Method for generating a model to support a lateral guidance assistant, method for operating a lateral guidance assistant in a vehicle, and lateral guidance assistant for a vehicle

Abstract

The invention relates to a method for generating a model (10) to support a lateral guidance assistant (1), wherein data from a vehicle fleet are collected and / or collected data from the vehicle fleet are obtained, wherein the data are collected or were collected while the lateral guidance assistant (1) is or was active, wherein time points in the data are identified at which a driver-initiated target course change of the lateral guidance assistant (1) has occurred, wherein time series for segments of the data around the respective time point are generated for at least some of the identified time points, wherein commonalities in the time series of the data are sought, and wherein, based on the commonalities found, the model (10) is generated and provided, wherein the model (10) is designed or is configured such thatthat, based on at least some of the quantities (5-x) contained in the data, this estimates and provides at least one probability (4) for a driver-initiated change of target course. Furthermore, the invention relates to a method for operating a lateral guidance assistant (1) in a vehicle (50) and a lateral guidance assistant (1) for a vehicle (50).

Description

[0001] The invention relates to a method for generating a model for supporting a lateral guidance assistant, a method for operating a lateral guidance assistant in a vehicle and a lateral guidance assistant for a vehicle.

[0002] Assistance systems for lateral vehicle guidance are common in current production vehicles and are designed to make driving safer and more comfortable. These systems typically provide more lane-centering guidance and, when the driver exerts a directed hand torque on the steering wheel for a longer period, offer the option of providing an adapted desired trajectory as the target course. The goal of such a target course adjustment is to ensure increased driver comfort through the adjusted target course and, in particular, to prevent the driver from feeling like they are working against the system.

[0003] The adjustment of the target course is initiated by the driver and requires a certain amount of time until the target course is adjusted, which can initially result in an uncomfortable feeling of the situation, since the driver and the assistance system adjust to a different target course (i.e. different target trajectories) during this period.

[0004] Only in exceptional situations can a set course adjustment be performed automatically, such as in a traffic jam to create an emergency lane. These exceptional situations are usually easily described by experts and result from legal requirements and the desire for increased safety. Beyond the current state of technology, however, various situations increase the likelihood of a driver-initiated set course adjustment.

[0005] It is therefore desirable to be able to adjust the target course more quickly in other situations in order to further increase the driver's sense of comfort.

[0006] US 2023 / 0 322 234 A1 describes a personalized lane change prediction for a vehicle. A learning-based lane change prediction algorithm, as well as systems and methods for implementing the algorithm, are described. The prediction algorithm evaluates the driving behavior of a human target driver and predicts lane change maneuvers based on this personalized driving behavior. The algorithm may include an online prediction phase for lane change decisions and an offline prediction training and cost function recovery phase. During the offline training phase, a machine learning model may be trained based on historical vehicle states.During the online validation phase, driving data can be collected and fed into the trained model to predict a driver's lane change maneuver, identify possible vehicle trajectories, and determine the most likely vehicle trajectory based on a driver cost function recovered during the offline phase.

[0007] DE 10 2023 000 315 A1 describes a method for recommending the use of an assistance system for vehicles in a fleet that are data-linked to a central computer unit. It is proposed that an input from a vehicle driver regarding the use of the assistance function be stored in a data set with information about a system state of the vehicle, information about a driver state, and driving context information. The vehicle data sets are transmitted to the central computer unit and collected. For each driver input, a clustering method is applied by the central computer unit to form clusters with identical or similar system states of the vehicles, driver states, and driving context information, if a cluster has a certain relevance.the cluster driver's input is saved as a preferred vehicle setting and made available to the vehicle fleet, and if a vehicle has a similar system state, a driver of the vehicle has at least a similar state, and / or the driving context information is similar, a recommendation for using the assistance function of the assistance system is issued to the driver of the vehicle and / or the assistance system is automatically parameterized.

