DE102021101718A1 - Method for predicting a failure of automated vehicle guidance, assistance system and motor vehicle - Google Patents

Abstract

The invention relates to a method for predicting a failure of an automated vehicle guidance of a motor vehicle (12), a corresponding assistance system (18) and a motor vehicle (12) equipped therewith. In the method, continuously planned trajectories (32, 34, 36, 38) for the motor vehicle (12) are determined in a conditionally autonomous driving mode by a vehicle guidance system (24). During driving operation, an error prediction model (26) is used to continuously update sequences of planned trajectories (32, 34, 36, 38) determined up to the current time (t, t+1, t+2, t+3) with regard to a variance pattern of the trajectories (32, 34, 36, 38) analyzed. Based on this, it is then automatically predicted whether a failure of the automated vehicle guidance of the motor vehicle (12) by the vehicle guidance system (24) is imminent.

Description

The present invention relates to a method for predicting a failure of an automated vehicle guidance of a motor vehicle. The invention further relates to an assistance system set up for this purpose and a motor vehicle equipped therewith.

In an autonomous operation of a motor vehicle, one or more trajectories must be determined, which describe a movement of the motor vehicle starting from its current position. Such trajectories must take into account the respective local conditions in order to ensure that the motor vehicle moves in accordance with traffic regulations and to avoid obstacles or even accidents. With the systems available today, however, this is not always completely reliable in all situations, although there are already a large number of approaches and suggestions for improvement.

For example, in the US 10,133,275 B1 described a generation of trajectories for an autonomous vehicle using temporal logic and a tree search. Therein, a search algorithm such as MCTS (Monte Carlo Tree Search) can be used to search for possible trajectories, while temporal logic formulas such as LTL (Linear Temporal Logic) can be used to validate or reject the possible trajectories. The trajectories can be optimized based on various performance-optimized costs and constraints. Determining a trajectory may include determining a current state of the vehicle including static and dynamic symbols in an environment. A context of the environment can then be populated with appropriate symbols, features, predictions, and LTL formulas. For example, Rabin automata based on an LTL formula can be used to evaluate the trajectory candidates. Generating nodes of the MCTS and exploring possible actions can be done based on machine learning, for example using a neural network.

Another approach is in the U.S. 8,706,394 B2 describes a control system for a vehicle. This includes an object sensor and a processor that is set up to receive and process object signals from the object sensor. An object in the vicinity of the vehicle is to be identified by means of the processor and its relative position is to be predicted taking into account a movement of the object, a planned movement path and a speed of the vehicle. Based on this, a compensation path should then be calculated in order to avoid a collision with the identified object. The processor should then take over control of the vehicle in order to follow the calculated compensation path if the predicted position of the object lies on the planned path of movement. If the object cannot be avoided by subsequent changes in the speed of the vehicle, a deviation of the vehicle from the avoidance path should be calculated and the vehicle should be guided accordingly.

Next is in the US 9,690,293 B2 describes a system for autonomously navigating an autonomous vehicle along a road segment. With this, images representing an environment of the vehicle are to be analyzed in order to determine a trajectory traveled by the vehicle along the road segment. A current position of the vehicle along a predetermined road model trajectory is also to be determined based on the images. A direction of travel for the autonomous vehicle is to be determined based on the specific traveled trajectory. A steering device for the vehicle is then determined relative to this direction of travel by comparing the traveled trajectory with the predetermined road model trajectory at the current position of the vehicle.

The object of the present invention is to enable safe operation of a vehicle that cannot cope with all driving situations in autonomous operation.

According to the invention, this object is achieved by the subject matter of the independent patent claims. Possible refinements and developments of the present invention are disclosed in the dependent patent claims, in the description and in the drawing.

