By taking into account external data, the forecast horizon or planning horizon for the release of the driving function can be extended.

The method further comprises determining, based on a classifier and based on the first driving scenario data set, whether the driving function will release (or possibly be degraded) for the preceding roadway section. The classifier may have been determined by the method described in this document. In particular, the classifier may be configured to assign different reliability data sets to different driving scenario data records, wherein a reliability data record indicates a reliability with which a driving situation described by a corresponding driving scenario data record is managed by the driving function. In other words, the classifier may be configured to indicate reliability information (e.g., a reliability record) for a particular driving situation (described, for example, by a particular driving scenario record). In this case, the reliability information can indicate a reliability with which the specific driving situation is managed by the driving function.

It can thus be determined by means of the classifier, a first reliability record (or first reliability information) for the first driving scenario record or for the first driving situation. It can then be decided based on the first reliability record, whether the driving function is enabled or degraded. The consideration of a classifier thus enables a flexible and reliable release of driving functions during the driving operation of a vehicle.

The method may include determining an actual reliability record indicative of reliability actually coping with the first driving situation of the driving function described by the first driving scenario data set. In particular, it can be determined after release of the driving function with which reliability the driving function has actually mastered the first driving situation. The classifier may then be adjusted depending on the first driving scenario record and the actual reliability record. Thus, the quality of the classifier and thus the robustness of the driving function release can be continuously improved.

In another aspect, a control unit for a vehicle configured to execute the method of releasing a driving function described in this document is described.

According to a further aspect, a vehicle (in particular a road motor vehicle, for example a passenger car, a lorry or a motorbike) is described which comprises the control unit described in this document.

In another aspect, a software (SW) program is described. The SW program can be set up to run on a processor, thereby performing the procedures described in this document.

In another aspect, a storage medium is described. The storage medium may include a SW program that is set up to run on a processor and thereby perform the methods described in this document.

It should be understood that the methods, devices and systems described herein may be used alone as well as in combination with other methods, devices and systems described in this document. Furthermore, any aspects of the methods, devices, and systems described herein may be combined in a variety of ways. In particular, the features of the claims can be combined in a variety of ways.

Furthermore, the invention will be described in more detail with reference to exemplary embodiments. Show

1 an exemplary system for determining reliability data relating to a driving function;

2a a flowchart of an exemplary method for determining a reliability classifier; and

2 B a flowchart of an exemplary method for determining reliability data with respect to a driving function.

As stated above, the present document deals with the robust and possibly predictive determination of reliability data with respect to an at least partially autonomous driving function of a vehicle. In this context shows 1 exemplary components of a system 100 , which can be used to determine reliability data for a driving function. The system 100 includes a vehicle 110 with one or more at least partially autonomously driving driving functions. The vehicle 110 includes one or more environmental sensors 114 (For example, one or more cameras, radar sensors, ultrasonic sensors and / or LIDAR sensors), which are arranged, sensor data relating to an environment of the vehicle 110 to determine (also referred to as environment data). The vehicle 100 further includes a position sensor 115 that is set up position data (eg, GPS coordinates) with respect to a position of the vehicle 110 capture. In addition, the vehicle includes 100 a storage unit 112 on, for example, digital map information relating to one of the vehicle 110 traveled roadway are stored.

• • die Position und die (prädizierte) Trajektorie von ein oder mehreren Objekten (z.B. anderen Verkehrsteilnehmern) in dem Umfeld des Fahrzeugs 110;
• • die Position und die (geplante) Trajektorie des Fahrzeugs 110; dabei kann die geplante Trajektorie des Fahrzeugs 110 von einer aktivierten Fahrfunktion bestimmt werden; und/oder
• • die Witterungsbedingungen.

A control unit 111 of the vehicle 110 Based on the environmental data, based on the position data and on the basis of the digital map information, driving scenario data (in particular a driving scenario data record) can be put together that show a current driving situation of the vehicle 110 describe. The driving scenario data can include, for example, static and dynamic data for a specific planning horizon (of a specific number of seconds or meters). The static data (eg in the form of different characteristics) can display fixed characteristics (such as the number of lanes, the course, etc.) of a lane traveled in the planning horizon. The dynamic data (eg in the form of different characteristics) can be displayed, for example

• The position and the (predicated) trajectory of one or more objects (eg other road users) in the environment of the vehicle 110 ;
• • the position and the (planned) trajectory of the vehicle 110 ; doing this may be the planned trajectory of the vehicle 110 be determined by an activated driving function; and or
• • the weather conditions.

The driving scenario data may be presented in structured form as an expression of a feature vector having a plurality of dimensions (and a corresponding plurality of features). In other words, the driving scenario data may include feature values for a plurality of different features, a current driving situation of the vehicle 110 to describe.

