Computer-implemented method, device and computer program for controlling one or more settings of a vehicle

Abstract

A computer-implemented method for controlling one or more settings of a vehicle (100), the method comprising: identifying (110) one or more occupants of a vehicle; determining (120) environmental indicators based on at least one of vehicle sensor data, vehicle actuator data, online data and time data, wherein the environmental indicators indicate an environmental context in which the vehicle and / or the one or more vehicle occupants are located;Determining (130) the one or more settings of the vehicle using an output of a machine learning model, wherein the machine learning model is trained to output setting values ​​for the one or more settings of the vehicle based on the environmental indicators, taking into account the one or more occupants, wherein the machine learning model is trained at least partially based on previous actions or feedback from the one or more occupants; and providing (140) a control signal to control the one or more settings of the vehicle based on the previously determined one or more settings of the vehicle;characterized in that, in the case of multiple occupants, separate settings are determined for each of the occupants and the interior of the vehicle is divided into at least two regions, wherein the control signal for controlling the one or more settings of the vehicle is provided in such a way that the settings for the respective region are provided according to one or more occupants who are located in the respective region of the vehicle.

Description

The present invention relates to a computer-implemented method, a device and a computer program for controlling one or more settings of a vehicle, as well as to a corresponding vehicle. Today's vehicles are generally personalized through direct user input, meaning a user selects their preferred settings via a user interface. In existing vehicles, it's possible for occupants to create a user profile in which various parameters of the vehicle environment are configured. For example, climate control, seat comfort, or ambient lighting can be saved in the profile. These profiles are stored in an app or on the key so they can be accessed when the respective occupant is driving. However, each user must configure the settings themselves, and complete individualization is difficult to achieve. To improve familiarization, presets can be selected that are suitable for the majority of users, but this lacks the individuality of each user, such as their personal preference for a comfortable temperature.This also requires a large number of user inputs. In the future, cars will be equipped with machine learning to collect data that will make them smarter. But even then, cars will not be individualized products; they will be delivered with a specific set of functions already present from the factory. Patent applications EP 3 751 465 A1 and US 2015 / 0158486 A1 disclose concepts for imitating a user's driving style in autonomous driving or for adapting a generic machine learning model for autonomous driving to local conditions. This can be achieved either by adapting the training of the generic model based on the driver's driving behavior, or by training a separate model based on the driver's driving behavior, which then acts as a "filter" on the output of the generic model. DE 10 2020 115 726 A1 demonstrates automated differentiation and automatic learning of vehicle profiles. There is a need to provide an improved concept that better adapts a vehicle to the needs of individual occupants. This need is met by the subject matter of the independent claims. The present invention is based on the understanding that vehicles can be highly personal spaces where drivers and passengers may spend a considerable amount of time. Over time, a certain routine develops, corresponding to the preferences of the driver or other vehicle occupants. The preferences and routines of different occupants can vary considerably – for example, the perception of warmth often differs between male and female occupants, and if there is a small child on board, their needs must also be taken into account. Similarly, preferred suspension firmness and / or seat settings can differ depending on the occupant. The present invention utilizes occupant recognition to identify the vehicle's occupant(s). A machine learning model is trained to select preferred settings for one or more occupants based on environmental indicators.Based on the output of the machine learning model, the settings are then applied. This allows the vehicle's settings to be automatically adjusted situationally and tailored to the specific occupant(s), creating a concept that better adapts the vehicle to the needs of individual occupants. This automatic adjustment of settings also reduces the driver's workload, thus increasing driving safety. Several aspects of this disclosure relate to a computer-implemented method for controlling one or more vehicle settings. The method includes identifying one or more vehicle occupants. It further includes determining environmental indicators based on at least one of the vehicle's sensor data, vehicle actuator data, online data, and time data. The environmental indicators specify the environmental context in which the vehicle and / or the one or more vehicle occupants are located. The method also includes determining the vehicle's one or more settings using the output of a machine learning model. The machine learning model is trained to output setting values ​​for the vehicle's one or more settings based on the environmental indicators and taking into account the one or more occupants.The machine learning model is at least partially trained based on previous actions or feedback from one or more occupants. The process further includes providing a control signal to manage one or more vehicle settings based on previously defined settings. Thus, the vehicle's settings are automatically adjusted situationally and tailored to the specific occupant(s), creating a system that better adapts the vehicle to the needs of individual occupants. This automatic adjustment of settings also reduces the driver's workload, thereby increasing driving safety. When there are multiple occupants, individual settings are determined for each of them. This allows the needs of different occupants to be met separately, even when there are several people in the vehicle. This can be achieved, for example, by applying the machine learning model separately to the environmental indicators for each occupant in a vehicle with multiple occupants. The term "machine learning model" is not limited to a single model. For instance, the machine learning model could be a collection of machine learning models, with a separate model for each occupant. The separately calculated settings can now also be applied separately to multiple occupants. The