The method according to the invention is used to predict a failure, that is to say an error or a so-called disengagement, of automated vehicle guidance of a motor vehicle. In particular, this can relate to a lateral and/or longitudinal guidance in a conditionally automated or highly automated driving mode of the motor vehicle, ie for example in a driving mode according to SAE J3016 Level 3 or 4. In a method step of the method according to the invention, continuously planned trajectories or trajectory sections for the motor vehicle are determined or recorded during driving operation of the motor vehicle controlled by a vehicle guidance system for lateral and/or longitudinal guidance. The vehicle guidance system can be, in particular an onboard system of the motor vehicle itself, in order to enable the lowest possible latency. The vehicle guidance system can include one or more devices, such as one or more sensors, a data processing device, a control unit and/or the like.

The planned trajectories can extend from a current position of the motor vehicle in its current direction of travel or describe a planned movement of the motor vehicle during the next few seconds. An updated or new current trajectory can be planned in each case with a movement of the motor vehicle and/or a change in the situation surrounding the motor vehicle. This can be done at a predetermined frequency or clock speed depending on the implementation or available computing capacity. In contrast to a conventional navigation route, which can extend over many kilometers or hours and, for example, only describes roads to be traveled on, but not a more detailed specific movement on the respective road, the trajectories determined here can, for example, be several meters up to some hundred meters or about up to 30 seconds into the future. Compared to a conventional navigation route of a satellite or map data-based navigation system, the planned tractors describe a relatively short-term, detailed movement plan, in particular a planned movement of the motor vehicle within the surroundings of the motor vehicle that can be detected by an ambient sensor system of the motor vehicle.

In a further method step of the method according to the invention, correspondingly updated sequences of such planned trajectories are continuously recorded or provided during the respective driving operation. In this case, these sequences can have a predetermined length, that is to say comprise a predetermined number of planned trajectories or, for example, comprise all planned trajectories that were determined within a sliding time window of predetermined length. The current sequence of trajectories in each case includes certain planned trajectories up to the current point in time during the current driving operation, ie for example the last N trajectories, where N can be a predefined or adjustable parameter. For example, N=10 can be specified for a practical application.

During the respective current driving operation, the respective current one of these sequences of trajectories is continuously analyzed with regard to a variance pattern of the trajectories of the respective current sequence. The sequences or their trajectories are thus evaluated, for example, with regard to changes or differences that have occurred or are developing between the respective trajectories in the course of the respective sequence, i.e. deviations from one another, with regard to a change or reversal of an intended direction of travel over several trajectories determined one after the other, with regard to patterns of analyzed or evaluated the same driving maneuvers and/or the like that were planned several times but not executed.

Based on this, in a further method step of the method according to the invention, it is then also predicted whether or not the automated vehicle guidance of the motor vehicle is about to fail by the vehicle guidance system. This can be done, for example, by a binary output or a respective probability value for the impending failure.

Previous approaches are based, for example, on recorded status data that indicate how the respective vehicle has actually moved in the past few seconds. This is a legitimate approach, since a failure of the automated vehicle guidance is often preceded by certain movement patterns of the motor vehicle. On the one hand, however, this is not always the case and, on the other hand, there is often only a relatively short period of time between the prediction of a failure and its actual occurrence. However, it is known that a human operator cannot instantaneously assume safe control of a motor vehicle that has previously been driving autonomously, but instead needs a certain amount of time to do so. In order to achieve improved safety in such cases, it is therefore desirable to give a respective operator as much time as possible to take control of the motor vehicle. For this purpose, an imminent failure of the automated vehicle guidance must ideally be predicted as early as possible. Precisely this is made possible by the present invention, since the patterns or changes in the trajectories planned for the future, according to a finding on which the present invention is based, can also have typical patterns or a specific variance behavior before an actual failure. Based on the planned trajectories, it is thus possible to predict the failure before the respective motor vehicle has actually traveled a corresponding route, that is to say before it has actually carried out corresponding driving maneuvers according to the planned trajectories. In other words, the present invention allows a predict impending failure of the automated vehicle guidance based on an anticipated future behavior of the motor vehicle before this behavior was actually realized.

In the present invention, in particular, a plurality of trajectories combined in the sequences of trajectories can be considered, which were planned one after the other, ie at different points in time, by one and the same model or vehicle guidance system. This enables improved security and robustness with reduced effort, since, for example, the outputs of two different models do not have to be compared with one another, as in known approaches, so that no implicit assumptions regarding the model or system used in each case have to be made.

