Procedure for operating a motor vehicle

Abstract

The invention relates to a method (24) for operating a motor vehicle (2), in particular one that can be moved at least partially automatically, in which method an error (32) of a further road user (30) is detected. The error (32) is classified, and on the basis of the classification a first forecast (44) for further behavior of the further road user (30) is created, for which purpose machine learning is used. On the basis of the classification a second forecast (50) for the further behavior of the further road user (30) is created, for which purpose a predefined set of rules is used. The intersection (56) of the two forecasts (44, 50) is determined, and a function (60) is determined and executed depending on the intersection (56). The invention further relates to an automated motor vehicle (2), a control unit (14), and a computer program product (20).

Description

[0001] The invention relates to a method for operating a motor vehicle. The motor vehicle is preferably movable at least partially automatically. Furthermore, the invention relates to an at least partially automated motor vehicle, a control unit, and a computer program product.

[0002] Motor vehicles are increasingly being designed for automated driving. Certain driving functions are controlled automatically using one or more assistance systems, so that the driver only needs to intervene partially. Fully autonomous vehicles are also known, where a driver is not necessarily required. In these autonomous vehicles, which in particular meet Level 5 requirements, the desired destinations are entered. Using the vehicle's assistance systems, a suitable route to the destination is selected and dynamically adjusted, for example, if any disruptions occur along the route. The vehicle then moves autonomously along the route using autonomous driving maneuvers.

[0003] Other road users are typically also moving along the route, such as autonomous vehicles, manually controlled vehicles, and / or pedestrians. To avoid an accident with these, the vehicle must move in a manner that is coordinated with the other road users. Consequently, it is necessary to estimate where these other road users will be, at least in the near future. The longer the forecasts are, the better the vehicle's behavior can be adapted accordingly, preventing relatively jerky driving maneuvers and thus improving comfort.

[0004] To determine the forecast, it can be assumed, for example, that the other road users are moving essentially uniformly along their current route, which is predetermined, for example, by a roadway. To improve the accuracy of the forecast, applicable traffic regulations can then be taken into account, whereby it is assumed that the other road users adhere to them. When determining the forecast, it is assumed that the other road users will stop at a stop sign along their route. However, it is also possible that another road user designed as an automated vehicle might malfunction, leading to a failure to observe the traffic regulations. Even in a manually controlled vehicle, the driver may in principle break traffic regulations.If the forecast is prepared in the same way in these cases, the forecast is predominantly incorrect, which is why the driving maneuvers of the vehicle determined on the basis of the forecast do not correspond to the situation that actually occurs.

[0005] DE 10 2017 223 621 A1 relates to a method and a control unit for controlling a function, in particular an output function and / or a guidance function, of an at least partially automated vehicle.

[0006] DE 10 2019 215 147 A1 relates to a method for controlling an ego vehicle, a computer program product, a driver assistance device and a vehicle with such a driver assistance device.

[0007] DE 10 2017 204 404 B3 relates to a method and a device for predicting the behavior of an object in the environment of a motor vehicle, as well as to a vehicle equipped with such a device. Furthermore, the invention relates to a method for controlling a driver assistance system of a motor vehicle based on a predicted behavior.

[0008] The invention is based on the object of specifying a particularly suitable method for operating a motor vehicle as well as a particularly suitable automated motor vehicle as well as a particularly suitable control device and a particularly suitable computer program product, wherein a reaction to the current situation is advantageously improved.

[0009] With regard to the method, this object is achieved according to the invention by the features of claim 1, with regard to the automated motor vehicle by the features of claim 8, with regard to the control unit by the features of claim 9, and with regard to the computer program product by the features of claim 10. Advantageous further developments and refinements are the subject of the respective subclaims.

[0010] The method is used for operating a motor vehicle. The motor vehicle is preferably land-based and preferably movable independently of rails or the like, and thus not rail-guided. Preferably, the motor vehicle comprises a number of wheels by means of which contact with the ground, in particular a lane, is established.

[0011] The motor vehicle is preferably a commercial vehicle, such as a lorry or a bus. However, the motor vehicle is particularly preferably a passenger car.

[0012] The motor vehicle expediently comprises a drive, which comprises, for example, an internal combustion engine, an electric motor, or a combination thereof. The drive is expediently operatively connected to at least some of the wheels, if any. Particularly preferably, the motor vehicle has a steering system that is operatively connected to at least some of the wheels. By actuating the steering system, it is possible to adjust the direction of travel of the motor vehicle. Upon actuation of the steering system, the wheels that are operatively connected to the steering system are suitably pivoted relative to any body of the motor vehicle.

