Vehicle and method for outputting a recommendation to a vehicle driver to take over vehicle control

Abstract

According to the present invention, the vehicle has a recommendation system including a data collection module, a prediction module and a recommendation module, the data collection module being configured to collect vehicle data, surrounding data and / or environmental data, the prediction module being configured to read a driver profile from a plurality of driver profiles, each driver profile including a machine learning model trained individually for each respective driver profile, the model being configured to read vehicle data, surrounding data and / or environmental data for at least the section located ahead and to output a predicted index value as an output variable, the recommendation module being configured to compare the predicted index value with an index threshold and configured to prompt the output of a recommendation to take over vehicle control by the vehicle driver or by the driver assistance system depending on the position of the predicted index value relative to the index threshold.

Description

[Technical Field]

[0001] The present invention relates to a vehicle with an at least partially automated driving mode and to a method for outputting recommendations to a vehicle driver of a vehicle of the type defined in detail in the preamble of claim 10.

[0002] Modern vehicles are increasingly equipped with driver assistance systems. Driver assistance systems support the vehicle driver when he or she takes over the driving task. This can increase road traffic safety, vehicle driving efficiency, and vehicle user comfort. In this case, various driver assistance systems can take over control of at least partially automated, and in the future, autonomous, vehicles. For example, a lane-following assist can keep the vehicle in its current lane when driving on a highway, an adaptive cruise control can maintain a distance from a preceding vehicle, or a parking assist can take over lateral control of the vehicle when reversing into a parking space, while the vehicle driver takes over longitudinal control.

[0003] To utilize a driver assistance system that enables at least partially automated vehicle control, boundary conditions must exist in a driving situation that allow the use of the driver assistance system. For example, sensors used for perimeter detection must not be hindered and must provide sufficient usable sensor data, or certain driving conditions must exist, such as traveling within a specified speed range on a specific type of road. If an at least partially automated assistance function is available, it typically must first be manually activated. However, in such cases, the vehicle driver may not use the assistance function even though it is available. This may be due to, for example, a loss of driving enjoyment or the pleasure of driving independently, or a lack of confidence in the safety of the corresponding driver assistance system or a loss of control. This would result in the corresponding benefits in terms of safety, efficiency, and comfort that may arise from using the assistance function being lost.

[0004] Patent Document 1 discloses a driving system and a method for activating a driving function for automated driving. The method disclosed in this document facilitates the operation of a driving system that provides an automated driving function. After a vehicle driver inputs a request to use the automated driving function, the method checks the availability of the corresponding automated driving function and, if available, provides the driving function to the vehicle driver. If the automated driving function is provided, the vehicle driver can accept the provision, i.e., activate the automated driving function, in a particularly comfortable manner. For this, it may be sufficient for the vehicle driver to release their hands from the steering wheel to activate the automated driving function. This eliminates the need for the vehicle driver to first manually check the general availability of the automated driving function while using their vehicle, and then, if available, manually enter an operating action to activate it. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] German Patent Application Publication No. 102017208504 Summary of the Invention [Problem to be solved by the invention]

[0006] The present invention is based on the problem of providing an improved vehicle with an at least partially automated driving mode and a method for outputting recommendations to the vehicle driver to take over control of the vehicle, and by using these, increasing the proportion of use of at least partially automated driving of the vehicle during travel. [Means for solving the problem]

[0007] According to the invention, this problem is solved by a vehicle having the features of claim 1 and a method for outputting a recommendation to a vehicle driver of a vehicle for the vehicle driver to take over manual vehicle control or for a driver assistance system to take over at least partially automated vehicle control, having the features of claim 10. Advantageous embodiments and developments are evident from the dependent claims.

[0008] According to the present invention, a vehicle with an at least partially automated driving mode has a recommendation system including a data collection module, a prediction module and a recommendation module, wherein the data collection module is configured to collect vehicle data, surrounding data and / or environmental data, the prediction module is configured to read one driver profile from a plurality of driver profiles, each driver profile including a machine learning model individually trained for each respective driver profile, the model is configured to read vehicle data, surrounding data and / or environmental data for at least a section located ahead and to output a prediction index value as an output variable, the prediction index value being a numerical value, the recommendation module is configured to compare the prediction index value with an index threshold and, when the prediction index value is in a first range compared to the index threshold, to prompt the vehicle driver to take over manual vehicle control at the vehicle, and when the prediction index value is in a second range compared to the index threshold, to prompt the driving assistance system to output a recommendation to take over at least partially automated vehicle control.

[0009] The vehicle according to the present invention can suggest to the vehicle driver to take over manual vehicle control, taking into account the driving style of the vehicle driver, precisely when the vehicle driver is likely to willingly want to take over control of the vehicle in accordance with his / her personal preferences for vehicle use, and can output a recommendation to take over vehicle control via the driver assistance system when a corresponding driver assistance function is available and the vehicle driver is likely not wanting to control the vehicle himself / herself. In other words, when the vehicle driver is highly likely to comply with a request to take over at least partially automated vehicle control, the vehicle outputs a recommendation to the vehicle driver to take over at least partially automated vehicle control, thereby increasing the utilization of at least partially automated vehicle control while traveling in the vehicle.