vehicle's interior is divided into at least two regions. The control signal for managing the vehicle's one or more settings is provided in such a way that the settings for each region are made available to the one or more occupants located in that region. This allows the needs of different occupants to be met separately, potentially without affecting the other occupants. In an unused configuration, the one or more settings can be chosen in such a way that, in the case of multiple occupants, one or more common settings are determined for all of them. In other words, a consensus can be reached that is equally practical for the different occupants. In an unclaimed embodiment, for example, the machine learning model can be applied separately to the environmental indicators for each occupant. The common one or more settings can then be combined from the respective outputs of the machine learning model based on one or more joining criteria. In other words, separate machine learning models can again be used for the different occupants. Determining the compromise can then be done subsequently based on the outputs of the machine learning model(s). Alternatively, in an unused configuration, a single machine learning model trained on a combination of occupants can be used. Particularly in a family setting, a limited number of different combinations are possible, allowing the preferences of individual occupants as well as combinations of occupants to be determined. The machine learning model can be trained to output setting values ​​for one or more vehicle settings based on environmental indicators and taking into account a combination of the multiple occupants. The present disclosure uses a machine learning model that maps the preferences of the respective inmates. Several approaches are possible for training the model on the inmates. In one approach, the machine learning model can comprise two sub-models. The first sub-model can be trained in a person-independent manner, and the second sub-model can be trained based on the previous actions or feedback of one or more inmates. In other words, the first sub-model can already be trained to determine "reasonable" attitudes. The second sub-model can then map the preferences of the inmates. Alternatively, an initially person-unspecific model can be further trained to reflect the preferences of the inmates. For example, the machine learning model can be based on a person-unspecific model that has been further trained based on the previous actions or feedback of one or more inmates. In some examples, the process involves training the machine learning model, or a sub-model of the machine learning model, based on the previous actions or feedback of one or more occupants. This allows the adaptation to the previous actions and feedback to take place. For training purposes, appropriate training data can be collected. The process can further include changing a vehicle setting by providing the control signal, receiving feedback from a vehicle occupant in response to the setting change, and storing this feedback as information about a previous action or response from the occupant. This information about the previous action or response from the occupant can then be used to train the machine learning model. Many methods are conceivable for capturing feedback. For example, the process could involve analyzing image data to obtain feedback. This could, for instance, record whether the respective occupant feels comfortable or not after the change. Alternatively or additionally, the process could involve analyzing biometric sensor data to obtain feedback. Similar to image analysis, heart rate data, for example, could be used to determine whether the respective occupant feels comfortable or not after the change. Alternatively or additionally, the occupant(s) could be asked whether they approve of the respective change. For example, the process could involve providing output via a text-to-speech interface and receiving feedback as a response to the output via speech recognition.Alternatively or additionally, the procedure can include providing an output via a screen of the vehicle and receiving feedback as a response to the output via an input device of the vehicle. For example, one or more vehicle settings can be changed by one or more vehicle occupants via a user interface. Information about previous actions can also be collected. The one or more vehicle settings can also relate to functions of the vehicle's interior. This is the focus of the present invention. For example, the one or more vehicle settings can include at least one of the following: air conditioning settings, seat settings, lighting settings, windshield wiper settings, and chassis settings. Various aspects of the present disclosure relate to a device comprising one or more processors and one or more storage devices, configured to carry out the previously presented method. Several aspects of the present disclosure relate to a program with program code for the previously presented method, when the program code is executed on a computer, a processor, a control module or a programmable hardware component. Some examples of devices and / or methods are explained in more detail below with reference to the accompanying figures. Figures 1a and 1b show flowcharts of examples of a computer-implemented method for controlling one or more vehicle settings; Figure 1c shows a block diagram of an example of a device for controlling one or more vehicle settings; Figure 1d shows a schematic drawing of an example of a vehicle; Figures 2a and 2b show a concept for personalizing vehicle settings; and Figure 3 shows a schematic example of an overview of various components of a system that can be used for personalizing settings. Some examples are now described in more detail with reference to the accompanying figures. However, other possible examples are not limited to the features of these detailed embodiments. These may include modifications of the features, as well as equivalents and alternatives to the features. Furthermore, the terminology used herein to describe certain examples should not be considered restrictive for other possible examples. Identical or similar reference symbols throughout the description of the figures refer to identical or similar elements or features, which may be implemented in an identical or modified form, while providing the same or a similar function. Furthermore, the thickness of lines, layers, and / or areas in the figures may be exaggerated for clarity. When two elements A and B are combined using "or," this is to be understood as revealing all possible combinations, i.e., only A, only B, and A and B, unless explicitly defined otherwise in a specific case. As an alternative formulation for the same combinations, "at least one of A and B" or "A and / or B" can be used. This applies equivalently to combinations of more than two elements. When