[0013] The motor vehicle preferably comprises an assistance system which serves for longitudinal and / or lateral control. In other words, the assistance system is suitable, in particular provided and configured, for this purpose. For example, the assistance system serves only for longitudinal control or only for lateral control. However, the assistance system particularly preferably serves both longitudinal and lateral control. During longitudinal control of the motor vehicle, a speed of the motor vehicle is set and regulated to a specific value, which is, for example, constant or variable over time. For this purpose, the assistance system is used in particular to adjust any drive of the motor vehicle. During lateral control, a direction of travel of the motor vehicle is set by means of the assistance system and regulated to a specific value, for which the steering system is expediently used.

[0014] The control is carried out in particular by means of control commands generated by the assistance system, which expediently actuate the steering system and / or the drive. In other words, the control commands are transmitted from the assistance system to the drive or steering system, thus resulting in appropriate control. For example, a bus system of the motor vehicle, such as a CAN bus system or a Flexray bus system, is used for this purpose.

[0015] The control commands for the regulation are expediently created based on input data, wherein the input data in particular at least partially depicts a current situation of the motor vehicle, such that the control commands are created depending on a current situation of the motor vehicle. The input data comprise, for example, sensor data generated by sensors of the motor vehicle. A radar sensor, a LIDAR sensor, or a camera is used as a sensor in this case. In particular, the motor vehicle has several such sensors, such that the input data comprise several sensor data. In a further alternative, for example, a current location of the motor vehicle, which is determined in particular by means of a GPS sensor, is used at least partially as input data. Alternatively, or in combination with this, signals transmitted to the motor vehicle via radio are used as input data.In this case, the signals are transmitted to the vehicle using a Wi-Fi or mobile communications standard, for example. In another alternative, the input data includes map material and thus navigation data.

[0016] The control commands are generated based on the input data. For example, the assistance system and lateral control ensure that the vehicle is moving in a specific lane. In other words, the assistance system is a lane keeping assistant. The input data includes, for example, sensor data generated by a camera. This camera is used to detect any lane markings. In this case, the assistance system only serves for lateral control. If longitudinal control is performed using the assistance system, the current speed of the vehicle is used as input data, for example, whereby the current speed is expediently determined using a wheel speed sensor or GPS signals.

[0017] For example, longitudinal and lateral control is carried out using just a single assistance system. Alternatively, the motor vehicle has several individual systems that are combined to form an assistance system, with each of these systems providing corresponding control. Thanks to the assistance system, it is therefore possible to move the motor vehicle at least partially automatically. In this case, the motor vehicle in particular meets a so-called Level 1 or Level 2 according to the SAE definition. It is therefore not always mandatory for a driver of the motor vehicle to carry out the longitudinal or lateral control of the motor vehicle manually. Alternatively, for example, the control system and / or the assistance system are not present. Preferably, however, the motor vehicle meets an automation level corresponding to Level 3 or preferably higher, particularly due to the assistance system.In particular, the motor vehicle is an autonomous motor vehicle (Level 4) or a highly autonomous motor vehicle (Level 5).

[0018] The method detects an error made by another road user. For this purpose, the other road user is first detected, expediently using any sensors of the motor vehicle, such as the camera and / or radar sensor. Furthermore, it is detected that the other road user, who is expediently at a distance from the motor vehicle, has made an error, i.e., in particular, that they are behaving in a manner that deviates from expected behavior. In other words, the other road user did not behave in an expected manner due to the error, so the error exists. If the other road user is a motor vehicle, this constitutes, in particular, a driving error.

[0019] To determine whether such an error exists, for example, the other road user is recorded only once or, preferably, over a longer period of time. In particular, the surroundings of the motor vehicle are recorded, for which purpose the sensor(s) are used, for example. Alternatively or in combination with this, map material is analyzed, from which in particular the expected behavior of the other road user is derived. Alternatively or in combination with this, the expected behavior was derived, for example, on the basis of the actual surroundings, in particular as recorded by the camera. For this purpose, in particular road markings and / or lanes of road users were analyzed, such as ruts or tracks in snow or other surfaces. Based on these and the recorded behavior of the other road user, it is then checked whether the error exists.

[0020] If no error is detected, the procedure is terminated. Alternatively, the procedure is continued until the other road user has committed the error, which is then recorded. After the error is recorded, a further step involves classifying the error. In other words, the error is assigned to one of several classes. The classes are suitably predefined, in particular by a vehicle manufacturer.

[0021] In a next step, an initial forecast for the further behavior of the additional road user is created based on the classification. In other words, the first forecast describes the assumed further behavior of the additional road user, in particular in the near future, for example for the next second, the next 5 seconds, or the next 10 seconds. In particular, the length of the first forecast is less than 5 minutes, 2 minutes, or 1 minute. Preferably, the first forecast directly follows the current point in time. In particular, the further behavior explicitly or at least implicitly describes the future activity of the additional road user, so that an assumption is made regarding the future activities of the additional road user.