[0010] When a situation exists in which at least partially automated vehicle control is possible, the vehicle driver is prompted to hand over vehicle control to the driver assistance system, thereby preventing at least partially automated vehicle control from remaining unused during a section of the journey in which this function is possible. However, the vehicle driver may not want to hand over control of the vehicle despite the availability of at least partially automated driving functions, for example, because the vehicle driver is experiencing driving pleasure during that section. In such a case, a recommendation to hand over vehicle control to the vehicle is not output, thereby preventing unnecessary recommendations from being output and increasing the comfort of the vehicle driver. It is also possible to output a recommendation to the vehicle driver when at least partially automated vehicle control is enabled but the vehicle still takes over manual vehicle control. The vehicle determines, according to the vehicle driver's personal preferences, which sections the vehicle driver is likely to want to control the vehicle himself / herself and outputs a corresponding recommendation. This ensures that the vehicle driver will control the vehicle himself / herself when traveling through a corresponding section that generates driving pleasure. This will lead to increased acceptance of vehicle operators' use of at least partially automated driving functions, resulting in increased use of at least partially automated driving functions when these functions are available.

[0011] For example, if the vehicle driver encounters a monotonous or boring section during the journey, the vehicle may output a recommendation to take over at least partially automated vehicle control, and the vehicle driver may then willingly hand over vehicle control to the at least partially automated driver assistance system. In this case, the vehicle may read the vehicle driver's preferences from the driver profile and thereby recognize situations ahead in which the vehicle driver would normally be willing to control the vehicle manually, but does not currently want to do so due to changed boundary conditions, for example, because there is too much traffic and / or bad weather conditions and / or the vehicle is currently in a phone conversation and / or there are other passengers in the vehicle. Thus, the vehicle recognizes possible driving situations in which automatic control can increase driving safety through at least partially automated control compared to manual control, and based on this, the vehicle driver may hand over vehicle control to the driver assistance system, thereby increasing the vehicle driver's sense of security.

[0012] In this case, the vehicle can have any degree of automation. That is, the vehicle can be controlled at least partially autonomously, which corresponds to vehicle control assisted by a driver assistance system. For example, the driver assistance system can take over lateral or longitudinal control of the vehicle, with the corresponding longitudinal or lateral control of the remaining vehicle being taken over by the vehicle driver. However, the vehicle can also be controlled highly automatically, fully automatically, or even autonomously. In an autonomous driving mode, intervention by the vehicle driver is no longer required, so the vehicle can generally perform its driving task even when the vehicle driver is absent or distracted.

[0013] In this case, the data collection module collects vehicle data, surrounding data, and / or environmental data. Vehicle data includes information related to the vehicle, detected by sensors on the vehicle itself, such as the forward speed, vehicle acceleration, vehicle position, steering angle, accelerator pedal position, etc. Environmental data includes information related to the vehicle's immediate surroundings, determined by the vehicle itself, such as the course determined by evaluating camera images generated by a surrounding camera, the self-measured road surface condition, road width, measured ambient temperature, traffic regulations indicated by traffic signs, construction sites and other obstacles in the roadway area also recognized in camera images generated by the surrounding camera, and precipitation detected using a humidity sensor. Environmental data includes information related to the vehicle's surroundings, obtained from outside the vehicle or read from a storage device. This includes, for example, the course and traffic regulations applied to the course read from a digital road map, traffic information from a traffic service, weather forecasts, etc. The surrounding data can be used to verify the environmental data. Furthermore, corresponding data can also be estimated for the section ahead. For example, it can read a route from a digital road map or determine the next 200 meters from a camera image and predict the steering angle, acceleration values ​​and forward speed to expect when driving the corresponding section. To process this "heterogeneous" data, the prediction module can also use semantic network methods such as knowledge graphs and / or ontologies.

[0014] Vehicle data, surrounding data, and / or environmental data may affect the vehicle driver's control behavior and are therefore taken into account to determine a recommendation to take over manual or at least partially automated vehicle control. For example, a sporty driver may want to drive the vehicle manually, especially on sections with many curves. However, if slippage is expected, for example, due to wetness or sub-zero temperatures, the vehicle driver may want at least partially automated vehicle control. Another example of a driving situation in which at least partially automated vehicle control is desired by the vehicle driver is a monotonous driving situation, such as a traffic jam or traffic stagnation, or driving on a particularly long road section with a certain speed limit. Therefore, various driving situations can be recognized and distinguished in particularly different ways, especially taking into account vehicle data, surrounding data, and environmental data.