[0008] DE 10 2019 123 899 B3 describes a method and a device for autonomous driving of a vehicle on a roadway in a direction of travel, wherein a trajectory for driving on the roadway in the direction of travel is determined, wherein a bending band limited in its length defines the trajectory, wherein the bending band is fixed at one end in a node that defines a starting point of the trajectory, wherein a course of the trajectory starting from the starting point is determined depending on a bending line of the bending band that extends from the node to the other end of the bending band, wherein a representation of a roadway boundary defines a boundary condition for determining the trajectory, wherein a quality measure is defined depending on a property of the bending band, in particular a bending stiffness, a tension or a centering of the bending band with respect to the representation of the roadway boundary,wherein, in particular, according to the principle of virtual displacement, preferably by means of the finite element method, the bending line is determined which satisfies the boundary condition and for which the quality measure has an extreme value.

[0009] The invention is based on the object of providing a solution to the aforementioned problem.

[0010] The object is achieved according to the invention by a method having the features of patent claim 1, a method having the features of patent claim 7, and a lateral guidance assistant having the features of patent claim 10. Advantageous embodiments of the invention emerge from the subclaims.

[0011] In particular, in a first aspect, a method is provided for generating a model to support a lateral guidance assistant, wherein data from a vehicle fleet are collected and / or collected data from the vehicle fleet are obtained, wherein the data are collected or were collected while the lateral guidance assistant is or was active, wherein points in time are identified in the data at which a driver-triggered target course change of the lateral guidance assistant occurred, wherein for at least some of the identified points in time, time series are generated for sections of the data around the respective point in time, wherein commonalities are sought in the time series of the data, and wherein, based on the commonalities found, the model is generated and provided, wherein the model is or is designed in such a way,that, based on at least some of the variables contained in the data, it estimates and provides at least one probability for a driver-triggered target course change.

[0012] Furthermore, in a second aspect, a method for operating a lateral guidance assistant in a vehicle is provided, wherein at least one probability for a driver-triggered target course change in the automated lateral guidance is estimated by means of a model on the basis of currently recorded and / or received variables, wherein an adjustment of the target course in the automated lateral guidance takes place taking into account the estimated at least one probability.

[0013] Furthermore, in a third aspect, a lateral guidance assistant for a vehicle is created, comprising a control device, wherein the control device is configured to provide a model, wherein the model estimates at least one probability for a driver-triggered target course change in the automated lateral guidance based on input data, further to feed currently recorded and / or received variables to the model as input data, and to estimate the at least one probability for a driver-triggered target course change in the automated lateral guidance based on the currently recorded and / or received variables by means of the model, and to adapt a target course in the automated lateral guidance taking into account the estimated at least one probability.

[0014] The methods and the lateral guidance assistant enable a faster adjustment of the target course if the driver requests a target course adjustment. For this purpose, a model is created according to the first aspect, which is used in the method according to the second aspect and the lateral guidance assistant according to the third aspect. The model is designed such that, based on recorded and / or received current variables, such as sensor data of the vehicle, control data of the vehicle, environmental data and / or context data, etc., it estimates at least one probability for a driver-triggered adjustment of the target course, i.e., for a change of target course. The adjustment of the target course of the lateral guidance assistant then takes into account the at least one probability. This allows the adjustment of the target course to take place, in particular, taking into account the current environment and / or a current context.

[0015] The method for generating a model to support a lateral guidance assistant is executed, for example, on a computing device, for example a central server and / or a cloud system.

[0016] The lateral guidance assistant controls the lateral guidance of the vehicle, in particular along a target course (i.e., a target trajectory). If no target course is specified by the driver, control is generally based on a reference course, for example, a virtual center line between two lane markings. If the driver wishes to adjust this target course, they can apply a steering torque to the steering wheel, which will cause the lateral guidance assistant to adjust the target course to a new target course after a while. For example, the adjustment of the target course can consist of shifting it 10, 20, or 50 cm to the left or right with respect to the virtual center line. According to the aspects described, this is done taking into account at least one probability.