a singular form, e.g., "ein, eine" and "der, die, das," is used, and the use of only a single element is neither explicitly nor implicitly defined as mandatory, further examples may also use multiple elements to implement the same function. If a function is subsequently described as being implemented using multiple elements, further examples may implement the same function using a single element or a single processing entity.It is further understood that the terms "include", "comprehensive", "exhibit" and / or "exhibit" when used describe the presence of the specified features, integers, steps, operations, processes, elements, components and / or a group thereof, but do not exclude the presence or addition of one or more other features, integers, steps, operations, processes, elements, components and / or a group thereof. Figures 1a and 1b show flowcharts of examples of a computer-implemented method for controlling one or more settings of a vehicle 100 (shown in Figure 1d). The method includes identifying 110 one or more occupants of a vehicle. The method further includes determining 120 environmental indicators based on at least one of the vehicle's sensor data, vehicle actuator data, online data, and time data. The environmental indicators indicate an environmental context in which the vehicle and / or the one or more vehicle occupants are located. The method further includes determining 130 the one or more settings of the vehicle using an output from a machine learning model. The machine learning model is trained to output setting values ​​for the one or more settings of the vehicle based on the environmental indicators, taking into account the one or more occupants.The machine learning model is trained, at least in part, based on previous actions or feedback from one or more occupants. The method further comprises providing a control signal 140 to control one or more vehicle settings based on the previously determined one or more vehicle settings. The method is executed by the vehicle 100, for example, by a corresponding device 10 for controlling the one or more vehicle settings, as shown in Fig. 1c and Fig. 1d. Fig. 1c shows a block diagram of an example of the device 10 for controlling one or more settings of the vehicle 100. The device 10 comprises one or more processors 14 and one or more storage devices 16, which are coupled to the one or more processors 14. Optionally, the device further comprises at least one interface 12, which is also coupled to the one or more processors 14. The one or more processors are fundamentally configured to provide the functionality of the device by means of the at least one interface (for communication and data exchange with the sensors 102, the actuators 104, and / or a remote counterpart via a wireless communication network) and / or by means of the one or more storage devices (for storing and retrieving data). The device is configured to perform the method of Fig. 1a and / or 1b.The process can, in principle, be executed by one or more processors, using at least one interface for communication and one or more storage devices for storing and retrieving data. This device is generally part of a vehicle. Fig. 1d shows a schematic drawing of an example of a vehicle 10, which includes the device 10 as well as the sensors 102 and / or actuators 104. The following describes the features of the method, the corresponding device, a corresponding computer program, and the vehicle in relation to the method. Features described in connection with the method can also be incorporated into the corresponding device, computer program, and vehicle. Several aspects of this disclosure relate to a computer-implemented method for controlling one or more vehicle settings, as well as to a corresponding device and computer program. The method, device, and computer program can be used to adjust (i.e., change) the one or more settings based on the following criteria. Not all vehicle settings are equally suitable for this concept. In particular, this concept is suitable for one or more settings that can be changed by the vehicle's one or more occupants via a user interface (of the vehicle), i.e., settings that can be changed by the users themselves without requiring any modification to the vehicle's hardware or software.A possible restriction to such settings also has the advantage that the actions of the occupants, i.e., the changes to the settings made by the occupants, can be recorded and used to train the machine learning model. Consequently, the method can include recording changes to the settings made by the occupants, along with corresponding environmental indicators. For example, the one or more vehicle settings could relate to functions of the vehicle's interior, such as comfort features. These could include, for example, one or more settings such as the vehicle's climate control, seat settings, lighting, or windshield wiper settings.However, the vehicle's suspension settings can also be considered settings, such as switching between a comfort and a sport suspension setting. Other settings are also conceivable. The proposed concept begins with identifying the one or more occupants of a vehicle. This is done for two reasons: firstly, it allows the appropriate settings to be determined, and secondly, the actions and feedback of the occupants can be attributed to the correct occupants to enable further training of the machine learning model. Occupants can be identified using various techniques, such as radio signals from a mobile device belonging to the respective occupant, identification of a vehicle key used by a specific occupant, determination of the occupants' weight, analysis of camera data (for visual occupant recognition), or analysis of biometric data (e.g., based on heart rhythm or voice analysis).In addition to the identity of the occupants, the position of the occupants within the vehicle can also be determined, i.e., which occupant is sitting in which seat of the vehicle. The machine learning model has two sets of inputs: the occupants and the environmental indicators. The environmental indicators are derived from at least one of the vehicle's sensor data, actuator data, online data, and time data, and represent the environmental context in which the vehicle and / or its one or more occupants are located. In other words, the environmental indicators provide the machine learning model with a way to adapt its settings to the environmental context. The environmental context can also be viewed as a feature vector for mapping the sensor data, actuator data, online data, and / or time data to the machine learning model. In some examples, the process also includes converting the sensor data, actuator data, online data, and / or time data into a standardized format, allowing the machine learning model to be used in other vehicles. Environmental indicators can also be used as triggers for changing vehicle