[0015] In abstract terms, a driver profile stores the associations between a person with a particular driving style and manually controlling the vehicle according to boundary conditions representing each driving situation. In this case, the boundary conditions are represented by vehicle data, surrounding data, and / or environmental data. The boundary conditions are then processed by a machine learning model that is trained to suit the driving style of each driver profile and stored in the predictive model. In this case, the vehicle driver first determines, for example, when purchasing the vehicle or at the start of each journey, which driver profile he or she feels is assigned to the vehicle. Next, the predictive module selects and loads a machine learning model or heuristic model belonging to the selected driving style from the storage device assigned to it. The entire driver profile can be specified, for example, by the automobile manufacturer. In this case, a certain number of standard driver profiles can be predefined. Examples of driver profiles can be sporty, proactive, reliable, a novice driver, an older driver, or a physically disabled driver.

[0016] The machine learning model reads vehicle data, surrounding data and / or environmental data as input data, i.e., for example, the brake pedal position predicted for the section ahead, steering angle, acceleration values, position values, traffic congestion reports, altitude profile, course, weather information, etc. As an output variable, a prediction index value is output, which corresponds to a numerical value within a specified value range, for example, between 0 and 1 or between -1 and 1. In this case, different trained machine learning models are provided for different driver profiles.

[0017] To train the various machine learning models, different drivers are tasked with driving tasks for the various courses, indicating whether manual or at least partially automated vehicle control is desired for each section. The various machine learning models then learn how the vehicle should be optimally controlled according to the driving style associated with the driver profile, depending on each course and the boundary conditions represented by the vehicle data, surrounding data, and / or environmental data. Correspondingly, the various machine learning models then output predictive index values ​​appropriate for each course.

[0018] The predicted index value is then compared to an index threshold by a recommendation module, and depending on the result, a recommendation for manual vehicle control or at least partially automated vehicle control is output for the section ahead.

[0019] The index threshold can be specified as a fixed value. For example, the index threshold can be 0.5. For example, the first range can be a range between 0.5 and 1, and the second range can be a range between 0 and 0.5. The first or second range can also include the value "0.5." For example, the first range can be a range of values ​​from 0.5 to 1, and the second range can be a range of values ​​from 0 to 0.499, or vice versa. In this case, the first and / or second ranges do not necessarily need to be above or below the index threshold. The first and second ranges can also be interpreted as percentages. For example, the first range can correspond to a range of ≧1, and the second range can correspond to a range between 0 and 1. It is also possible that the first range corresponds to a positive range and the second range corresponds to a negative range, or vice versa.

[0020] The recommendation to take over vehicle control can be output in various ways in the vehicle, for example acoustically, visually and / or haptically. For example, the steering wheel of the vehicle can vibrate when a new recommendation is output. The recommendation can then be displayed acoustically and textually, for example as a voice message such as: "There is a winding section ahead. It is recommended that you take over vehicle control manually." A corresponding message, possibly supplemented by pictograms, symbols, images, videos, etc., can also be displayed on any display device of the vehicle, for example the display of the instrument cluster or head unit.

[0021] A vehicle can determine the section ahead in various ways. For example, the vehicle can have a navigation system with a pre-programmed route. In this case, the route is divided into sections. The corresponding section ahead is then determined from the route the vehicle must follow further. If route guidance is not available, the vehicle can also use a so-called most likely route, which the navigation system determines based on historical data. If route guidance is not available, the vehicle can determine the section immediately ahead as the section ahead. For example, the next 100 meters of travel can be determined as the closest section ahead. If the section ahead includes an intersection, an exit road, an access road, etc., the closest section ahead may extend to such a road element. The vehicle can also determine the section ahead without using a navigation system. For this purpose, the vehicle can evaluate vehicle data and, for example, determine the course itself. For this purpose, the vehicle can recognize the course immediately ahead using depth information sensors, such as a stereo camera, a LIDAR, an ultrasonic sensor, and / or a radar system.

[0022] An advantageous development of the vehicle contemplates that the recommendation system further includes a driver monitoring module, which is configured to monitor the vehicle driver using at least one sensor, determine a current control behavior and / or a current driver state from sensor data generated by the sensor, and assign the vehicle driver to one of the driver profiles according to the determined current control behavior and / or the current driver state. In this way, the vehicle driver's driving behavior is gradually recognized, and a driver profile that is sufficiently suitable for the corresponding driving behavior is selected by the vehicle, thereby outputting other predictive indicator values. This can further improve the comfort of the vehicle driver by increasing the degree of agreement between the vehicle driver's actual driving style and the driving style estimated by the vehicle, thereby allowing the recommendation system to output a more accurate recommendation to take over vehicle control. This increases the likelihood that the recommendation to take over vehicle control output by the vehicle will actually be implemented by the vehicle driver.