The present invention is based on the idea that the planned trajectories reflect a respective understanding of the current traffic situation or traffic scene or its complexity from the perspective of the motor vehicle or its vehicle guidance system. If a specific situation is correctly perceived or understood by the vehicle guidance system and a corresponding solution for automated vehicle guidance is found in this situation, the latest, i.e. current, planned trajectory will significantly, in particular largely, overlap with one or more previous planned trajectories. In this case, there are no or only relatively small deviations between these trajectories planned at successive times. On the other hand, if the situation is complex or demanding for the vehicle guidance system, a previously planned trajectory may no longer be considered safe, so that the next or current trajectory is planned with a significant deviation or change. If the vehicle guidance system plans such changing trajectories, i.e. trajectories that differ significantly from one another, this is an indication that the respective traffic or driving situation is dynamic or generally demanding, in particular too demanding, for the vehicle guidance system, so that a failure of the automated vehicle guidance contributes further approach to the respective situation, a respective obstacle, other road users, a specific route feature and/or the like is to be expected.

In one possible embodiment of the present invention, the respective current sequence is composed of between 10 and 50 trajectories, in particular of at least approximately 30 trajectories. With a calculation/or update frequency of the planned trajectories, for example of the order of 10 Hz, this can allow a period of several seconds, for example the previous 3 s, to be taken into account or covered. This can enable a robust and reliable prediction of the failure of the automated vehicle guidance with a data processing effort that can be managed in a practicable manner.

Likewise, the predefined length of the respectively analyzed sequence of trajectories can be predefined dynamically, for example as a function of a current speed of the motor vehicle. For example, at a higher speed of the motor vehicle, fewer planned trajectories per sequence can be analyzed than at a lower speed. It can thereby be taken into account that the motor vehicle travels different distances at different speeds per unit of time and thus, with an unchanged sensor range, a different amount of time can elapse before a possible failure occurs at a point just entering the sensor range.

In a further possible embodiment of the present invention, when analyzing the trajectories for the prediction, only or predominantly their variance pattern in an overlapping area extending from the current position of the motor vehicle in its direction of travel, which is penetrated by several of the trajectories of the current sequence, taken into account. In other words, a section that is added to the current trajectory compared to the previous trajectory at the maximum distance from the motor vehicle can remain unconsidered or be underweighted. As a result, it can be taken into account that significant changes in direction can occur in the respectively newly added section with a higher probability than in the overlapping area, even in driving situations that are free of errors or failures. The overlapping area thus forms a spatial area extending in front of the motor vehicle, for which several trajectories have already been planned. The parts or sections of these planned trajectories lying within this overlap area can then be used exclusively or overweighted as a basis for predicting the failure. This enables a particularly robust and reliable prediction of the failure based on the knowledge that if the vehicle guidance system correctly understands the respective situation within the overlapping area, it will consistently come to the same result or the same solution to the driving task, i.e. at least essentially the same Trajectory or plan the same trajectory course.

In a further possible embodiment of the present invention, the error prediction model is designed as a machine learning device and is trained to analyze sequences of trajectories and to predict whether the automated vehicle guidance system is about to fail. In other words, the error prediction model can be or include an in particular deep, i.e. multi-layered, artificial neural network that learns or has learned patterns of changes or deviations in the sequences of trajectories as part of its training, which are likely to lead to a failure of the automated vehicle guidance lead, i.e. precede such a failure. The error prediction model can be trained in particular introspectively with regard to the respective vehicle guidance system. In other words, the error prediction model can be trained on the basis of actual sequences of planned trajectories generated by the respective vehicle guidance system, which can be marked, i.e. labeled or annotated, as preceding a failure or as driving through the respective traffic situation without errors, i.e. preceding a successful driving maneuver . This enables a particularly robust and reliable prediction of imminent disengagements of the respective vehicle guidance system. The use of machine learning allows an imminent failure to be predicted reliably in a large number of different situations in a relatively simple manner. This is the case because the error prediction model can independently learn generalizations of concrete training scenarios and thus does not have to be manually programmed explicitly for any potentially possible situation, which seems impractical.