settings. Depending on the environmental context, a change or re-evaluation of one or more settings is triggered. For example, vehicle brightness sensor data or time data can indicate that it is dark outside (and therefore also inside) the vehicle. This can trigger an adjustment of the vehicle's exterior and / or interior lighting. Similarly, heart rate sensor data can indicate that an occupant (such as a small child) is asleep. This can also trigger an adjustment of the vehicle's interior lighting. Activation of the windshield wipers can trigger an adjustment of the infotainment system's volume or the wiper rate according to the driver's preferences.Activating the side windows can also trigger an adjustment of the infotainment system's volume. Receiving an internet notification about a traffic jam or the availability of charging stations can trigger a re-evaluation of the route plan. Therefore, the process can further involve determining a trigger to change one or more settings based on environmental indicators. Trigger determination can also be based on machine learning, whereby another machine learning model, or the existing machine learning model, can be trained to learn a relationship between a change in an environmental indicator and a setting change by an occupant (for example, through unsupervised learning to determine commonalities in input data).Alternatively or additionally, when an environmental indicator changes, a trigger can be determined using a database or list of common triggers. Therefore, the environmental indicators reflect, on the one hand, the situation in which the vehicle and / or the one or more occupants find themselves, and on the other hand, they can be used as a trigger for determining one or more settings. The environmental indicators (including, for example, an indicator for the specific trigger) are now used as input for the machine learning model. The identity of the one or more occupants also influences the machine learning model. In some examples, the identity of the one or more occupants (in coded form) can also be used as an input feature for the machine learning model. Alternatively, the machine learning model corresponds to a collection of machine learning models, where each occupant (and optionally each combination of occupants) is assigned a machine learning model from the collection, selected based on the identity of the one or more occupants. The machine learning model is trained to output setting values ​​for one or more vehicle settings based on environmental indicators and taking into account the one or more occupants. The machine learning model is at least partially trained based on previous actions or feedback from the one or more occupants. The following section first briefly outlines the fundamentals of machine learning and different learning approaches. It then demonstrates how the present machine learning model can be designed and trained. Machine learning refers to algorithms and statistical models that computer systems can use to perform a specific task without the use of explicit instructions. For example, instead of a rule-based transformation of data, machine learning can use a data transformation derived from an analysis of historical and / or training data. For instance, the content of images can be analyzed using a machine learning model or a machine learning algorithm.To enable the machine learning model to analyze the content of an image, it can be trained using training images as input and information about the image content as output. By training the machine learning model with a large number of training images and / or training sequences (e.g., words or sentences) and associated information about the image content (e.g., labels or annotations), the machine learning model "learns" to recognize the image content. This allows the machine learning model to recognize the content of images not included in the training data.The same principle can be used for other types of sensor data: By training a machine learning model using training sensor data and a desired output, the machine learning model "learns" a conversion between the sensor data and the output. This can then be used to provide an output based on non-training sensor data supplied to the machine learning model. The supplied data (e.g., sensor data, metadata, and / or image data) can be preprocessed to obtain a feature vector, which is used as input for the machine learning model. In this case, the environmental indicators, such as a version of the feature vectors converted into a standardized format, will be used as the feature vector. Machine learning models can be trained using training input data. The examples above use a training method called supervised learning. In supervised learning, the machine learning model is trained using a plurality of training samples (data sets), where each sample can include a plurality of input data values ​​and a plurality of desired output values. By providing both input data values ​​and desired output values, the machine learning model "learns" which output value to provide based on an input data value that is similar to the input data values ​​provided during training. In addition to supervised learning, semi-supervised learning can also be used. In semi-supervised learning, some of the training samples lack a desired output value. Supervised learning can be based on a supervised learning algorithm (e.g., ...).a classification algorithm, a regression algorithm, or a similarity learning algorithm). Classification algorithms can be used when the outputs are restricted to a limited set of values ​​(categorical variables), i.e., the input is classified as one from the limited set of values. Regression algorithms can be used when the outputs are any numerical value (within a range). Similarity learning algorithms can be similar to both classification and regression algorithms, but they are based on learning from examples using a similarity function that measures how similar or related two objects are. In addition to supervised or semi-supervised learning, unsupervised learning can be used to train the machine learning model.In unsupervised learning, only input data may be provided, and an unsupervised learning algorithm can be used to find a structure in the input data (e.g., by grouping or clustering the input data, finding commonalities in the data). Clustering is the assignment of input data, which comprises a plurality of input values, into subsets (clusters) such that input values ​​within the same cluster are similar according to one or more (predefined) similarity criteria, while they are dissimilar to input values ​​included in other clusters. Reinforcement learning is a third group of machine learning algorithms. In other words, reinforcement learning can be used to train the machine learning model. In reinforcement learning, one or more software actors (so-called "software agents") are