[0022] Machine learning, specifically a so-called "leverage machine learning model," is used to create the initial forecast. For example, a neural network is used. For example, each fault class is assigned a corresponding neural network. However, it is particularly preferred to use the same neural network, with the weighting factors of the neural network adjusted depending on the fault classification. This reduces the effort required. For example, the neural network is already fully trained and is not further modified during the operation of the vehicle. Alternatively, the neural network or other parts of the machine learning system are continually trained, particularly after the process has been implemented.In particular, after the procedure has been completed, the actual subsequent behavior of the road user is compared with the initial prediction, and the neural network or other machine learning algorithm is adjusted based on this. In summary, the initial prediction appropriately takes into account the habitual error-making of all or at least many other road users, particularly in this area.

[0023] In a subsequent step, a second prediction for the subsequent behavior of the other road user is created based on the classification. For example, the two predictions are created simultaneously, so that they are essentially available at the same time. Alternatively, they are created consecutively, thus reducing the required hardware resources. In particular, the second prediction also follows the current time and / or the length of the second prediction is equal to the length of the first prediction.

[0024] A predefined set of rules is used to create the second prediction. The set of rules comprises one or more rules, in particular a rule set / set of rules. These expediently include traffic regulations and / or rules specified by a vehicle manufacturer. In particular, the set of rules is rigid or can be changed as part of an update, for example, during a workshop visit or "over the air." For example, the set of rules includes a decision tree. In this case, the weighting factors, in particular, depend on the classification of the fault. Alternatively, a different decision tree is used for each class.

[0025] For example, the first and / or second forecast only covers a single manifestation of the further behavior. In other words, the respective forecast only covers one future activity or a sequence of future activities. Alternatively, the respective forecast includes several alternative manifestations of the further behavior, i.e. several activities that would have to be carried out in parallel. In particular, it is assumed that the further road user carries out either one of these activities or one of the other activities. In other words, the future activities are mutually exclusive. Preferably, different probabilities are assigned to each of the manifestations, or such an assignment of a probability is omitted.

[0026] After completing these two procedures, two forecasts for further behavior are available, which are, for example, the same or partially or completely different. Once the two forecasts are available, the intersection of the two forecasts is determined. If each of the forecasts only shows a single manifestation of the further behavior, but with multiple activities, the intersection only includes the activities that are present in both manifestations. If, on the other hand, several of the forecasts comprise multiple manifestations, the intersection only includes, for example, the manifestations that are present in both forecasts. Alternatively, all activities that are present in both forecasts are included in the intersection.Thus, it is possible, for example, that there are activities in the intersection that lead to characteristics that are not present in either of the two forecasts.

[0027] In a further step, a function is determined and executed depending on the intersection. In particular, it is checked whether the further behavior of the other road user contained in the intersection could endanger the motor vehicle and / or its occupants. If this is the case, the issuing of a warning is expediently used as the function, particularly if the motor vehicle is not autonomous. If, however, the motor vehicle is autonomous, the function used in particular is to bring the vehicle into a safe state, for which purpose the motor vehicle is brought to a standstill or the distance to the other road user is increased. Based on the function, an appropriate longitudinal and / or lateral acceleration is carried out.If, however, there is no danger and / or the distance to the other road user needs to be maintained, no warning is issued and / or the vehicle continues driving in the same manner. Alternatively, or in combination with this, depending on the intersection, the function used is to query an operator, particularly if the vehicle is autonomous. In this case, the preferred function is to request manual intervention from the operator.

[0028] In the method, if the other road user makes the error, the probability of another error being committed is checked, in particular based on the classification, which is stored in the prediction of the respective subsequent behavior. Since the initial prediction is created using machine learning, an adaptation to the current conditions and / or surroundings of the motor vehicle is carried out. This allows, for example, an adaptation to country-specific customs. Therefore, a warning is not issued and / or the direction of movement of the motor vehicle is not changed with excessive caution.Since the rule set is used for the second prediction, it is ensured that, for example, relatively serious errors are not ignored, even if they are committed by several other road users in quick succession, and continuous learning takes place during machine learning. In summary, the method thus estimates how the other road user will behave if it is already known that they have committed the error, i.e., deviated from the expected behavior. By selecting the appropriate function, it is possible to respond appropriately to the current situation, thus improving the response to the current situation.

[0029] The further behavior preferably comprises a trajectory of the further road user. In other words, the trajectory of the further road user is used in particular or at least also as further behavior. Consequently, the trajectory of the further road user is determined / estimated using the forecasts. For example, several trajectories are determined with each or one of the forecasts, so that several manifestations of the further behavior, i.e. several trajectories, are available. To determine the trajectory, the current position of the further road user is expediently first recorded and the trajectory is created based on this. The trajectory suitably comprises several activities, such as stopping at a certain point, one or more changes of direction, or changing and / or maintaining speed.To create the intersection, for example, it is checked which trajectory is completely present in both the first and second forecasts. Alternatively, only the individual activities, such as stopping at a certain point, are compared between the two forecasts and, if necessary, included in the intersection.