In one possible development of the present invention, the error prediction model is trained both with appropriately annotated sequences of planned trajectories that have resulted in a failure of the vehicle guidance system and with appropriately annotated sequences of trajectories that have resulted in an error-free driving maneuver. This training with negative examples and positive examples enables particularly effective training and learning. This in turn can lead to particularly reliable and robust predictions of the error prediction model. The error prediction model can be trained before the method according to the invention, that is to say separately from it, or it can be carried out as part of the method according to the invention. For example, it may be possible to further train the error prediction model after or during its use in the motor vehicle, for example using sequences of planned trajectories recorded during operation of the respective motor vehicle. This can allow individualization or individual adaptation of the error prediction model to the respective motor vehicle or its driving behavior or typical environment.

In a possible development of the method according to the invention, a recurrent artificial neural network (RNN) is used as the error prediction model or as part thereof. This recurrent neural network can in particular include at least one LSTM module (Long Short-Term Memory). This makes it particularly easy and effective to make the respective prediction based on time patterns or dynamic developments in the sequences of trajectories. This may allow for a more reliable prediction than would be possible based on a single planned trajectory, for example.

In a further possible embodiment of the present invention, in the event that a failure of the vehicle guidance system is or was predicted, a request to take over manual control of the motor vehicle is automatically issued. In other words, in such a case, a human operator can take control of the motor vehicle, in particular before the failure of the vehicle guidance system has actually occurred. As a result, safe, continuous operation of the motor vehicle can be made possible. The request for manual control takeover can be issued to a vehicle occupant or driver of the motor vehicle and/or to an operator external to the vehicle, ie a teleoperator. In this way, the vehicle guidance system can already be used in practice despite not being absolutely reliable in all possible traffic situations, whereby, for example, a gain in comfort or even a gain in safety can be achieved compared to permanently completely manual guidance of the motor vehicle.

In a further possible embodiment of the present invention, if a failure of the vehicle guidance system is or was predicted, a control signal for initiating a safety maneuver that brings the motor vehicle into a safe state, in particular to a standstill, is automatically output . Since this is based on the forecast, i.e. based on the planned, trajectories that have not yet been completely traversed, the safety maneuver can also be initiated if the vehicle guidance system at the time the failure is predicted still delivers or finds trajectories that it believes are permissible, i.e. solutions for trajectory planning that correspond to given requirements. This saves time for the motor vehicle to react, so that the safety maneuver can be less abrupt or violent, for example. This can contribute to further improved safety, since not only is the motor vehicle itself put into a safe state as early as possible, but other road users can also react more easily and safely to the safety maneuver of the motor vehicle. The safety maneuver can include, for example, a reduction in the speed of the motor vehicle, a braking or stopping maneuver, evading onto a hard shoulder and/or the like. Such a safety maneuver can take into account that, for example, a manual takeover of control of the motor vehicle by an operator can take a significant amount of time, for example up to 30 s according to studies, but the impending failure of the automated vehicle guidance according to the forecast will already be shorter distance in time can occur. It can thus be ensured that the motor vehicle does not get into a situation in which the vehicle guidance system fails and at the same time no operator has yet taken over manual control of the motor vehicle.

Another aspect of the present invention is an assistance system for a motor vehicle. The assistance system according to the invention has an input interface for detecting trajectories of the motor vehicle planned by an automated vehicle guidance system, a computer-readable data memory, a processor device connected thereto and an output interface. An error prediction model for predicting, based on sequences of recorded planned trajectories, whether a failure of the vehicle guidance system is imminent is stored in the data memory. The processor device can, for example, be or comprise a microchip, a microprocessor, a microcontroller or the like. For example, the respective prediction or a control or notification signal generated based thereon can be output via the output interface. The assistance system according to the invention is set up to carry out, in particular automatically, at least one variant of the method according to the invention. Accordingly, the assistance system according to the invention can in particular be the assistance system mentioned in connection with the method according to the invention. The same also applies to the other named components of the assistance system according to the invention, such as the named error prediction model and the named vehicle guidance system. For example, a corresponding operating or computer program can be stored in the data memory that represents, i.e. encodes or implements, the method steps or sequences of the method according to the invention and can be executed by the processor device in order to execute the corresponding method or cause it to be executed.