trained to perform actions in an environment. Based on the actions performed, a reward is calculated using a reward function. Reinforcement learning relies on training the one or more software agents to select actions in such a way that the cumulative reward increases, resulting in software agents that become better at the task they are given (as evidenced by increasing rewards). In the present concept, the machine learning model can be trained using supervised learning and / or reinforcement learning. Accordingly, the procedure can involve training the machine learning model or a submodel of the machine learning model based on the previous actions or feedback of one or more occupants. If the machine learning model is trained using supervised learning, a desired output value can be generated based on the previous actions or feedback of one or more occupants, and the feature vector determined based on environmental indicators can be used as training input data. If a machine learning model is used for multiple occupants, the identified characteristics of one or more occupants can also be used as additional features in the training input data. For example, when a vehicle setting is changed (such as changes that are not reversed within a short time), a snapshot of the environmental indicators can be saved as a feature vector for the training input data, and the changed setting can be stored as the desired output value. Once a relevant amount of training input data has been collected, training of the machine learning model can begin.Over time, the training can be continued based on subsequently collected training input data and desired output values. If the machine learning model is trained using reinforcement learning, the previous actions or feedback of one or more inmates can be used to define the reward function. The training concepts described above can each be applied to the approach in which the occupant's identity is used as input for the machine learning model, as well as to the approach in which a separate machine learning model is used for each occupant. Only the selection of the training data to be used can be adapted to the respective occupant. The training data can also be collected specifically by recording and evaluating the reaction of one or more occupants to changes in one or more settings. The procedure can further include changing a vehicle setting by providing the control signal, receiving feedback from a vehicle occupant in response to the setting change, and storing this feedback as information about a previous action or response from the occupant. For example, it can be determined whether the change in settings is perceived positively or negatively by the one or more occupants, which is then considered feedback from them. To obtain this feedback, the procedure can include evaluating image data (such as a facial expression of the one or more occupants).Happy or relaxed facial expressions can be interpreted as positive or neutral feedback, while tense or angry facial expressions can be interpreted as negative feedback. Alternatively or additionally, the method can include the evaluation of biometric sensor data (such as heart rate data or voice recordings) to obtain feedback. A decrease in heart rate can be interpreted as positive feedback, while an increase can be interpreted as negative feedback (in the case of an angry facial expression) or positive feedback (in the case of a pleased facial expression). Speech recognition can be used to distinguish between a pleased (positive feedback) and an angry (negative feedback) tone of voice. Alternatively or additionally, the one or more occupants can be asked directly whether they approve of a change to a setting. The procedure can involve providing output via a text-to-speech interface and receiving feedback in response to the output via speech recognition. A vehicle screen can also be used. The procedure can involve providing output via a vehicle screen and receiving feedback in response to the output via an input device, such as a touchscreen, in the vehicle. In general, it can be impractical to train a machine learning model from scratch based on the data collected in the vehicle. Therefore, a pre-trained, generic machine learning model provided by the vehicle manufacturer can be used as a foundation. This model can be trained to select one or more settings based on environmental indicators that are perceived as appropriate by a majority of different occupants. It can also be trained to detect triggers in the environmental indicators. This pre-trained model can then be used as the basis for the machine learning model. Two approaches are possible here. In the first approach, the pre-trained, generic machine learning model can be used as one of two models that are linked together. The machine learning model could, for example, comprise two sub-models.The first submodel can be trained in a person-independent (i.e., generic) manner, and the second submodel can be trained based on the previous actions or feedback of one or more inmates. Such an approach is shown, for example, in Figures 2a to 3, where the pre-trained, generic machine learning model is referred to as the standard model and the inmate-based model as the behavioral model. Alternatively, the machine learning model can be based on a person-independent model that has been further trained based on the previous actions or feedback of one or more inmates. Therefore, training the machine learning model can be equivalent to further training the pre-trained, generic machine learning model. The trained machine learning model can be used in various ways. In some examples, the machine learning model can be used to determine individual settings for each occupant. In other words, if there are multiple occupants, individual settings can be determined for each one. This can be achieved by applying the machine learning model separately to the environmental indicators for each occupant. This can be done either by using the occupant's identity as an input for the machine learning model, or by using a separate machine learning model for each occupant. Alternatively or additionally, common settings can be defined directly to best suit all occupants. For example, one or more settings can be chosen so that, in the case of multiple occupants, common settings are defined for all of them. Here, two approaches are possible. In the first approach, the machine learning model can be applied separately to the environmental indicators for each occupant. The common settings can then be combined based on one or more merging criteria derived from the respective outputs of the machine learning model. These merging criteria can, for example, specify whether a minimum, average, or maximum value should be used for different setting values ​​(such as for interior temperature or noise level).In a second approach, the machine learning model can be