To extend the planning horizon can be used for the driving scenario data on external data, the external data information with respect to an underlying environment of the vehicle 110 show that is not (yet) or not sufficient with the one or more environmental sensors 114 of the vehicle 110 can be detected. The preceding environment of the vehicle 110 can do this on a planned driving route of the vehicle 110 lie. The external data can be from a central unit 130 (eg from a central data server) (typically via a wireless communication link) and one or more external sensors 130 (eg from the environmental sensors of other road users).

On the storage unit 112 of the vehicle 110 For example, a classifier or observation unit may be stored that is configured to provide reliability data (in particular a reliability data set or a reliability value) based on the driving scenario data, indicating the reliability with which a driving function is performed within the framework of the Driving scenario data described driving situation can be operated. In particular, the reliability data can indicate with which reliability the trajectory of the vehicle planned in the driving scenario data 110 can be implemented by the driving function. It can then be decided depending on the reliability data, whether the driving function is carried out further, or whether possibly a deactivation or degradation of the driving function takes place. In particular, it can be decided on the basis of the reliability data whether the planned trajectory is implemented by the driving function or not.

At a certain point in time, a plurality of driving scenario data records can be determined, wherein each driving scenario data record describes a possible driving situation. The different driving scenario data sets can be, for example, in relation to the planned trajectory of the vehicle 110 and / or with respect to an active driving function and / or with respect to a degree of autonomy of the vehicle 110 differ. By means of the classifier or observer can then be determined for each of the driving scenario data sets a reliability record. Based on the plurality of reliability data records, it is then possible to decide which trajectory and / or which driving function and / or which degree of autonomy by the control unit 111 of the vehicle 110 is implemented.

A classifier or observer can be trained as part of the development of a driving function. In doing so methods of machine learning can be used. In particular, a model (eg, a neural network) may be learned based on training data to provide a classifier configured to assign a reliability record to a driving scenario record (with a particular expression of characteristics) indicates how well a driving scenario described by the driving scenario record by the vehicle 110 at least partially autonomous.

The training data includes a variety of training records (typically 100,000, 1 million or more training records). Here, a training record comprises a (real measured or a synthetically generated) driving scenario record and an associated (real measured or simulated) reliability record (e.g., a value indicating whether the driving scenario could be implemented properly or not). The training datasets are typically recorded as part of the development of a driving function for test purposes.

The training data may then be used to train the model (e.g., a neural network) for the classifier.

By using a classifier or observer, the function release of driving functions in a vehicle can 110 (especially in connection with (possibly highly- or fully) automated driving (eg HAF)) dynamically released. For this purpose, cooperation between a local database on the vehicle 110 and a backend 120 respectively.

To teach the classifier or the observer, real and typically virtual driving scenarios can be taken into account. By means of the driving scenarios, the stochastic space of possible driving situations (also referred to as a possible driving scenario space) should be covered as well as possible with a specific planning horizon. In particular, a particular focus may be placed on driving scenarios that could lead to difficulties (e.g., relatively low reliability) for a driving function.

Such a trained classifier or observer can then in the vehicle 110 used to determine a certain reliability value (on a certain scale or in a specific value range) for a current driving scenario. This allows a dynamic function release. In the case of the HAF as a driving function, it is typically to be noted that a predictive function release takes place (eg with a horizon of at least 10 seconds), in order to initiate a controlled transfer by the driver if necessary. For this purpose external data can be taken into account (as explained above).

The classifier or observer can be in a storage unit 112 of the vehicle 110 get saved. The classifier or observer can display the performance and / or the reliability of a driving function for different driving scenarios.

During operation of the vehicle 110 can the control unit 111 the current state of the vehicle 110 and the external effects on the vehicle 110 determine. For example, the vehicle 110 Automated driving at one time, it can rain and it can be seen a curve in the course of the route. This information may be described by a driving scenario record. By means of the learned classifier or observer, a specific risk or reliability value can be determined for such a driving situation or for such a driving scenario. To increase the risk or reliability value, the driving function can be degraded (eg to "keep track"). As a result, the reliability of the driving function can be increased.

The planning horizon for the release of driving function can vary. In particular, by considering external data, the planning horizon can possibly be substantially extended. By extending the planning horizon, the safety of (partially) autonomous driving functions can be increased.

The use of a classifier or observer allows driving functions for certain driving scenarios during the operation of a vehicle 110 to evaluate and release or degrade. In this case, the classifier or observer can be designed to be learning, so that current (ie actual) driving scenarios and (actual) conversions of driving functions can be used as training data for further training of the classifier or observer. Thus, the robustness of the driving function release can be further increased.

2a shows a flowchart of an exemplary method 210 for determining a classifier for evaluating a driving function of a vehicle 110 (In particular, to determine the reliability of a driving function for a particular driving scenario or for a particular driving situation). The procedure 210 can be performed by a server or computer (eg from a central processing unit 120 ).