[0030] For example, the function is determined only directly as a function of the individual elements, such as the activities or the characteristics. However, it is particularly preferred to determine a confidence for the intersection, for example for each element of the intersection or for the intersection in general. In other words, a measure of the probability of the elements contained in the intersection occurring is thus specified. The confidence can expediently assume a value between 0 and 1. The function is preferably determined as a function of a comparison of the confidence with a threshold value. If the confidence is comparatively low, a different function is expediently used than if the conference is comparatively large.For example, in the case of a low confidence level which is smaller than the threshold value, a demand from a person, for example the driver or the operator of the motor vehicle, is used as a function, even if, for example, the further behavior contained in the intersection would show that there is no danger.

[0031] If the intersection is empty, for example, the confidence is preferably 0 or at least smaller than the threshold. In one further development, for example, the confidence is determined as a function of the size of the intersection. Alternatively, or in combination with this, the confidence is determined based on the ratio of the individual elements of the intersection with respect to the corresponding elements of the first and / or second forecast, for example, how many trajectories are in the intersection compared to the first forecast or the second forecast. Alternatively, for example, the confidence is determined as a function of a ratio of the individual activities in the intersection with respect to the number of activities in the two forecasts.Due to the confidence, it is thus ensured that randomly occurring elements of the intersection, which arise, for example, due to artifacts, do not lead to undesired behavior of the vehicle.

[0032] For example, the configuration of the other road user is not taken into account when classifying the error. However, it is particularly preferred that the other road user be further classified. This is done, for example, visually, in particular using a camera or, for example, by means of radio communication. In particular, it is checked whether the other road user is a pedestrian or, expediently, another motor vehicle. If this is a motor vehicle, it is checked in particular whether it is being piloted manually or whether it is an automatically moved or movable motor vehicle. This is taken into account when classifying the error. In particular, if the other motor vehicle drives into a prohibited area, in the case of an automatically moved motor vehicle, a classification is carried out that a malfunction has occurred.In the case of a manually piloted motor vehicle, for example, the classification that the driver was distracted is used. In particular, the further classification is also taken into account in the forecasts. If, for example, a malfunction occurs, the forecasts indicate that such a malfunction continues to exist as further behavior. If the further classification revealed that the error was made due to inattention, it is assumed in at least one of the forecasts that such an error will not occur again. Thus, in particular, the further classification of the other road user is also taken into account in the forecasts, at least indirectly via the classification. However, this is particularly preferably done directly when creating the forecasts. In this case, for example, any weighting factors of the decision tree and / or the neural network are adjusted accordingly.Due to the further classification, it is taken into account that automated motor vehicles and manually piloted motor vehicles make errors for different reasons, which is also reflected in future behavior.

[0033] Particularly preferred during classification is to check whether the error constitutes a violation of applicable traffic regulations or a deviation from normal behavior. In particular, normal behavior may also constitute at least a partial violation of applicable traffic regulations, such as failing to activate a turn signal or driving over a marking in a curve (in particular “cutting the curve”). In this case, all, many, or at least some of the road users present there commit the rule violation, which is why it is assumed that this is normal behavior. In particular, normal behavior is covered by the applicable traffic regulations. Such an error is an indication that the error was made unintentionally, which is taken into account in particular in the forecasts.

[0034] A deviation from normal behavior may occur, for example, in the event of a malfunction, when the vehicle is being driven automatically, or due to a driver impairment, such as a heart attack, or when the driver is drunk. Such an error could include, for example, comparatively slow driving, at least compared to other road users in the area, and / or comparatively fast / abrupt acceleration and / or deceleration. Another such error could be if the other road user is swerving without crossing the lane markings of the assigned lane. Staying correctly in the lane can also constitute such an error, namely when all other road users leave the lane, for example, in a curve.

[0035] A violation of applicable traffic regulations, on the other hand, indicates that the error was committed intentionally, such as driving too fast or leaving a lane, such as crossing a solid line or leaving the paved area, crossing stop lines, or failing to yield the right of way. In other words, such an error is predominantly committed by a so-called traffic offender. In particular, machine learning is trained regarding the behavior of a traffic offender and / or the rules are adapted accordingly.

[0036] Thanks to the classification, the predictions make it easier to assess whether the respective error was committed intentionally or unintentionally. This makes it easier to determine whether the other road user will continue to make errors. This improves the accuracy of the predictions.