trained from the outset to output setting values ​​for one or more vehicle settings based on environmental indicators and taking into account a combination of the multiple occupants. This can be achieved either by using the occupants' identities as input for the machine learning model, or by using a machine learning model from a collection of machine learning models that is trained on a specific combination of occupants. Hybrid forms are also possible. For example, some settings can be defined as common settings (such as settings that must be set the same for every occupant), while other settings can be defined separately for each occupant (or for subgroups of occupants). The machine learning model can, for example, be executed and / or trained on at least one of the one or more processors or on an acceleration unit for the execution or training of machine learning models. The method further includes providing the control signal 140 to control one or more vehicle settings based on the previously determined one or more vehicle settings. The control signal can, for example, be provided to change the one or more settings. The control signal can, for example, be provided to an actuator 104. Some settings can be determined separately for each occupant. This is particularly true when these settings can be adjusted individually for each occupant. Examples include seat settings, the volume or tone of the infotainment system (within limits), ventilation, and interior lighting. For instance, the vehicle's interior can be divided into at least two regions, such as front and rear, first, second, and third row, or front left, front right, rear left, and rear right. The control signal for managing one or more vehicle settings can be provided in such a way that the settings for each region are made available to one or more occupants located in that region. The previously determined position of the occupants within the vehicle can then be used for this purpose. Machine learning algorithms are typically based on a machine learning model. In other words, the term "machine learning algorithm" can refer to a set of instructions that can be used to create, train, or use a machine learning model. The term "machine learning model" can refer to a data structure and / or a set of rules that represents the learned knowledge (e.g., based on the training performed by the machine learning algorithm). In some examples, the use of a machine learning algorithm may imply the use of an underlying machine learning model (or multiple underlying machine learning models). The use of a machine learning model may also imply that the machine learning model and / or the data structure / set of rules that constitutes the machine learning model are trained by a machine learning algorithm. For example, the machine learning model could be an artificial neural network (ANN). ANNs are systems inspired by biological neural networks, such as those found in a retina or brain. ANNs comprise a plurality of interconnected nodes and a plurality of connections, called edges, between the nodes. There are typically three types of nodes: input nodes, which receive input values; hidden nodes, which are connected (only) to other nodes; and output nodes, which provide output values. Each node can represent an artificial neuron. Each edge can transmit information from one node to another. The output of a node can be defined as a (nonlinear) function of its inputs (e.g., the sum of its inputs). The inputs of a node can be used in the function based on a "weight" assigned to the edge or the node providing the input.The weight of nodes and / or edges can be adjusted during the learning process. In other words, training an artificial neural network can involve adjusting the weights of the nodes and / or edges of the artificial neural network, i.e., to achieve a desired output for a given input. Alternatively, the machine learning model can be a support vector machine, a random forest model, or a gradient boosting model. Support vector machines (i.e., support vector networks) are supervised learning models with associated learning algorithms that can be used to analyze data (e.g., in classification or regression analysis). Support vector machines can be trained by providing an input with a plurality of training input values ​​belonging to one of two categories. The support vector machine can be trained to assign a new input value to one of the two categories. Alternatively, the machine learning model can be a Bayesian network, which is a probabilistic directed acyclic graphical model. A Bayesian network can represent a set of random variables and their conditional dependencies using a directed acyclic graph.Alternatively, the machine learning model can be based on a genetic algorithm, which is a search algorithm and heuristic technique that imitates the process of natural selection. The at least one interface 12 can, for example, correspond to one or more inputs and / or one or more outputs for receiving and / or transmitting information, such as digital bit values, based on a code, within a module, between modules, or between modules of different entities. The at least one interface 12 can, for example, comprise an interface circuit arrangement configured to receive input data and provide output data. The one or more processors 14 can correspond to any controller, processor, or programmable hardware component. For example, the functionality of the one or more processors 14 can also be implemented as software programmed for a corresponding hardware component. In this respect, the one or more processors 14 can be implemented as programmable hardware with appropriately adapted software. Any processor, such as digital signal processors (DSPs), can be used. The implementation examples are not limited to a specific type of processor. Any processor, or even multiple processors, are conceivable for implementation. The one or more storage devices 16 can, for example, include at least one element of the group consisting of computer-readable storage medium, magnetic storage medium, optical storage medium, hard disk, flash memory, floppy disk, random access memory, programmable read only memory (PROM), erasable programmable read only memory (EPROM), electronically erasable programmable read only memory (EEPROM), and network storage. Vehicle 100 can, for example, correspond to a land vehicle, a watercraft, an aircraft, a rail vehicle, a road vehicle, a car, an off-road vehicle, a motor vehicle, or a truck. Further details and aspects of the method, apparatus, computer program, and vehicle are mentioned in connection with the concept or examples described before or after (e.g., Figures 2a to 3). The method, apparatus, computer program, and vehicle may include one or more additional optional features corresponding to one or more aspects of the proposed concept or the examples described before or after. Several aspects of