[0023] To detect the current control behavior, the steering behavior and acceleration or braking behavior of the vehicle driver can be analyzed. For this purpose, the steering angle and pedal position, which are adjusted accordingly during driving, are detected and analyzed by the driver monitoring module.

[0024] The driver state represents, among other things, the vehicle driver's emotional mood, level of attention and / or cognitive load. To this end, vital parameters of the vehicle driver are monitored, such as pulse rate, breathing rate, skin conductance, skin temperature, gaze direction, blinking frequency, and frequency of changes in gaze direction. For example, an increase in pulse rate, skin conductance and / or body temperature can indicate increased concentration and / or the experience of driving enjoyment.

[0025] For example, a vehicle driver who remains calm in congested traffic situations and / or feels stressed in monotonous traffic situations may be assigned to a sporty driver profile, whereas a vehicle driver who becomes excited in congested traffic situations and / or feels relaxed in monotonous traffic situations may be assigned to a comfortable driver profile.

[0026] The prediction module can further be configured to estimate a predicted control behavior and / or a predicted driver state for the section located ahead, i.e., an expected current control behavior and / or a driver state that is likely to occur in the section located ahead. Thus, the predicted control behavior can be compared with the current control behavior and / or the predicted driver state with the current driver state in order to evaluate the predictive quality of the prediction module, in particular of the respective machine learning model underlying the corresponding driver profile.

[0027] In order to assign a suitable driver profile to a vehicle driver, the implementation behavior of the recommendation to take over vehicle control output by the vehicle can also be taken into consideration. If the vehicle driver follows the recommendation of the vehicle more frequently than a defined threshold, the vehicle considers that a suitable driver profile has been selected. On the other hand, if the vehicle driver does not follow the recommendation to take over vehicle control made by the vehicle or follows it less frequently than a defined threshold, it is preferable to set a new driver profile.

[0028] The predefined "standard driver profile" can be further adapted to the actual driving style of the vehicle driver over the vehicle's usage period, depending on the vehicle driver's actual control behavior and / or driver state, and its predictive quality can be further improved by continuous training. For this purpose, according to another advantageous embodiment of the vehicle, the driver monitoring module is further configured to determine a current indicator value from the current control behavior and / or the current driver state, and the prediction module is further configured to read the current control behavior, the current driver state and / or the current indicator value and further train the machine learning model using this. The machine learning model or heuristic can be further trained using the above-mentioned current data as well as data about the surroundings and / or environment of the currently driven section, which is currently read by the data acquisition module. This further improves the predictive ability of the machine learning model for the section ahead, thereby making the recommendations output by the recommendation system more frequently implemented by the vehicle driver.

[0029] The predicted control behavior and / or predicted driver state can also be input to the machine learning model as input variables for determining a prediction index value. By comparing with the current control behavior and / or current driver state, the prediction module can check how accurately the predicted control behavior and / or predicted driver state was predicted. The machine learning model learns accordingly and improves its prediction quality.

[0030] Similarly, a machine learning model trained for each driver profile can be integrated into the driver monitoring module to determine current index values ​​from current data, and the current index values ​​output by the driver monitoring module can also be taken into account by the prediction module to check the predictive quality of predicted index values ​​based on the current index values ​​and to adapt the machine learning model to minimize errors.

[0031] Another advantageous embodiment of the vehicle further provides that the prediction module is further configured to further train the machine learning model taking into account actual implementation of the recommendation to take over vehicle control made by the recommendation module, whereby the corresponding result regarding whether the recommendation to manually or at least partially automatedly take over vehicle control was actually implemented or rejected by the vehicle driver is returned as feedback by the control device to the prediction module, thereby further improving the machine learning or heuristic model and the predictions made therewith.

[0032] According to another advantageous embodiment of the vehicle, the recommendation system is configured to receive fleet data, the fleet data including aggregated amounts of user-specific control behaviors and / or driver states of users of a fleet of vehicles, and to derive a driver profile from the fleet data and / or update an existing driver profile, such that similar control behaviors and / or driver states within a defined limit are assigned to the same driver profile. By collecting and using fleet data, a particularly comprehensive data set can be used to train a corresponding machine learning model for each derived driver profile. This makes it possible to develop a particularly wide range of different driver profiles, thereby allowing a particularly appropriate machine learning model to be developed for each vehicle driver or assigned to each individual vehicle user. Furthermore, the use of these large data sets also allows particularly appropriate training of machine learning models specific to each individual driver profile. This means that the predictive quality of each machine learning model for each driver profile can be improved, thereby ensuring that each recommendation output by the recommendation system is implemented in practice as much as possible.

[0033] In this case, in addition to the user-specific control behavior, the respective user-specific driver state can also be taken into account. In this way, the control behavior can be linked to the driver state, so that there are also driver profile-specific driver states for a specific section or vehicle data, surrounding data, and / or environmental data. For example, on a particularly steep and winding route, with frequent and large steering and strong acceleration and deceleration behavior, a first driver state indicates high cognitive load and anxiety, and a second driver state indicates medium cognitive load and enjoyment. Therefore, a person with the first driver state can be assigned to a timid driver profile, and a person with the second driver state can be assigned to a sporty driver profile.