[0017] The vehicle fleet data particularly includes collected swarm data, where the data may include variables such as vehicle sensor data, vehicle control data, environmental data and / or context data, etc. The data is recorded and / or was recorded while the lateral guidance assistant is or was activated. Times are identified in the data at which a driver-triggered target course change of the lateral guidance assistant occurred. For at least some of the identified times, time series are generated for excerpts of the data around the respective time. For example, it can be provided to place the time window symmetrically around the respective time, for example with a width of 10, 30 or 60 seconds each. The time series then comprise excerpts of the data in the respective time windows. In these time series, commonalities in the time series of the data are sought.The idea behind it is to find patterns in the data that occur frequently in the time series and that correlate in time with the driver-triggered target course change. This way, based on the patterns found, a driver-triggered target course change can later be predicted during lateral guidance. For example, such a pattern can include the following characteristics or criteria: the vehicle is driving on the motorway in a construction zone with narrower lanes and a truck is driving next to the vehicle on the right. In this case, it is more likely that a driver wants a greater distance from the truck. This greater distance can be achieved by changing the target course from the virtual center line to a target course that is, for example, 20 cm further to the left. The model is generated and provided based on the similarities found, in particular based on the patterns found.In other words, the model specifically maps the commonalities or patterns. The model is or will be designed in such a way that, based on at least some of the variables contained in the data, it estimates and provides at least one probability for a driver-triggered target course change. In the example mentioned, the model would then, for example, estimate a higher probability for a driver-triggered target course change based on environmental data (here: a truck detected to the right of the vehicle) and a context (here: a construction zone on a highway) than if the vehicle were driving on an empty highway outside of a construction zone.

[0018] Possible categories and information that can be used when evaluating the data to find commonalities (or patterns) include the following: - a vehicle type (e.g. vehicle dimensions): drivers of smaller vehicles may be less inclined to change course than drivers of larger and / or wider vehicles; - Information from the (ego) vehicle, such as speed, acceleration, steering interventions (steering angle, steering angle change, steering torques), etc.; - Map information: roadway, road type, lanes, speed limit, construction site, hazard area, road condition, lane marking condition, lane width, etc.; - Trajectory information: curvature, curvature prediction, etc.; - Traffic information: objects around the (ego) vehicle, in particular an object type, an object speed, a relative speed, a traffic flow, a traffic density, etc.; - Weather conditions: rain (e.g. detected by a rain sensor), clouds, sun, outside temperature, slippery road warning, friction coefficient, etc.

[0019] It may be intended that the search for commonalities be carried out at least partially with the help of human expert knowledge. In particular, it may be intended that human experts search the data for patterns, identify these patterns, and create the model based on them. The model can then be generated, for example, using rules. In the aforementioned example, a rule may state, for example, that the probability of a driver-triggered target course change is high if the vehicle is driving on a highway in a construction zone and a truck is driving alongside the vehicle.

[0020] Alternatively or additionally, the search for commonalities can be carried out using a data mining method. This allows commonalities to be identified automatically. In particular, it can also identify patterns that cannot be recognized and / or described / modeled by human experts, for example, due to their complexity and / or number of dimensions.

[0021] It can be provided that the data, in particular the collected swarm data, is designed in such a way that in addition to a drivable path that is uploaded based on the vehicle fleets, it is also detected and / or recorded whether a target course change was initiated by a driver during a journey, i.e. whether a deviation was made from a path originally used as a reference course (e.g. by means of a marker in the data: Flag_Sollkurswechsel == TRUE, if the lateral guidance assistant is active AND a target course change > limit value has occurred; the limit value can be a few centimeters, for example 10 cm; the limit value can also be a function of a road class and / or a lane width). This can improve the subsequent analysis. In particular, it can be directly identified whether a target course change has been initiated or not.