Another aspect of the present invention is a motor vehicle that has a sensor system for recording environmental data that depicts or characterizes a particular area surrounding the motor vehicle, a vehicle guidance system connected thereto for automated lateral and/or longitudinal guidance of the motor vehicle, and the assistance system according to the invention. Accordingly, the motor vehicle according to the invention is set up for executing or using the method according to the invention. The motor vehicle according to the invention can therefore in particular be the motor vehicle mentioned in connection with the method according to the invention and/or in connection with the assistance system according to the invention. Accordingly, the motor vehicle according to the invention can have some or all of the properties and/or features mentioned in these contexts.

Further features of the invention can result from the claims, the figures and the description of the figures. The features and feature combinations mentioned above in the description and the features and feature combinations shown below in the description of the figures and/or in the figures alone can be used not only in the combination specified in each case, but also in other combinations or on their own, without going beyond the scope of the invention to leave.

The drawing shows in 1 a schematic representation of a traffic situation in which a motor vehicle is moving autonomously towards an obstacle.

In 1 a traffic situation is shown in a schematic plan view in which a motor vehicle 12 is driving towards a hazard spot 14 on a road 10 in autonomous or conditionally automated driving mode. The danger point 14 can be caused, for example, by an unclear traffic situation, an accident, a construction site, a other road users who do not behave in accordance with the rules and/or the like.

The motor vehicle 12 includes an environment sensor 16, which detects an area surrounding the motor vehicle 12, including the danger point 14 in the present case. The motor vehicle 12 also has an assistance system 18 which receives and processes environmental data recorded by means of the environmental sensors 16 . Depending on the configuration of the environmental sensor system 16, this environmental data can be or include, for example, camera images, radar echoes, lidar echoes and/or the like. The assistance system 18 can be set up, for example, to control a vehicle component 20 or the motor vehicle 12 as a whole autonomously, at least at times, based on the environmental data or a processing result generated therefrom.

For this purpose, the assistance system 18 has a data memory 22 in which a predefined vehicle guidance model 24 and a predefined error prediction model 26 are stored. The error prediction model 26 includes an artificial neural network 28, indicated here schematically, which is trained to predict whether a failure of the vehicle guidance model 24, ie the automated vehicle guidance or vehicle control of the motor vehicle 12, is imminent. Assistance system 18 also includes a processor 30 for executing the content stored in data memory 22 .

The assistance system 18 or the vehicle guidance model 24 is set up to continuously plan trajectories for the motor vehicle 12 during driving operation. These trajectories each extend from a current position x of motor vehicle 12 in its direction of travel. For example, a first trajectory 32 is planned here at a current point in time t, starting from the current position x. This first trajectory 32 extends in a straight line from the current position x to a probable or planned future position y of the motor vehicle 12 and then turns to the right in the direction of travel in order to avoid the hazardous area 14 . If motor vehicle 12 is now moving along first trajectory 32, a second trajectory 34 is planned at a later point in time t+1. This second trajectory also leads straight ahead to the future position y, but then turns left in the direction of travel in order to avoid the danger point 14 . Accordingly, further trajectories are continuously planned while the motor vehicle 12 is driving, for example a third trajectory 36 at the even later point in time t+2 and a fourth trajectory 38 at the even later point in time t+3. The third trajectory 36 also turns to the left here like the second trajectory 34 planned beforehand, but has a significantly greater curvature, while the fourth trajectory 38 planned afterward to avoid the danger point 14 again turns to the right like the first trajectory 32 .