[0037] For example, the forecast is created solely based on the classification. However, it is particularly preferred to also use an environment. In this case, the environment is expediently first determined and the forecasts are then created based on this. The environment is determined, for example, using a camera or, preferably, at least also using map material. The map material contains, in particular, applicable traffic regulations / traffic infrastructure that the other road user must / should adhere to. Depending on the assessment of the other road user, preferably based on the further classification, their further behavior can then be better deduced, which increases the accuracy of the forecasts.

[0038] For example, the forecasts are created solely based on the classification or, particularly preferably, also based on the determined environment. Particularly preferably, at least the current position and / or speed of the other road user is also used to determine the forecasts. In other words, these are first determined and then used to create one of the forecasts, or expediently both forecasts. Thus, based in particular on the current position and speed of the other road user, it is possible to deduce where the other road user will be in the near future, particularly if one of the forecasts assumes that the other road user will continue to move at a constant speed.

[0039] Alternatively or in combination with this, a traffic situation and / or the behavior of other road users is used to create the forecast. To do this, the traffic situation, expediently the other road users, is first recorded, in particular using sensors such as a camera and / or a radar sensor. Suitably, a corresponding forecast for the further behavior of the other road users is also created. If it is then assumed that the other road user and the further road user will encounter each other, the forecast is used to estimate how the latter will behave, for example whether they will yield right of way to the other road user. Depending on this, the period of time that the further road user will spend in the area of ​​the encounter is then specified. The accuracy when creating the forecast(s) is therefore further improved.

[0040] The motor vehicle can be moved at least partially automatically and is, for example, a Level 1 or Level 2 motor vehicle. However, the motor vehicle can preferably be moved autonomously or highly autonomously and thus meets the classification according to Level 4 or Level 5. The motor vehicle is, for example, a commercial vehicle (truck, bus) or a passenger car. The motor vehicle is in particular land-based and has a number of wheels by means of which contact with the ground is established. In particular, the motor vehicle has a drive and / or a steering system which are coupled by means of at least some of the wheels. Particularly preferably, the motor vehicle also comprises a braking system by means of which individual wheels or all of the wheels can be braked. Furthermore, the motor vehicle has an assistance system for longitudinal and / or lateral control. By means of the assistance system, the drive orThe steering system or the braking system is subjected to control commands. In particular, the assistance system makes it possible to change the direction of travel of the vehicle and / or its speed, i.e., to accelerate or decelerate it.

[0041] The motor vehicle is operated according to a method in which an error of another road user is detected, preferably by means of a sensor of the motor vehicle. The error is classified, and based on the classification, a first prediction of the further behavior of the other road user is created. Machine learning is used for this purpose. Furthermore, based on the classification, a second prediction of the further behavior of the other road user is created, for which a predefined set of rules is used. The intersection of the two predictions is determined, and a function is determined and executed depending on the intersection.

[0042] The motor vehicle expediently has a control unit that is suitable, in particular intended and configured, for at least partially carrying out the method. The control unit is, for example, a component of the assistance system or of another assistance system. The control unit has, for example, an application-specific integrated circuit (ASIC) that serves to carry out the method. Alternatively or in combination with this, the control unit comprises a microprocessor that is, in particular, designed to be programmable. The microprocessor in this case forms, in particular, a computer, and the method is expediently stored in a computer program product that can be carried out by means of the computer. In the assembled state, the control unit is expediently a component of a motor vehicle and is suitable, in particular provided and configured, for this purpose. Expediently, the control unit is a component of an assistance system and preferably the (sole or main) control unit of the assistance system. In this case, the assistance system expediently serves for the longitudinal and / or lateral control of the motor vehicle. Alternatively, the assistance system serves in particular to warn a driver of the motor vehicle, i.e. in particular to output information. The control unit comprises, for example, an application-specific integrated circuit (ASIC) and / or a microprocessor, which is expediently designed to be programmable. Expediently, the control unit has a memory which is suitable, in particular provided and configured, for storing a computer program product on it.The control unit is provided and configured to carry out a method in which an error of another road user is detected, preferably by means of a sensor of the motor vehicle. The error is classified, and based on the classification, a first prediction for further behavior of the other road user is created. Machine learning is used for this purpose. Furthermore, based on the classification, a second prediction for the further behavior of the other road user is created, for which a predefined set of rules is used. The intersection of the two predictions is determined, and a function is determined and executed depending on the intersection.

[0043] The computer program product comprises a number of instructions which, when executed by a computer, cause the computer to carry out a method for operating an at least partially automated motor vehicle, in which method an error of another road user is detected, preferably by means of a sensor of the motor vehicle. The error is classified, and based on the classification, a first forecast for further behavior of the other road user is created. Machine learning is used for this purpose. Furthermore, based on the classification, a second forecast for the further behavior of the other road user is created, for which a predefined set of rules is used. The intersection of the two forecasts is determined, and depending on the intersection, a function is determined and executed.