[0022] Parts of the lateral guidance assistant, in particular the control device, can be implemented individually or collectively as a combination of hardware and software, for example, as program code executed on a microcontroller or microprocessor. However, it can also be provided that parts are implemented individually or collectively as an application-specific integrated circuit (ASIC) and / or a field-programmable gate array (FPGA).

[0023] In one embodiment, the search for commonalities is carried out using K-means clustering. This allows commonalities to be identified automatically. In this case, the time series identified from swarm data are used, and the categories described above are passed to the clustering algorithm as differentiating features (algorithm inputs = number of categories). Subsequently, n clusters are formed that combine similarities between the categories. The number n of clusters can be determined either by experts or automatically (e.g., using the elbow criterion). The result represents combinations of categories for which the probability of a target course change (target course adjustment) is increased.

[0024] Another possibility is to divide the time series into subsets, e.g. 1. slight adjustment to the left; 2. strong adjustment to the left; 3. slight adjustment to the right; 4. strong adjustment to the right.

[0025] In principle, other clustering or classification methods can also be used.

[0026] In one embodiment, at least part of the model is configured as a decision tree. This allows values ​​of the recorded and / or obtained variables or combinations thereof to be directly assigned to values ​​for the at least one probability.

[0027] In one embodiment, the model comprises a multidimensional characteristic map. This allows values ​​of the recorded and / or obtained variables to be assigned to values ​​for the at least one probability for different situations. The characteristic map has the particular advantage that it can be generated and interpreted by a human. The assignment can be performed, for example, using a multidimensional function with respect to the input parameters (i.e., the recorded and / or obtained variables).

[0028] In one embodiment, the model comprises a machine learning method, a recurrent neural network, and / or a long short-term memory (LSTM). This allows the at least one probability to be estimated even for more complex scenarios with a multitude of variables that must be processed as input data. The machine learning method, in particular a neural network, is trained in a conventional manner, in particular based on the identified commonalities, in particular the identified patterns. In particular, data originating from situations in which no driver-triggered target course change has occurred can also be used in order to also represent the opposite case in the training data.

[0029] In one embodiment, the model estimates at least values ​​for the following probabilities as output: an adjustment to the left, an adjustment to the right, and no adjustment. This allows a statement to be made about the type of target price change. In particular, this allows for automated adjustment of the target price.

[0030] In one embodiment, the model estimates values ​​for further adjustment gradations as output. This allows, in particular, the extent of an automated adjustment of the target price to be fine-tuned. For example, the following adjustments or probabilities can be provided: a slight adjustment to the left, a strong adjustment to the left, a slight adjustment to the right, a strong adjustment to the right, and no adjustment.

[0031] In one embodiment of the method for operating a lateral guidance assistant in a vehicle, the set course is automatically adjusted if at least one probability estimated by the model exceeds a predetermined threshold. This can further increase convenience, as the driver no longer has to initiate the set course adjustment at all; instead, it is carried out automatically by the lateral guidance assistant. For example, if the probabilities are for a slight adjustment to the left, a strong adjustment to the left, a slight adjustment to the right, a strong adjustment to the right, and no adjustment, an adjustment to the left or right by, for example, 10 or 20 cm (slight adjustment) or 50 cm (strong adjustment) can be made if the respective probabilities exceed a threshold of, for example, 0.8.

[0032] In a further embodiment of the method for operating a lateral guidance assistant in a vehicle, it is provided that a time until the driver-triggered adjustment of the target course is determined taking into account the at least one probability. This allows preconditioning of the lateral guidance assistant. In particular, this can further increase comfort for the driver. If an estimated probability for an adjustment to the left has a high value, for example, the target course can be changed more quickly if it has been detected that the driver wants to change the target course to the left than if this probability has a lower value. In the latter case, the driver would then have to apply the manual torque to the steering wheel for a longer time to effect a target course change.It can be provided that a dependency of the time until the driver-triggered target course change is selected to be inversely proportional to the estimated probability for the requested target course change.