The trajectories 32 to 36 indicated here as examples are only shown offset transversely to the direction of travel of motor vehicle 12 for the sake of clarity, i.e. they can be at least essentially identical in their sections lying between the current position x and the future position y, i.e. in a corresponding overlapping area be, i.e. actually lie on top of each other.

Corresponding trajectories can, for example, be newly determined or planned or updated several times per second. At a typical driving speed of the motor vehicle 12, several planned trajectories can therefore differ drastically from one another, at least in some areas, although the motor vehicle 12 moves at least apparently stringently or constantly, i.e. consistently with a single, permanently constant trajectory, during a period in which these trajectories were planned. Despite the trajectories 32 to 36 that differ significantly from each other here according to the future position y, the motor vehicle 12 can drive straight ahead up to the future position y, so that there is no indication of an impending failure of the automated vehicle guidance outwards or based on the distance traveled up to that point , ie the vehicle guidance model 24, is recognizable.

Based on the differences and changing planned directions of travel according to the trajectories 32 to 36 after the future position y, however, it can be seen that the vehicle guidance model 24 is obviously uncertain as to how the motor vehicle 12 is or can be guided safely or optimally in view of the danger zone 14 ahead. In the present case, this is exploited in that a current sequence of planned trajectories, in this case for example the trajectories 32 to 36, are supplied to the error prediction model 26 as input. Based on the variation of the trajectories 32 to 36, this predicts whether or not a failure of the vehicle guidance model 24 is imminent or the probability of such a failure.

In the present case, based on the differences or the reversal of the directions of the planned trajectories 32 to 36, it can be deduced that an upcoming traffic situation around the area of the danger point 14 is complex and demanding and therefore a failure of the vehicle guidance model 24 is to be expected with a certain probability. The planned trajectories 32 to 36 each extend several seconds into the future, so that they therefore represent a possibility of anticipating the future behavior of the motor vehicle 12 before the planned trajectories 32 to 36 have actually been driven on.

It should be pointed out that the present case is not about predicting how the actual movement trajectory of the motor vehicle 12 or the planned trajectories 32 to 36 will develop in the future. Rather, the error prediction model 26 analyzes here how the trajectories planned in the immediate past up to the current point in time have developed. A current sequence of trajectories planned up to that point is thus provided as input to the error prediction model 26 in each case. This current sequence can include, for example, the first trajectory 32, the second trajectory 34 and the third trajectory 36 at time t+2. At time t+3, however, the then current sequence can include, for example, the second trajectory 34, the third trajectory 36 and the fourth trajectory 38. However, this is to be understood purely schematically and as an example here, since in a practical application other numbers of trajectories, for example the 10 most recently planned trajectories, can also be used or taken into account for the current sequence of trajectories.

The correspondingly trained error prediction model 26 can assign each input, ie each sequence of trajectories, a probability that this sequence will lead to a failure, ie a so-called disengagement, of the vehicle guidance model 24 . This probability can then be compared to a predetermined probability threshold. If the predicted probability reaches or exceeds this probability threshold value, then the assistance system 18 can react to this in a predetermined manner. For example, a request signal 40 can be output, which requests manual monitoring or control of motor vehicle 12 by an operator, for example a driver 42 of motor vehicle 12 or a vehicle-external teleoperator 44 . The teleoperator 44 can then, for example, transmit a control signal 46 to the motor vehicle 12 in order to take manual remote control of the latter around the danger point 14 . The teleoperator 44 or the driver 42 can then transfer control of the motor vehicle 12 back to the assistance system 18 or to the vehicle guidance model 24 .

Due to the fact that the request signal 40 is based on a prediction and not on a failure of the vehicle guidance model 24 that has actually already occurred, the respective operator can have sufficient time to take over the manual control of the motor vehicle 12 . Likewise, the assistance system 18 can automatically initiate a safety maneuver of the motor vehicle 12 in the event of a predicted failure of the vehicle guidance model 24 in order to enable the respective operator to take over control manually even more safely. The failure can typically be predicted several seconds in advance, since the respective current sequence of the planned trajectories typically shows variance patterns indicating this, ie changes in the planned trajectories, several seconds before an actual failure of the vehicle guidance model 24 .