[0034] In another advantageous additional or alternative embodiment of the vehicle, the recommendation system is configured to receive fleet data, which further includes implementation behaviors by users of the fleet of vehicles of recommendations made by the recommendation module of the fleet of vehicles, and the prediction module is further configured to improve or further train the machine learning model taking into account the implementation behaviors of users of the driver profile assigned to the vehicle driver. This can be understood as a filter to ensure that, before the machine learning model is further trained, the corresponding input data used for further training actually leads to an improvement in the implementation of the output recommendation in multiple users of the fleet. This prevents a deterioration in the quality of the machine learning model, which may lead to a decrease in the implementation of the corresponding recommendation. For example, depending on the vehicle data, surrounding data, and / or environmental data, or predicted control behaviors or driver states derived therefrom, sections can be determined in which at least a predetermined number of users, for example at least 75% of the users, actually implement the output recommendation.

[0035] According to another advantageous embodiment of the vehicle, the machine learning model comprises a neural network, which, but also an equivalent machine learning method, can be used to particularly ensure that the machine learning model is able to determine suitable predictive indicator values ​​depending on the input parameters.

[0036] Another advantageous embodiment of the vehicle further contemplates that the vehicle control device is configured to automatically implement the recommendation made by the recommendation system to have the vehicle driver or driver assistance system take over vehicle control. This can further improve the comfort of the vehicle driver. Therefore, the vehicle driver does not need to input a manual operating action each time a corresponding recommendation is output to follow the recommendation. That is, the vehicle can automatically implement the corresponding recommendation, i.e., enable or disable at least partially automated vehicle control. The vehicle driver can, for example, set in a configuration menu that automatic takeover of the recommendation should occur in the case of certain boundary conditions. In this case, in certain situations, the vehicle driver may still need to first respond to the output recommendation before the recommendation is even implemented. For example, at least partially automated vehicle control can always be automatically enabled, and manual vehicle control takeover can be enabled only after confirmation by the vehicle driver.

[0037] Preferably, the recommendation module is further configured to adapt the index threshold value depending on the driver profile, vehicle data, surrounding data, and / or environmental data read by the prediction module. This allows for particularly simple, fast, and flexible reactions to determine which vehicle driving mode should be recommended. For example, if the index threshold value has an initial value of, for example, 0.5, it can be increased or decreased depending on the driver profile. If a manual vehicle control recommendation is output with an index value relatively close to 1, it can be decreased, for example, to 0.3 for a "sporty" driver profile and increased, for example, to 0.8 for a "safe" driver profile. Thus, the index threshold value can also be adapted depending on the vehicle data, surrounding data, and / or environmental data. The vehicle can analyze the section ahead and, for example, set the index threshold to 0.7 for a monotonous section and 0.33 for an exciting section.

[0038] According to the present invention, a method for outputting a recommendation to a vehicle driver of a vehicle as described above for taking over manual vehicle control by the vehicle driver or for taking over at least partially automated vehicle control by a driver assistance system comprises the following steps: - collecting vehicle data, surrounding data and / or environmental data by a data collection module; - reading, by a prediction module, one driver profile from a plurality of driver profiles, each driver profile including a machine learning model individually trained for the respective driver profile, the model being configured to read vehicle data, surrounding data and / or environmental data for at least a section located ahead and to output a prediction index value as an output variable, the prediction index value being a numerical value; - determining, by a prediction module, a prediction indicator value for the preceding interval; - comparing, by a recommendation module, the predicted indicator value with an indicator threshold, and prompting output of a recommendation for the vehicle driver to take over manual vehicle control in the vehicle if the predicted indicator value is within a first range relative to the indicator threshold, and prompting output of a recommendation for the driver assistance system to take over at least partially automated vehicle control if the predicted indicator value is within a second range relative to the indicator threshold.

[0039] Using the method according to the present invention, a recommendation to take over manual or at least partially automated vehicle control is output to the vehicle driver in the vehicle. In this case, the output of each recommendation is adjusted to the respective driver profile, so that a recommendation for manual vehicle control is output when the vehicle driver is particularly likely to actually want to control the vehicle manually, and a recommendation for at least partially automated control is output correspondingly when the vehicle driver wants at least partially automated vehicle control or does not object to it. This improves the acceptance of driver assistance systems that enable at least partially automated vehicle control, thereby increasing the proportion of at least partially automated vehicle control used during driving.