[0033] Further features of the device design will become apparent from the description of embodiments of the method. The advantages of the device are the same as those of the embodiments of the method.

[0034] The invention will be explained in more detail below using preferred embodiments with reference to the figures. Fig. 1 a schematic flow diagram to illustrate embodiments of the method for generating a model to support a lateral guidance assistant; Fig. 2 a schematic flow diagram to illustrate embodiments of the method for operating a lateral guidance assistant in a vehicle; Fig. 3 a schematic representation to illustrate embodiments of the lateral guidance assistant for a vehicle; Fig. 4 a schematic representation to illustrate an embodiment; Fig. 5 a schematic representation to illustrate embodiments.

[0035] The Fig. 1 shows a schematic flow diagram of an embodiment of the method for generating a model for supporting a lateral guidance assistant.

[0036] In a measure 100, data from a vehicle fleet is collected. Alternatively or additionally, collected data from the vehicle fleet can be obtained, for example, if it has already been collected. The data includes, in particular, sensor data of the vehicle, control data of the vehicle, environmental data and / or context data, etc. The data is or was collected while the lateral guidance assistant is or was active. It can be provided that the times at which the driver initiated a driver-triggered target course change are additionally recorded or were recorded directly in the data. For this purpose, information marking this time can be recorded in the data, for example.

[0037] In a measure 101, points in time in the data at which a driver-triggered target course change of the lateral guidance assistant occurred are identified. This can be done either by evaluating a driven trajectory, for example, based on a change in the trajectory with respect to a lane marking, and / or based on information recorded directly in the data, as described above.

[0038] In a measure 102, time series are generated for at least some of the identified points in time for data segments around the respective point in time. For this purpose, the data around the points in time are extracted, for example, within a time window of + / - 30 seconds around the respective point in time. Subsequently, a large number of segments are available, in which it can be assumed that the data depicted therein coincide with the driver-triggered target course change.

[0039] In measure 103, commonalities are searched for in the time series of data. This can be done manually by experts. Alternatively or additionally, the search for commonalities can be carried out using a data mining method. In particular, the search for commonalities can be carried out using K-means clustering. In particular, the aim is to find patterns in the sections that are characteristic of situations in which a driver-triggered target course change is requested by the driver.

[0040] In a measure 104, the model is generated and provided on the basis of commonalities found, wherein the model is designed in such a way that, based on at least some of the variables contained in the data, it estimates and provides at least one probability for a driver-triggered target course change.

[0041] In a measure 105, it can be provided that the model is loaded into a memory of a lateral guidance assistant. In particular, a structural description of the model and parameters of the model are loaded into the memory so that the generated model can be used by the lateral guidance assistant.

[0042] The Fig. 2 shows a schematic flow diagram of an embodiment of the method for operating a lateral guidance assistant in a vehicle.

[0043] In a measure 200, at least one probability for a driver-triggered target course change during automated lateral guidance is estimated using a model based on currently recorded and / or received variables. The model is, in particular, a model that was generated and provided according to the method described in this disclosure.

[0044] In action 201, it is checked whether a driver has requested a set course change. This can be done, for example, by checking whether a manual torque detected on a steering wheel exceeds a specified threshold.

[0045] If the check in action 201 reveals that the driver desires a change of target course, the target course for the automated lateral guidance is adjusted in action 202, taking into account the at least one probability estimated in action 200. The process then returns to action 200.

[0046] However, if the review in action 201 shows that no change in target course is desired, the system returns to action 200 without further action.

[0047] Measure 202 may, in particular, provide for a time until the driver-triggered adjustment of the target course to be determined taking into account at least one probability. In particular, the greater the estimated probability, the shorter the time until the driver-triggered adjustment.

[0048] Alternatively, in action 201, it can also be checked whether the estimated probability exceeds a specified threshold. If this is the case, the target price is automatically adjusted. Otherwise, the system returns to action 200 without further action.