the present disclosure relate to a behavioral model trained using deep learning, such as reinforcement learning, and suitable for autonomous vehicles. To enhance the occupant experience and foster a stronger connection with the vehicle, a deep learning-based behavioral model (such as a machine learning model) can be integrated into the car, complementing existing machine learning models used for tasks like vehicle control. Personalization is a major focus for occupants, suggesting that fully personalized vehicles are likely the future. Such a vehicle can adapt to the user in any situation and understands their habits. For example, every car can be delivered from the factory with a generic machine learning model that is suitable for a large number of people but doesn't cater to the personal preferences of each individual. A behavioral model of the vehicle, going beyond the standard machine learning model, can be introduced in a neural processing unit within the vehicle. This behavioral model can include a constantly updated, unique, and personalized combination of features to adapt to all occupants driving the vehicle. Therefore, the vehicle can be uniquely adapted to its occupants. This can be achieved, for example, through deep learning, such as supervised learning or reinforcement learning, on a deep neural network (i.e., a neural network with at least one hidden layer) of the environment.The environment is captured by the available sensors in the car and provided to the machine learning model as environmental indicators. User reactions can also be tracked, allowing the level of satisfaction to be determined and fed into the behavioral model to personalize it for each user. Several examples are thus based on adding this extra layer to the behavioral model, creating a unique, individualized model for the vehicle's occupants. Figures 2a and 2b show a concept for personalizing vehicle settings. Personalization 210 is achieved by overlaying a behavioral model 240 onto an existing model 230 of an autonomous vehicle with factory settings. The behavioral model is operated based on environmental factors 230, such as driving behavior, fuel level, air conditioning, navigation, parking, infotainment, cleaning, number of drivers, etc. That is, one or more of the aforementioned environmental factors / indicators can be provided as input features to the behavioral model 240 (and to the existing model 220), and an output of the machine learning model can be used to control one or more vehicle settings. Figure 2b shows an approach for training the behavioral model 240. Here, the previously mentioned environmental factors / indicators 230 are used as training data for deep reinforcement learning-based training 230 to train the behavioral model 240. Alternatively, other training methods, such as supervised learning, can also be used. The behavioral model is trained to provide 250 unique functionality for the inmates. Fig. 3 shows a schematic example of an overview of various components of a system that can be used to personalize settings. The user and the vehicle's environment generate an input 300 for the sensor network 310 in the vehicle, which generates environmental information (such as environmental indicators) for the neural unit 320. The sensor network can include, for example, a heart rate sensor 312, a radar sensor 314, a GPS (Global Positioning System) sensor 316, and / or other sensors 318. The neural unit passes the data to a deep learning model (the machine learning model), such as a reinforcement learning-based or supervised learning-based model, the so-called behavioral model 324, which begins to learn the occupants' habits. To ensure that no laws are broken by learning bad habits, the standard model, i.e.,In a generic submodel 322, a predefined parameter space is defined. The behavioral model can monitor the standard model and decide how the standard model should interact with the actuators in the vehicle to control and manage the human-machine interfaces. The neural unit 320, such as the standard model 322 executed by the neural unit, interacts with one or more actuators 330, such as a climate control unit 332, an accelerator pedal unit 334 (to make settings related to the accelerator pedal), an infotainment system 336, or other actuators 338. One example could be the climate control: If the driver is alone in the car, the model would set the temperature to 18°C. If the whole family, including a small child, is traveling with them, the temperature will be set to a higher value that corresponds to the consensus of all passengers. The behavioral model can be trained in such a way that it can also be transferred to other vehicles. For example, the environmental indicators can be converted into a standardized format before being used for training and executing the behavioral model. The proposed concept is applicable to vehicles with a neural unit, i.e., vehicles capable of running machine learning models, for example, on a processor or accelerator for neural applications. More generally, the proposed concept can be applied to any adaptable device with neural units. Further details and aspects of the behavioral model are mentioned in connection with the concept or examples previously described (e.g., Figs. 1a to 1d). The behavioral model may include one or more additional optional features that correspond to one or more aspects of the proposed concept or the described examples, as presented before or after. The aspects and features described in connection with one of the previous examples can also be combined with one or more of the further examples to replace an identical or similar feature of that further example or to additionally introduce the feature into the further example. Examples may also include a (computer) program with program code for executing one or more of the above procedures, or refer to such a program when executed on a computer, processor, or other programmable hardware component. Steps, operations, or processes of various procedures described above may therefore also be executed by programmed computers, processors, or other programmable hardware components. Examples may also include program storage devices, such as digital data storage media, that are machine-, processor-, or computer-readable and encode or contain machine-executable, processor-executable, or computer-executable programs and instructions. The program storage devices may, for example,Digital storage devices include or may include magnetic storage media such as magnetic disks and magnetic tapes, hard disk drives, or optically readable digital data storage media. Further examples may also include computers, processors, control units, (field) programmable logic arrays ((F)PLAs), (field) programmable gate arrays ((F)PGAs), graphics processing units (GPUs), application-specific integrated circuits (ASICs), integrated circuits (ICs), or system-on-a-chip (SoCs) programmed to perform the steps of the procedures described above. It is further understood that the disclosure of several steps, processes, operations, or functions disclosed in the description or claims should not be interpreted as necessarily occurring in the described sequence, unless explicitly stated in a specific case or required for technical reasons. Therefore, the preceding description does not restrict the execution of multiple steps or functions to a specific sequence. Furthermore, in other examples, a single step, function, process, or operation may include and / or be broken down into multiple sub-steps, functions, processes, or operations. If certain aspects described in the preceding sections relate to a device or system, these aspects should also be understood as a description of the corresponding procedure. For example, a block, device, or functional aspect of the device or system may correspond to a feature, such as a process step, of the corresponding procedure. Similarly, aspects described in relation to a procedure should also be understood as a description of a corresponding block, element, property, or functional feature of that device or system. The following claims are hereby included in the detailed description, each claim being a separate example. It should also be noted that—although a dependent claim may refer to a specific combination with one or more other claims—other examples may include a combination of the dependent claim with the subject matter of any other dependent or independent claim. Such combinations are hereby explicitly proposed unless it is stated in a specific case that a particular combination is not intended. Furthermore, features of a claim are also to be included for each other independent claim, even if that claim is not directly defined as dependent on that other independent claim. Reference symbol list 10 Device 12 Interface 14 Processor 16 Storage Device 100 Vehicle 110 Identifying one or more occupants of a vehicle 120 Determining environmental indicators 130 Determining one or more vehicle settings 140 Providing a control signal 150 Receiving feedback 152 Evaluating image data 154 Evaluating biometric sensor data 156 Providing output via a text-to-speech interface 158 Providing output via a screen 160 Training a machine learning model 210 Personalization 220 Predefined model 230 Environmental factors 235 Deep reinforcement learning 240 Behavioral model 250 Providing unique functionality 300 Input 310 Network of sensors 312 Heart rate sensor 314 Radar sensor 316 GPS 318 Other sensors 320 Neural unit 322 Standard model 324 Behavioral model 330 Actuators 332 Climate control 334 Accelerator pedal unit 336 Infotainment system 338 OtherThe actuator

Claims

A computer-implemented method for controlling one or more settings of a vehicle (100), the method comprising: identifying (110) one or more occupants of a vehicle; determining (120) environmental indicators based on at least one of vehicle sensor data, vehicle actuator data, online data and time data, wherein the environmental indicators indicate an environmental context in which the vehicle and / or the one or more vehicle occupants are located;Determining (130) the one or more settings of the vehicle using an output of a machine learning model, wherein the machine learning model is trained to output setting values ​​for the one or more settings of the vehicle based on the environmental indicators, taking into account the one or more occupants, wherein the machine learning model is trained at least partially based on previous actions or feedback from the one or more occupants; and providing (140) a control signal to control the one or more settings of the vehicle based on the previously determined one or more settings of the vehicle;characterized in that, in the case of multiple occupants, separate settings are determined for each of the occupants and the interior of the vehicle is divided into at least two regions, wherein the control signal for controlling the one or more settings of the vehicle is provided in such a way that the settings for the respective region are provided according to one or more occupants who are located in the respective region of the vehicle. The method according to claim 1, wherein, in the case of multiple occupants, the machine learning model is applied separately for each of the occupants to the environmental indicators. The method according to one of claims 1 or 2, wherein the machine learning model comprises two sub-models, wherein a first sub-model is trained in a person-independent manner and the second sub-model is trained based on the previous actions or feedback of one or more occupants. The method according to one of claims 1 or 2, wherein the machine learning model is based on a person-independent model that has been further trained based on the previous actions or feedback of one or more occupants. The method according to any one of claims 1 to 4, comprising training (170) the machine learning model or a sub-model of the machine learning model based on the previous actions or feedback of one or more occupants. The method according to any one of claims 1 to 5, further comprising changing a setting of the vehicle by providing the control signal, receiving (150) feedback from an occupant of the vehicle in response to the change of the setting, and storing (160) the feedback as information about a previous action or feedback from the occupant. The method according to claim 6, wherein the method comprises evaluating (152) image data to obtain feedback, and / or wherein the method comprises evaluating (154) biometric sensor data to obtain feedback, and / or wherein the method comprises providing (156) an output via a text-to-speech interface and obtaining feedback as a response to the output via speech recognition, and / or wherein the method comprises providing (158) an output via a screen of the vehicle and obtaining feedback as a response to the output via an input device of the vehicle. The method according to any one of claims 1 to 7, wherein the one or more settings of the vehicle can be changed by the one or more occupants of the vehicle via a user interface, and / or wherein the one or more settings of the vehicle relate to functions of an interior of the vehicle, and / or wherein the one or more settings of the vehicle comprise at least one of a climate control setting of the vehicle, a seat setting of the vehicle, a lighting setting of the vehicle, a windshield wiper setting of the vehicle and a chassis setting of the vehicle. A device comprising one or more processors and one or more storage devices, configured to perform the method according to any one of claims 1 to 8. A program comprising program code for performing the method according to any one of claims 1 to 8, when the program code is executed on a computer, a processor, a control module or a programmable hardware component.

Images