[0040] Further advantageous embodiments of the vehicle according to the invention will become apparent from the examples which are described in detail below with reference to the drawings. [Brief explanation of the drawings]

[0041] [Figure 1] 1 is a schematic diagram of a vehicle according to the present invention; [Figure 2] FIG. 1 is a schematic diagram of the information pathways of variables read and output by the machine learning model executed in the prediction module. DETAILED DESCRIPTION OF THE INVENTION

[0042] 1 shows a schematic front view of a vehicle 1 according to the invention. The vehicle 1 is equipped with a recommendation system 2 which includes a data collection module 2.1, a prediction module 2.2, a recommendation module 2.3 and a driver monitoring module 2.4.

[0043] The vehicle 1 has an at least partially automated driving mode. Depending on various boundary conditions, the at least partially automated driving mode is available, allowing the vehicle driver 4 to at least partially automatically control the vehicle 1. For example, this can be an adaptive cruise control, a lane following assist, a parking assist, or even an autopilot, i.e., an autonomous control of the vehicle 1. The vehicle 1 according to the present invention can output a recommendation to the vehicle driver 4 whether manual vehicle control should be performed by the vehicle driver 4 or whether at least partially automated vehicle control should be performed by a driver assistance system in the section ahead during the journey performed by the vehicle 1. In this case, the vehicle 1 outputs a recommendation for manual vehicle control precisely when the vehicle driver 4's preferences are met and it is highly likely that the vehicle driver 4 wants to control the vehicle 1 himself, and outputs a recommendation for at least partially automated vehicle control when at least partially automated vehicle control contributes to increasing the user comfort of the vehicle driver 4.

[0044] For this purpose, the data collection module 2.1 collects vehicle data D-FZG, surroundings data D-UMG, and environmental data D-UMW. The vehicle data D-FZG is information related to the vehicle 1, such as its forward speed, its position or orientation in space, acceleration values, pedal position, etc., detected by the vehicle 1 itself. The surroundings data D-UMG is information related to the vehicle 1's surroundings, such as its course, applicable traffic regulations, the surrounding temperature, and the current amount of precipitation, detected by the vehicle 1. To determine its course, the vehicle 1 can scan its surroundings using surrounding sensors, such as mono or stereo cameras, LIDAR, ultrasonic sensors, and / or radar systems, and determine its course from the corresponding sensor data. The environmental data D-UMW includes information related to the surroundings read by the vehicle 1 from an external device or data storage device, such as the course read from a digital road map, the traffic regulations stored in the digital road map, particularly for the section ahead, traffic information from a traffic service, and weather forecasts from a weather service.

[0045] Furthermore, the driver monitoring module 2.4 is configured to monitor the vehicle driver 4 and detect a current control behavior D-SV-AKT and a current driver state D-ZUS-AKT using various sensors. The current control behavior D-SV-AKT includes, for example, the steering behavior and / or acceleration or braking behavior of the vehicle driver 4, derived from a steering angle sensor and / or a pedal position sensor. The current driver state D-ZUS-AKT represents, for example, the emotional mood, attention level, and cognitive load of the vehicle driver 4 due to the driving task. For this purpose, various sensors can be used to monitor the vehicle driver 4, thereby detecting vital parameters. Corresponding variables that are considered for determining the driver state can then be derived from the vital parameters. Examples of vital parameters include pulse rate, eye blink frequency, facial expressions of the vehicle driver 4, such as whether the vehicle driver 4 is smiling, skin conductance, skin temperature, gaze direction, and the frequency of gaze direction changes, derived from visual monitoring using an image recognition algorithm.

[0046] Furthermore, the recommendation system 2 can also receive fleet data D-FLO, for example via the data collection module 2.1, which comprises the control behavior and / or driver states of different users of the vehicles 1 of the fleet, from which different driver profiles can be derived that describe their driving style and the emotions they experience when doing so.

[0047] For example, before using the vehicle 1, the vehicle driver 4 selects which driver profile he or she feels he or she belongs to, such as a sporty driver profile or a comfortable driver profile. Then, depending on the selected driver profile, the prediction module 2.2 selects a trained machine learning model or heuristic model that determines whether the vehicle driver 4 prefers to control the vehicle 1 manually or at least partially automatically for the section ahead. The selected machine learning module or heuristic model of the prediction module 2.2 processes at least the vehicle data D-FZG, surrounding data D-UMG, and / or environmental data D-UMW predicted for the next section ahead, thereby enabling a realistic estimation of the driving situation that will occur in the section ahead. Each driver profile represents the expected manual control behavior of the vehicle driver 4 depending on the expected driving situation in the section ahead. This allows the prediction module 2.2 to estimate the manual control behavior that the vehicle driver 4 is expected to perform in the section ahead. Additionally or alternatively, the prediction module 2.2 can also estimate the expected driver state. The prediction module 2.2 compares the expected manual control behavior, the expected selection of the automated driving mode, and / or the expected driver state with the expected driving situation in the section ahead to estimate whether manual vehicle control is likely, desirable, or safer, or whether at least partially automated vehicle control is preferred. To perform these abstracted method steps, the prediction module 2.2 includes, for each driver profile, a machine learning model 3, shown in FIG. 2, that is individually trained on the driver profile. The machine learning model 3 receives as input data at least the vehicle data D-FZG, the surrounding data D-UMG, and / or the environmental data D-UMW. The machine learning model 3 provides, as an output variable, a prediction indicator value, which is a simple numerical value. This can be a natural number, an integer, a rational number, or a rounded irrational number. The prediction indicator value can be assigned to a range of allowable values, such as a value range from -1 to 1 or from 0 to 1.The predictive index value may be interpreted as a value representative of the vehicle driver's 4 affinity for controlling the vehicle 1 either manually or at least partially automated by a driver assistance system.