[0049] The Fig. 3 shows a schematic representation to illustrate an embodiment of the lateral guidance assistant 1. The lateral guidance assistant 1 is arranged in particular in a vehicle 50 and, in the active state, serves for automated lateral guidance of the vehicle 50. For this purpose, the lateral guidance assistant 1 receives, for example, sensor data 5-x from at least one sensor 51 of the vehicle 50 and generates control data 20 which are fed to a steering system 52 of the vehicle 50.

[0050] The lateral guidance assistant 1 comprises a control device 2. The control device 2 comprises in particular a computing device with a memory (both not shown), which executes program code designed to carry out measures of the method.

[0051] The control device 2 is configured to provide a model 10. Based on input data 3, the model 10 estimates at least one probability 4 for a driver-triggered target course change during automated lateral guidance. The model 10 was generated, in particular, according to the method described in this disclosure.

[0052] The control device 2 is further configured to supply currently recorded and / or received variables 5-x to the model 10 as input data 3 and to estimate the at least one probability 4 for a driver-triggered desired course change during automated lateral guidance based on the currently recorded and / or received variables by means of the model 10.

[0053] Furthermore, the control device 2 is configured to adapt a target course during automated lateral guidance taking into account the estimated at least one probability 4.

[0054] It can be provided that the model 10 includes a multidimensional characteristic map 11. This is shown schematically in the Fig. 4. The recorded and / or received variables 5-x are fed to the model 10. For example, variable 5-1 can be a weather category (dry, rain, snow, etc.), variable 5-2 can be a road category (e.g., motorway, motorway with construction site, country road, country road with construction site, main road, secondary road, etc.), and variable 5-3 can be a distance to a vehicle, in particular a lorry, in an adjacent lane. Alternatively or additionally, other variables 5-x can also be taken into account. Based on these variables 5-x, the probability is determined or estimated using the multi-dimensional characteristic map 11. For example, it can be provided that, based on variables 5-1 and 5-2, a characteristic curve is selected from the characteristic map 11 (e.g., for the following combination: rain and motorway with construction site), which links a value of variable 5-3 with a value for probability 4.For example, the characteristic curve has a shape that means that the smaller the distance to a vehicle in the adjacent lane, the greater the probability 4 of a driver-triggered course change.

[0055] It can be provided that the model 10 comprises a machine learning method 12, a recurrent neural network 13 and / or a long short-term memory (LSTM) 14. This is shown schematically in the Fig.5. During an application phase, the recorded and / or received variables 5-x are fed to the model 10 as input data 3. The variables 5-x can include the variables 5-x described above. Alternatively or additionally, other variables 5-x can also be considered. Based on the variables 5-x, the machine learning method 12, the recurrent neural network 13, and / or the long short-term memory (LSTM) 14 estimate at least one probability 4 for the driver-triggered target course change during automated lateral guidance.

[0056] In the context of generating the model 10, the machine learning method 12, the recurrent neural network 13 and / or the long short-term memory (LSTM) 14 are trained, in particular, with the aid of training data. The training data are generated, in particular, based on the time periods. It can be provided, in particular, that equally sized time periods are added to the training data in which no driver-triggered target course change has occurred, in order to also take the opposite case into account in the training data. In this case, the ground truth in the training data is labeled, for example, with the labels 0. no adjustment; 1. slight adjustment to the left; 2. strong adjustment to the left; 3. slight adjustment to the right; 4. Strong right-adjustment, and model 10, for example, neural network 13, receives the ground truth as well as the time series containing the categories (descriptors). In other words, each time series is assigned one of the label instances as ground truth. Accordingly, model 10 learns a functional relationship between the categories and the presence of a target price change (a target price adjustment).

[0057] However, it can also be planned that the machine learning method 12 is trained using unsupervised learning. In the case of unsupervised learning, the time series identified from swarm data are used, and the categories described above are passed to the clustering algorithm as differentiating features (algorithm inputs = number of categories). Subsequently, n clusters are formed that combine similarities between the categories. The number n of clusters can be determined either by experts or automatically (e.g., using the elbow criterion). The result represents combinations of categories for which the probability of a target course change (target course adjustment) is increased.