[0044] The computer is expediently a component of a control unit or electronics and is formed, for example, by means of the latter. The computer preferably comprises a microprocessor or is formed by means of the latter. The computer program product is, for example, a file or a data carrier containing an executable program which, when installed on a computer, automatically executes the method. The invention further relates to a storage medium on which the computer program is stored. Such a storage medium is, for example, a CD-ROM, a DVD, or a Blu-ray Disc. Alternatively, the storage medium is a USB stick or other memory which is, for example, rewritable or can only be written to once. Such a memory is, for example, a flash memory, a RAM, or a ROM.

[0045] The further developments and advantages explained in connection with the method are also to be transferred analogously to the automated motor vehicle / the control unit / the computer program product / the storage medium as well as to each other and vice versa.

[0046] An embodiment of the invention is explained in more detail below with reference to a drawing. In the drawings: Fig. 1 schematically shows an at least partially automated motor vehicle, Fig. 2 a method for operating the at least partially automated motor vehicle, and Fig. 3 schematically shows a bird's eye view of the motor vehicle at different times during the procedure.

[0047] Corresponding parts are provided with the same reference numerals in all figures.

[0048] In Fig. 1 shows a simplified schematic representation of a motor vehicle 2 in the form of a passenger car (car). The motor vehicle 2 has a plurality of wheels 4, by means of which contact is made with a roadway (not shown in detail). Some of the wheels 4 are driven by a drive 6, so that the motor vehicle 2 can be moved. Some of the wheels 4 are also operatively connected by means of a steering system 8, by means of which an angular angle of these wheels 4 relative to a body of the motor vehicle 2 can be adjusted. Consequently, the direction of travel of the motor vehicle 2 can be adjusted by means of the steering system 6. In addition, the motor vehicle 2 comprises a braking system 10 by means of which all wheels 4 can be braked.

[0049] The drive 6, the steering system 8, and the braking system 10 are signal-connected to an assistance system 12 or are a component of the assistance system 12. The assistance system 12 comprises a control unit 14, by means of which control commands for the drive 6, the steering system 8, and the braking system 10 are generated. Human intervention is not required here, so that the motor vehicle 2 is designed to be movable automatically. The motor vehicle 12 comprises a plurality of sensors 16, by means of which the surroundings of the motor vehicle 2 can be detected. The control unit 14 generates the control commands depending on the surroundings detected by the sensors 16 and on the basis of map material (not shown in detail) stored in a memory 18 of the control unit 14. In summary, the assistance system 12 thus serves to longitudinally and laterally control the motor vehicle 2, which can be moved automatically.

[0050] A computer program product 20 is also stored on the memory 18. This includes several instructions which, when the program is executed, cause a computer 22 of the control unit 14 to Fig. 2 for operating the motor vehicle 2. In other words, the motor vehicle 2 is operated according to the method 24, and the control unit 14 is provided and configured to carry out the method 24.

[0051] In the method 24, in a first step 26, an environment 26 of the Fig.2 from a bird's-eye view. For this purpose, the surroundings 26 are directly detected by means of the sensors 16. Furthermore, the current position of the motor vehicle 2 is determined by means of a position sensor (not shown in detail), and the corresponding map material is retrieved and compared with the data generated by the sensors 16. Consequently, the course of the roadways 28 present in the surroundings 26 and the traffic regulations applicable there are known.

[0052] In the example, sensors 16 detect two additional road users 30, each of which is a motor vehicle, and each of which has committed an error 32 in the situation shown. One of the two additional road users 30 is moving in a serpentine pattern along the roadway 28, which merges into the roadway 28 on which the motor vehicle 2 is located. A desired path 34 of the motor vehicle 2 leads to the roadway 28 on which this additional road user 30 is currently traveling.

[0053] The other road user 30, however, turned from another street onto lane 28, along which the motor vehicle 2 is currently traveling, without activating the turn signal. Furthermore, the other road user 30 did not stop at a stop sign located there, but moved slowly into lane 28 without endangering other objects. This other road user 30 is approaching the motor vehicle 2 in the opposite direction. On the way there, however, the other road user 30 stopped to allow another road user 36, namely a pedestrian, to cross lane 28 safely. The current position and speed of the respective road user 30 are also recorded by the sensors 16. The behavior of the other road user 36 and therefore also the traffic situation were also recorded.If the other road users 30 continue to move unchanged and the motor vehicle 2 continues to move along the desired route 34, they would meet in the area of ​​the junction, whereby motor vehicle 2 would have the right of way due to traffic regulations not shown in detail.