[0048] The machine learning model 3, which is specific to the driver profile, is initially trained by the vehicle manufacturer. For this purpose, specified driving situations are driven by various people with different driving styles, and then it is checked for each driving situation whether manual vehicle control or at least partially automated vehicle control is desirable. The machine learning model 3 includes, in particular, an artificial neural network, whereby the individual neurons of the artificial neural network form corresponding links through a learning process, so that a predictive index value IND-PRA appropriate for each driving style and driving situation can be determined.

[0049] The predicted indicator value IND-PRA is then read by the recommendation module 2.3 and compared with the indicator threshold value IND-SW. For example, the value range of the predicted indicator value IND-PRA can be from −1 to 1, and the indicator threshold value IND-SW can be 0. In that case, if the predicted indicator value IND-PRA is in a first range compared to the indicator threshold value IND-SW, e.g., a range between 0 and 1, a recommendation is output for the vehicle 1 to take over manual vehicle control. On the other hand, if the predicted indicator value IND-PRA is in a second range compared to the indicator threshold value IND-SW, e.g., between −1 and 0, a recommendation is output for the vehicle 1 to take over at least partially automated vehicle control by the driver assistance system. If the predicted indicator value IND-PRA and the indicator threshold value IND-SW have the same value, for example, no recommendation for taking over vehicle control can be output, or a preset standard recommendation can be output.

[0050] Figure 2 shows a more precise diagram of the information pathways of the variables read and output by the machine learning model 3.

[0051] The recommendations output in the vehicle 1 can be haptic, acoustic and / or visual. The output is made on a corresponding output device 6, which can be for example an actuator, a speaker or a display device.

[0052] The driver monitoring module 2.4 can be used to determine the vehicle driver's 4 current control behavior D-SV-AKT and / or the current driver state D-ZUS-AKT. A correspondingly trained machine learning model 3, which determines the current indicator value IND-AKT from the current control behavior D-SV-AKT and / or the current driver state D-ZUS-AKT, can also be integrated into the driver monitoring module 2.4. These variables can be read by the prediction module 2.2 to further train the machine learning model 3 corresponding to the driver profile, thereby improving the predicted quality of the vehicle driver's 4's actual compatibility. Additionally or alternatively, the control device 7 can determine the vehicle driver's 4's implementation behavior, i.e., whether the vehicle driver 4 followed the recommendations output by the recommendation system 2, and transmit this information as feedback to the prediction module 2.2. The prediction module 2.2 can also use this information to further train the corresponding machine learning model 3.

[0053] Furthermore, in an alternative development, the driver monitoring module 2.4 can analyze the driving behavior of the vehicle driver 4 by detecting the current control behavior D-SV-AKT and / or the current driver state D-ZUS-AKT and thereby perform an initial assignment or reassignment of the vehicle driver 4 to a corresponding driver profile. The vehicle driver 4 can estimate which driver profile best suits him / her, and the driver monitoring module 2.4 can check this estimation and thereby select a driver profile that is more appropriate to the vehicle driver's 4's actual driving behavior. Through continuous further training of the machine learning model 3 and its personalized storage, the corresponding driver profile of the vehicle driver 4 is adjusted and adapted to the vehicle driver more accurately. This is also performed for all vehicles 1 of the vehicle fleet, so that multiple, particularly realistic, driver profiles can be defined and distributed to the individual vehicles 1 of the vehicle fleet. This gradually improves the predictive quality of the corresponding machine learning models 3 assigned to the various driver profiles, thereby greatly increasing the likelihood that the recommendations output in the vehicle 1 will actually be implemented by the respective vehicle driver 4.