[0058] Another possibility is to divide the time series into subsets, e.g. 1. slight adjustment to the left; 2. strong adjustment to the left; 3. slight adjustment to the right; 4. strong adjustment to the right.

[0059] Model 10 may be intended to estimate as output at least values ​​for the following probabilities: a left-hand adjustment, a right-hand adjustment, and no adjustment.

[0060] Furthermore, model 10 can be configured to estimate values ​​for further adjustment levels as output. For example, the following adjustments or probabilities can be configured: a slight adjustment to the left, a strong adjustment to the left, a slight adjustment to the right, a strong adjustment to the right, and no adjustment. List of reference symbols 1 lateral guidance assistant 2 Control device 3 Input data 4 Probability 5-x size / sensor data 10 Model 11 map 12 Machine Learning Methods 13 recurrent neural network 14 Long short-term memory (LSTM) 20 tax data 50 vehicles 51 Sensor 52 Steering 100-105 Measures of the procedure 200-202 Measures of the procedure

Claims

[1] Method for generating a model (10) for supporting a lateral guidance assistant (1), wherein data of a vehicle fleet are collected and / or collected data of the vehicle fleet are received, wherein the data are collected or were collected while the lateral guidance assistant (1) is or was active, wherein points in time are identified in the data at which a driver-triggered target course change of the lateral guidance assistant (1) has occurred, wherein for at least some of the identified points in time, time series are generated for sections of the data around the respective point in time, where commonalities are sought in the time series of the data, and wherein the model (10) is generated and provided on the basis of commonalities found, wherein the model (10) is designed or is configured in such a way that, based on at least some of the variables (5-x) contained in the data, it estimates and provides at least one probability (4) for a driver-triggered target course change. [2] Method according to claim 1, characterized by that the search for commonalities is carried out using K-Means clustering. [3] Method according to one of the preceding claims, characterized by that the model (10) comprises a multidimensional characteristic map (11). [4] Method according to one of the preceding claims, characterized by that the model (10) comprises a machine learning method (12), a recurrent neural network (13) and / or a long short-term memory (LSTM) (14). [5] Method according to one of the preceding claims, characterized bythat the model (10) estimates as output at least values ​​for the following probabilities (4): a left-hand adjustment, a right-hand adjustment, and no adjustment. [6] Method according to claim 5, characterized by that the model (10) estimates values ​​for further gradations of adjustments as output. [7] Method for operating a lateral guidance assistant (1) in a vehicle (50), wherein at least one probability (4) for a driver-triggered desired course change in the automated lateral guidance is estimated on the basis of currently recorded and / or received variables (5-x) by means of a model (10) generated according to a method according to one of claims 1 to 6, wherein an adjustment of the desired course in the automated lateral guidance takes place taking into account the estimated at least one probability (4). [8] Method according to claim 7, characterized bythat the target price is automatically adjusted if the at least one probability (4) estimated by the model (10) exceeds a predetermined threshold. [9] Method according to claim 7 or 8, characterized by that a time until the driver-triggered adjustment of the target course is determined taking into account at least one probability (4). [10] Lateral guidance assistant (1) for a vehicle (50), comprising: a control device (2), wherein the control device (2) is configured to provide a model (10) generated according to a method according to one of claims 1 to 6, wherein the model (10) estimates at least one probability (4) for a driver-triggered desired course change in the automated lateral guidance based on input data (3), further supplying currently recorded and / or received variables (5-x) to the model (10) as input data (3), and estimating the at least one probability (4) for a driver-triggered desired course change in the automated lateral guidance based on the currently recorded and / or received variables (5-x) by means of the model (10), and to adapt a target course for automated lateral guidance taking into account the estimated at least one probability (4).

Images