[0054] In a subsequent second work step 38, the other road users 30 are further classified. For this purpose, a visual check is performed using sensors 16 and a radio communication (not shown in detail), which is used to query the other road users 30 as to whether they are capable of automated movement. Based on corresponding feedback from the other road user 30 approaching from the left, this other road user 30 is classified as an automated vehicle, whereas due to a lack of feedback and a visual check, the other road user 30 approaching is classified as a manually piloted vehicle.

[0055] In a subsequent third step 40, the error 32 committed by each of the two additional road users 30 is classified. For the additional road user 30 approaching from the left, error 32 is zigzagging, but the additional road user 30 does not leave the assigned lane, so error 32 does not constitute a violation of applicable traffic regulations. Rather, it is a deviation from normal behavior. The classification also takes into account that the additional road user 30 is an automated motor vehicle, so error 32 is (sub-)classified as a malfunction.

[0056] The other road user 30, however, committed error 32 by not stopping at the stop sign and also not using the turn signal. This constitutes a violation of applicable traffic regulations, but not a deviation from normal behavior. Several road users, when there is no danger to other objects, do not stop completely at a stop sign and, in the majority of cases, do not use the turn signal. Since the other road user 30 is a manually piloted motor vehicle, it is also assumed that error 32 was made intentionally.

[0057] In summary, in the third step 40, an assumption is made based on the classification as to the reason why the respective error 32 was committed. Errors 32 that lead to a classification as a violation of applicable traffic regulations include, for example, breaking speed limits or leaving passable areas, such as leaving an asphalt area or crossing a solid line. Failure to yield right of way or ignoring signals such as traffic lights or a stop sign are also such errors. Driving in the wrong direction or changing lanes without activating a turn signal is also a corresponding error. It is possible that such an error 32 is committed intentionally, for example, due to a criminal act or because the driver is drunk.However, it's possible that the error is committed unintentionally or unconsciously, for example, due to health reasons such as a heart attack. However, the probability of a deliberate violation is increased.

[0058] However, classification as a deviation from normal behavior occurs, for example, when the other road user 30 is moving in a zigzag pattern. Another error 32 leading to this is driving too slowly or excessive acceleration and / or deceleration. Another corresponding error 32 is not negotiating curves when almost all road users are negotiating the curve. Such errors 32 occur for the same reasons as the errors 32 that lead to classification as a violation of applicable traffic regulations. However, the probability of unconscious behavior is increased here.

[0059] In a fourth work step 42, based on the classification of the respective error 32 and the further classification of the other road user 30, a first forecast 44 for the further behavior of each additional road user 30 is created. Trajectories 46 are used as further behavior, so that the trajectories 46 are created using the first forecast 44. Thus, based on the current position and speed of the respective additional road user 30 and the determined environment 26, namely the data recorded by the sensors 16 as well as the map material and the traffic rules stored therein, assumptions for their future movement are determined. The first forecast 44 uses machine learning, for which a neural network 47 stored in the memory 18 and already trained is used.

[0060] When creating the trajectories 46, the traffic situation and the behavior of the other road users 46 are also taken into account. Since the oncoming road user 30 allowed them to cross the roadway 28, this results in, for example, rule-compliant behavior for them, provided no other objects are endangered. As a result, only a single trajectory 46 results for them, with a stop at the junction, since both the motor vehicle 2 and the remaining road user 30 will be there. For the remaining road user 30, however, the neural network 47 determined that a sensor of the other road user 30 is malfunctioning, so that no specific direction in which the other road user 30 is moving can be specified.In summary, machine learning takes into account the position, speed, current traffic conditions, the environment 26, such as the presence of traffic signs, and the behavior of others in the environment. In summary, machine learning uses the position, speed, current traffic situation, the environment 26, such as the presence of traffic signs, and the behavior of others in the environment, to determine the trajectories 36.

[0061] In a fifth step 48, which is performed simultaneously with the fourth step 42, a second prediction 50 is created for the further behavior of the other road users 30, which also includes or can include multiple trajectories 46 of the respective road user 30. For this purpose, a decision tree 52 stored in the memory 18 is used, which maps a predefined set of rules. Several rules are stored in the set of rules so that even unusual situations can be queried.

[0062] The rules specify, for example, that it is assumed that the other road user 30 will not stop for pedestrians in the future and / or that they will use a comparatively dangerous trajectory if they have not stopped for pedestrians once. Another rule, for example, is that if the direction indicator is not used, it will also not be used in the future. Another corresponding rule is that comparatively late and sharp braking occurs if the other road user 30 does not maintain a safe distance. These rules are, in particular, summarized into a set of rules that is generally applicable. The rulebook also includes, in particular, a set of rules that only apply to certain situations that are specified, for example, on the basis of traffic rules, regulations and / or other standards.For example, it is assumed that if the other road user 30 violates a speed limit, they will continue to move at an increased speed, in particular, greater than the permitted speed. A corresponding rule is also used that if a solid line is crossed, the other road user 30 will continue to cross it, or that the other road user 30 will move in the lane not intended. If, however, the other road user 30 runs a red light, it is assumed based on the rules that traffic lights will continue to be ignored. When creating the second forecast 50, the environment 26 and the traffic signs stored / determined therein are also used.When creating the second forecast 50, the current position, speed, traffic situation and the behavior of other road users 36 are also taken into account.