Overall, the examples described show how a disengagement prediction can be realized based on an analysis of planned trajectories in order to enable a particularly safe operation of a vehicle in a conditionally autonomous driving mode.

Reference List

10

Street

12

motor vehicle

14

danger point

16

environmental sensors

18

assistance system

20

vehicle component

22

data storage

24

vehicle guidance model

26

failure prediction model

28

neural network

30

processor

32

first trajectory

34

second trajectory

36

third trajectory

38

fourth trajectory

40

request signal

42

driver

44

teleoperator

46

control signal

x

actual position

y

future position

t, t+1, t+2, t+3

times

QUOTES INCLUDED IN DESCRIPTION

This list of documents cited 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

• US10133275B1 [0003]
• US 8706394 B2 [0004]
• US 9690293 B2 [0005]

Claims (10)

Method for predicting a failure of an automated vehicle guidance of a motor vehicle (12), in which - continuously planned trajectories (32, 34, 36, 38) for the motor vehicle (12) are determined during driving operation of the motor vehicle (12) controlled by a vehicle guidance system (24) for lateral and/or longitudinal guidance, - during the driving operation by means of a predetermined error prediction model (26) continuously a correspondingly updated current sequence of predetermined length of planned trajectories (32, 34, 36, 38) determined up to the current time during the current driving operation with regard to a variance pattern of the trajectories (32, 34, 36, 38) of the respective current sequence is analyzed, and - Based on this, it is automatically predicted whether a failure of the automated vehicle guidance of the motor vehicle (12) by the vehicle guidance system (24) is imminent. procedure after claim 1 , characterized in that the respective current sequence is composed of between 10 and 50 trajectories (32, 34, 36, 38), in particular of 30 trajectories (32, 34, 36, 38). Method according to one of the preceding claims, characterized in that when the trajectories (32, 34, 36, 38) are analyzed for the prediction, only or predominantly their variance pattern in a range from the current position (x) of the motor vehicle (12) in whose overlapping area extending in the direction of travel, through which several of the trajectories (32, 34, 36, 38) of the current sequence pass, is taken into account. Method according to one of the preceding claims, characterized in that the error prediction model (26) is designed as a machine learning device (28) and is based on analyzing sequences of trajectories (32, 34, 36, 38) and predicting whether a failure of the automated vehicle guidance is imminent. procedure after claim 4 , characterized in that the error prediction model (26) both with appropriately annotated sequences of planned trajectories (32, 34, 36, 38), which have resulted in a failure of the vehicle guidance system (24), and with appropriately annotated sequences of trajectories (32 , 34, 36, 38) that resulted in a flawless driving manoeuvre. procedure after claim 4 or 5 , characterized in that the error prediction model (26) comprises a recurrent neural network (28), in particular with at least one LSTM module. Method according to one of the preceding claims, characterized in that if a failure of the vehicle guidance system (24) is predicted, a request (40) to take over manual control of the motor vehicle (12) is automatically issued. Method according to one of the preceding claims, characterized in that in the event that a failure of the vehicle guidance system (24) is predicted, a control signal for initiating a safety maneuver by which the motor vehicle (12) is brought into a safe state, in particular to a standstill, is automatically is brought, is spent. Assistance system (18) for a motor vehicle (12), having an input interface for detecting trajectories (32, 34, 36, 38) of the motor vehicle (12) planned by a vehicle guidance system (24), a data memory (22) in which an error prediction model (26) for predicting whether a failure of the vehicle guidance system (24) is imminent, based on sequences of recorded planned trajectories (32, 34, 36, 38), is stored, a processor device (30) connected thereto and an output interface, the assistance system (18) is set up to carry out a method according to any one of the preceding claims. Motor vehicle (12), having a sensor system (16) for recording environmental data, a vehicle guidance system (24) connected thereto for automated lateral and/or longitudinal guidance of the motor vehicle (12) and an assistance system (18). claim 9 .

Images