The procedure 210 includes providing 211 of training data with a variety of training records. In this case, a training record comprises a driving scenario data record which describes a driving situation that is to be handled autonomously by the driving function. Furthermore, a training record includes a reliability record that indicates how the driving situation was handled by the driving function. In addition, the process includes 210 the determining 212 of a classifier by teaching a model by means of the training data. The classifier can be determined in such a way that the classifier is designed to assign different reliability data sets to different driving scenario data sets. For this purpose, a suitable model (eg a neural network) can be selected. Furthermore, a relatively high number of different training records (eg, 1 million or more) may be used.

2 B shows a flowchart of an exemplary method 220 for releasing a driving function of a vehicle 110 for a preceding roadway section. The procedure 220 can by a processor (eg by a control unit 111 ) of a vehicle 110 be executed.

The procedure 220 includes determining 221 a first driving scenario data set describing a first driving situation, which is to be handled autonomously by the driving function in the preceding roadway section. The procedure 220 further comprises determining 222 on the basis of a classifier and on the basis of the first driving scenario data set, whether the driving function for the preceding roadway section is enabled (or possibly degraded). The classifier can according to procedures 210 be determined. Furthermore, the classifier may be configured to assign different reliability data sets to different driving scenario data records, wherein a reliability data record indicates a reliability or a probability with which a driving situation described by a corresponding driving scenario data record is handled by the driving function.

The procedures described in this document 210 . 220 allow a dynamic adjustment of the power spectrum of driving function during the operation of a vehicle 110 , Furthermore, by the described method 210 . 220 allows a continuous assessment of how driving functions work. In this case, the data processing for a classifier or observer on the control unit 111 of a vehicle 110 and / or in a central processing unit 120 respectively.

The present invention is not limited to the embodiments shown. In particular, it should be noted that the description and figures are intended to illustrate only the principle of the proposed methods, apparatus and systems.

Claims (10)

Procedure ( 210 ) for determining a classifier for evaluating a driving function of a vehicle ( 110 ); the method ( 210 ), - Provide ( 211 ) training data with a plurality of training records; wherein a training record comprises, - a driving scenario data record which describes a driving situation which is to be handled autonomously by the driving function; and a reliability record indicating how the driving situation was handled by the driving function; and - determining ( 212 ) of a classifier by teaching a model by means of the training data, such that the classifier is designed to provide reliability information for a specific driving situation. Procedure ( 210 ) according to claim 1, wherein - a driving scenario record comprises feature values for a plurality of features in order to determine a driving situation of a vehicle ( 110 ) to describe; and - the plurality of features comprises, - one or more features relating to a roadway on which the vehicle ( 110 ) moves; One or more features relating to an object in an environment of the vehicle ( 110 ); One or more characteristics related to the driving function of the vehicle ( 110 ) planned driving maneuver; and / or - one or more characteristics, the weather conditions surrounding the vehicle ( 110 ). Procedure ( 210 ) according to any one of the preceding claims, wherein the training data comprise, - real training data sets obtained from one or more sensors ( 114 ) of a vehicle ( 110 ) were recorded; and - synthetic training data sets obtained by simulating behavior of the driving function. Procedure ( 210 ) according to one of the preceding claims, wherein the model comprises a neural network. Procedure ( 210 ) according to one of the preceding claims, wherein the driving function is an autonomous longitudinal and / or transverse guidance of the vehicle ( 110 ). Procedure ( 210 ) according to one of the preceding claims, wherein the reliability record comprises a reliability value which indicates whether or not a driving situation described by the corresponding driving scenario data record has been mastered by the driving function. Procedure ( 220 ) for releasing a driving function of a vehicle ( 110 ) for a preceding lane section, the method ( 220 ), - determining ( 221 ) a first driving scenario data set describing a first driving situation to be autonomously managed by the driving function in the preceding roadway section; and - determining ( 222 ), based on a classifier and on the basis of the first driving scenario record, whether the driving function will release for the preceding roadway section; wherein the classifier is configured to provide reliability information for a particular driving situation; and wherein the reliability information indicates a reliability with which the particular driving situation is handled by the driving function. Procedure ( 220 ) according to claim 7, wherein the first driving scenario data record based on environmental data from one or more environment sensors ( 114 ) of the vehicle ( 110 ) is determined. Procedure ( 220 ) according to one of claims 7 to 8, wherein - the first driving scenario record based on external data from one or more vehicle-external sensors ( 130 ) is determined; and - the external data comprises information relating to the preceding lane section which is not or not adequately provided by one or more environmental sensors ( 114 ) of the vehicle ( 110 ) can be detected. Procedure ( 220 ) according to any one of claims 7 to 9, wherein the process ( 220 ), - determining an actual reliability record indicating a reliability with which the first driving situation described by the first driving scenario data set was actually handled by the driving function; and - adjusting the classifier in dependence on the first driving scenario record and the actual reliability record.

Images