[0063] In a sixth step 54, the intersection 56 of both forecasts 44, 50 is determined. This checks which of the trajectories 46 were determined in both the first forecast 50 and the second forecast 50 and correspond to each other. A confidence level is also determined for the intersection 56, namely for the trajectories 46 contained therein. The confidence level can assume values ​​between 0 and 1.

[0064] For the oncoming road user 30, each forecast 44, 50 contains only the individual trajectories 46 in which a stop occurs at the junction, so that the intersection 56 also only includes these. Since each forecast 44, 50 only includes a single trajectory 46, the confidence is comparatively high and amounts to 1 in the example shown.

[0065] For the other road user 30 coming from the left, however, each of the predictions 44, 50 includes a large number of different tractors 46, which partially overlap, so that the intersection 56 also has a large number of trajectories 46. However, the confidence is comparatively low and is, for example, equal to 0.05 for each of the trajectories 46.

[0066] In a subsequent seventh step 58, a function 60 is determined and executed depending on the intersection 56 and the comparison of the confidence with a threshold value. If only the oncoming road user 30 were present, the movement of the motor vehicle along the desired route 34 would be used as function 60 based on the intersection 56, namely the only trajectory 46 in which a stop occurs at the junction, as well as the comparison of the confidence, which is 1, with the threshold value, which is 0.5, especially since any danger to motor vehicle 2 is ruled out.

[0067] However, the other road user 30 approaching from the left is also present. The confidence of the intersection 56 is lower than the threshold, and based on the trajectories 46, no desired route 34 can be derived that excludes any danger to motor vehicle 2. Therefore, stopping and querying an operator of motor vehicle 2 are used as function 60.

[0068] The invention is not limited to the exemplary embodiment described above. Rather, other variants of the invention can also be derived therefrom by those skilled in the art without departing from the scope of the invention. In particular, all individual features described in connection with the exemplary embodiment can also be combined with one another in other ways without departing from the scope of the invention. List of reference symbols 2 motor vehicles 4 wheel 6 Drive 8 Steering system 10 Braking system 12 Assistance system 14 Control unit 16 sensors 18 storage 20 Computer program product 22 computers 24 procedures 26 Surroundings 28 Roadway 30 other road users 32 errors 34 desired route 36 other road users 38 second step 40 third step 42 fourth step 44 first forecast 46 Trajectory 47 neural network 48 fifth step 50 second forecast 52 Decision tree 54 sixth step 56 Intersection 58 seventh step 60 Function

Claims

[1] Method (24) for operating a motor vehicle (2), in particular one which is movable at least partially automatically, in which - an error (32) of another road user (30) is detected, - a classification of the error (32) is carried out, - based on the classification, a first prediction (44) for further behaviour of the other road user (30) is made, for which machine learning is used, - based on the classification, a second prognosis (50) is created for the further behaviour of the other road user (30), for which a predefined set of rules is used, - the intersection (56) is determined from the two forecasts (44, 50), and - a function (60) is determined and executed depending on the intersection (56). [2] Method (24) according to claim 1, characterized bythat the further behavior comprises a trajectory (46) of the further road user (30). [3] Method (24) according to claim 1 or 2, characterized by that a confidence is determined for the intersection (56), and that the function (60) is also determined depending on a comparison of the confidence with a threshold value. [4] Method (24) according to one of claims 1 to 3, characterized by that a further classification of the further road user (30) is carried out and taken into account in the classification of the error (32). [5] Method (24) according to one of claims 1 to 4, characterized by that for classification purposes it is checked whether the error (32) is a violation of applicable traffic rules or a deviation from normal behaviour. [6] Method (24) according to one of claims 1 to 5, characterized by that the forecasts (44, 50) are made based on an identified environment (26). [7] Method (24) according to one of claims 1 to 6, characterized by that a current position, speed, traffic situation and / or behavior of other road users (36) is used when creating the forecasts (44, 50). [8] Automated movable motor vehicle (2) which has an assistance system (12) for longitudinal and / or lateral control and which is operated according to a method (24) according to one of claims 1 to 7. [9] Control device (14) which is provided and arranged to carry out a method (24) according to one of claims 1 to 7. [10] A computer program product (20) comprising instructions which, when executed by a computer (22), cause the computer (22) to carry out a method (24) according to any one of claims 1 to 7.

Images