Claims

1. A vehicle (1) with an at least partially automated driving mode, A recommendation system (2) comprising a data collection module (2.1), a prediction module (2.2) and a recommendation module (2.3), the data collection module (2.1) is configured to collect vehicle data (D-FZG), surrounding data (D-UMG) and / or environmental data (D-UMW), The prediction module (2.2) is configured to read one driver profile from a plurality of driver profiles, each of which includes a machine learning model (3) or a heuristic model individually trained for each driver profile, and the model is configured to read the vehicle data (D-FZG), surrounding data (D-UMG) and / or environmental data (D-UMW) for at least a section located ahead, and to output a prediction index value (IND-PRA) as an output variable, the prediction index value (IND-PRA) being a numerical value; The recommendation module (2.3) is configured to compare the predicted index value (IND-PRA) with an index threshold (IND-SW), and to present a recommendation by the vehicle (1) to the vehicle driver (4) to hand over from at least partially automated vehicle control to manual vehicle control by the vehicle driver (4) when the predicted index value (IND-PRA) is in a first range compared to the index threshold (IND-SW), and to present a recommendation by the vehicle (1) to the vehicle driver (4) to hand over the at least partially automated vehicle control by a driver assistance system when the predicted index value (IND-PRA) is in a second range compared to the index threshold (IND-SW).

2. The recommendation system (2) further comprises a driver monitoring module (2.4), which is configured to monitor the vehicle driver (4) using at least one sensor, to determine a current control behavior (D-SV-AKT) and / or a current driver state (D-ZUS-AKT) from sensor data generated by the sensor, and to assign the vehicle driver (4) to one of the driver profiles depending on the determined current control behavior (D-SV-AKT) and / or the current driver state (D-ZUS-AKT). characterized in that A vehicle (1) according to claim 1.

3. The driver monitoring module (2.4) is further configured to determine a current indicator value (IND-AKT) from the current control behavior (D-SV-AKT) and / or the current driver state (D-ZUS-AKT), and the prediction module (2.2) is further configured to read the current control behavior (D-SV-AKT), the current driver state (D-ZUS-AKT) and / or the current indicator value (IND-AKT) and thereby further optimize the machine learning model (3) and / or the heuristic model. characterized in that A vehicle (1) according to claim 2.

4. The prediction module (2.2) is further configured to further train the machine learning model (3) and / or the heuristic model taking into account a current implementation of the recommendation to take over vehicle control made by the recommendation module (2.3). characterized in that A vehicle (1) according to any one of claims 1 to 3.

5. The recommendation system (2) is configured to receive fleet data (D-FLO), the fleet data (D-FLO) including aggregated amounts of user-specific control behaviors and / or driver states of users of a fleet of vehicles; and The system is configured to derive a driver profile from the fleet data (D-FLO) and / or update an existing driver profile, and to assign similar control behaviors and / or driver states to the same driver profile within defined limits. characterized in that A vehicle (1) according to any one of claims 1 to 3.

6. The fleet data (D-FLO) further comprises implementation behavior by users of the fleet of vehicles of recommendations made by the recommendation module (2.3) of the fleet of vehicles, and the prediction module (2.2) is further configured to further optimize the machine learning model (3) and / or the heuristic model taking into account the implementation behavior of users of the driver profile assigned to the vehicle driver (4). characterized in that A vehicle (1) according to claim 5.

7. The machine learning model (3) includes a neural network and / or is initialized based on a heuristic function. characterized in that A vehicle (1) according to any one of claims 1 to 3.

8. The recommendation module (2.3) is further configured to adapt the level of the indicator threshold (IND-SW) depending on the driver profile, the vehicle data (D-FZG), the surrounding data (D-UMG) and / or the environmental data (D-UMW) read by the prediction module (2.2). characterized in that A vehicle (1) according to any one of claims 1 to 3.

9. 4. A method for outputting a recommendation to a vehicle driver (4) of a vehicle (1) according to any one of claims 1 to 3 for the vehicle driver (4) to take over manual vehicle control or for a driver assistance system to take over at least partially automated vehicle control, comprising: - collecting vehicle data (D-FZG), surrounding data (D-UMG) and / or environmental data (D-UMW) by a data collection module (2.1), - reading, by a prediction module (2.2), one driver profile from a plurality of driver profiles, each driver module comprising a machine learning model (3) or a heuristic model individually trained for each of said driver profiles, said model being configured to read said vehicle data (D-FZG), surrounding data (D-UMG) and / or environmental data (D-UMW) for at least the section located ahead and to output a prediction index value (IND-PRA) as an output variable, said prediction index value (IND-PRA) being a numerical value; - determining, by said prediction module (2.2), said prediction indicator value (IND-PRA) for said section located ahead; - comparing, by a recommendation module (2.3), the predicted indicator value (IND-PRA) with an indicator threshold (IND-SW) and, if the predicted indicator value (IND-PRA) is in a first range compared to the indicator threshold (IND-SW), presenting a recommendation by the vehicle (1) to the vehicle driver (4) to hand over at least partially automated vehicle control to manual vehicle control by the vehicle driver (4), and, if the predicted indicator value (IND-PRA) is in a second range compared to the indicator threshold (IND-SW), presenting a recommendation by the vehicle (1) to the vehicle driver to hand over at least partially automated vehicle control by a driver assistance system; A method comprising:

Images