Method for using at least one driver model during at least partially automated driving of a vehicle

Abstract

The invention relates to a method (100) for using at least one driver model during at least partially automated driving of a vehicle (1), the method comprising: - providing (101) driver goals, wherein the driver goals characterise human driving behaviour in road traffic; - generating (102) at least one driver model on the basis of the provided driver goals, wherein the at least one driver model is a graphical representation (11) in which nodes (12) of the graphical representation (11) represent the driver goals and edges (13) of the graphical representation (11) represent relationships between the driver goals; and - using (103) the at least one generated driver model during the at least partially automated driving of the vehicle (1) in order to take into account human driving behaviour in road traffic during the at least partially automated driving of the vehicle (1). The invention also relates to a computer program, to a device, and to a storage medium for this purpose.

Description

[0001] R.415561

[0002] - 1 -

[0003] Description

[0004] title

[0005] Method for using at least one driver model in at least partially automated driving of a vehicle

[0006] The invention relates to a method for using at least one driver model in at least partially automated driving of a vehicle. The invention further relates to a computer program, a device, and a storage medium for this purpose.

[0007] State of the art

[0008] A necessary characteristic of automated vehicles is the ability to understand the intentions of other human drivers and to react in a predictable and interpretable manner. This is achieved, for example, with the help of driver models. Driver models are often learned through imitation learning or reinforcement learning using strategies based on safety risks.

[0009] Motion planning is one of the most challenging tasks in automated driving. Existing motion planning methods are often based on imitation learning, where a strategy is learned to approximate the behavior of experts, or on reinforcement learning, which is based either on maximizing a reward or maximizing a strategy. Newer methods also include multi-agent reinforcement learning or parallel learning, which combines real-world data with simulated data. However, these methods typically neglect the human factors that determine agent behavior. R.415561

[0010] - 2-

[0011] Agents' preferences can vary between selfish, cooperative, and competitive, and these preferences can constantly change between agents. However, current technology assumes the same behavioral model for every agent and performs both individual and global strategy optimization.

[0012] Current state-of-the-art approaches are therefore unable to represent the real-world behavior of human drivers. Human driving behavior is determined by more factors than just social values ​​and safety risks, and these factors are also context-dependent.

[0013] Disclosure of the invention

[0014] The invention relates to a method with the features of claim 1, a computer program with the features of claim 10, a device with the features of claim 11, and a computer-readable storage medium with the features of claim 12. Further features and details of the invention will become apparent from the respective dependent claims, the description, and the drawings. Features and details described in connection with the method according to the invention naturally also apply in connection with the computer program, the device, and the computer-readable storage medium according to the invention, and vice versa, so that a reciprocal reference is always possible with regard to the disclosure of the invention.

[0015] The invention relates in particular to a method for using at least one driver model in at least partially automated driving of a vehicle, comprising:

[0016] Providing driver objectives, wherein the driver objectives characterize human driving behavior in road traffic, wherein the driver objectives describe or can characterize a human driver of the vehicle or of surrounding vehicles, R.415561

[0017] - 3-

[0018] Generating at least one driver model based on the provided driver goals, wherein the at least one driver model is a graphical representation, in particular a knowledge graph, where nodes of the graphical representation represent the driver goals and edges of the graphical representation represent relationships or dependencies between the driver goals.

[0019] Using at least one generated driver model in the at least partially automated driving of the vehicle, in order to take into account human driving behavior in road traffic.

[0020] At least partially automated driving can take the form of automated maneuvers such as adaptive cruise control (ACC) or fully automated driving in the sense of autonomous driving. By considering driver objectives and the relationships or dependencies between them, at least partially automated driving can be performed more precisely, as human driving behavior is taken into account. This allows for better prediction of the actions of other road users.

[0021] Another possibility is that the provisioning process includes at least one of the following steps:

[0022] Determine at least one of the driver's objectives based on an analysis of sensor data, wherein the sensor data results from the acquisition of at least one sensor, in particular at least one sensor of the vehicle, wherein the sensor data may be, for example, digital images such as video, radar, LiDAR, ultrasonic, motion and / or thermal images and may result from the acquisition of a corresponding sensor such as a camera, radar, LiDAR, ultrasonic, motion and / or thermal imaging sensor, wherein, for example, the movement of surrounding road users can be analyzed in order to infer the at least one driver objective of the corresponding surrounding road users.

[0023] Define at least one of the driver objectives based on a safety requirement, where the safety requirement may relate, for example, to the safety of the vehicle's occupants and / or other road users, R.415561

[0024] - 4-

[0025] Automated or manual provision or modification of at least one of the driver's objectives based on an individual preference or driving characteristic of a driver of the vehicle and / or a driver of at least one other vehicle in the vicinity of the vehicle.

[0026] This allows for a comprehensive understanding of driving behavior, as it draws on both observed data and individual preferences and safety concerns. This enables more precise predictions and improves the vehicle's adaptation to different driving styles.

[0027] Furthermore, it is advantageous if the driver objectives relate to the safety of other road users, traffic regulations, the vehicle's purpose, the driver's social values, driving ability, driving style, and / or attention span. This allows the generated driver models to consider various aspects of human driving behavior in road traffic situations, such as priorities regarding safety, adherence to traffic regulations, and individual driving styles. In particular, this enables a realistic representation of human driving behavior.

[0028] For example, driver objectives can be subdivided into primary driver objectives and subordinate sub-driver objectives. Alternatively or additionally, driver objectives can be characterized as hard and soft driver objectives, with the hard driver objectives being prioritized and must be fulfilled within the framework of at least partially automated driving, and the soft driver objectives only being considered if the hard driver objectives are met. This allows for different priorities in driving behavior to be taken into account. Hard objectives represent particularly essential requirements, such as those related to safety, and must be fulfilled with priority. Soft objectives, such as a driver's individual preferences regarding achieving a driving destination, can be considered only after the hard objectives have been met.

[0029] It is also conceivable that the procedure may further include: R.415561

[0030] - 5-

[0031] Providing or determining at least one piece of contextual information regarding the vehicle's environment, wherein the at least one piece of contextual information is selected from the vehicle's location, the condition of a road, the current time of day, weather conditions, and / or current traffic, and wherein the at least one piece of contextual information is further taken into account during at least partially automated driving. This allows the driving behavior of other road users to be predicted more realistically and accurately, and the vehicle's own driving to be adjusted accordingly during at least partially automated driving. By considering factors such as location, road condition, time of day, weather, and traffic situation, the vehicle's surroundings can be comprehensively taken into account, thereby enabling more precise decisions.

[0032] It is also advantageous if the usage includes:

[0033] Modeling the human driving behavior of at least one driver in a current situation based on the at least one generated driver model and preferably further based on the at least one context information, wherein the at least one driver is a driver of the vehicle and / or a driver of at least one other vehicle in an environment of the vehicle.

[0034] The method according to the invention can thus enable the modeling of realistic driving behavior of individual drivers, i.e., of the vehicle and / or at least one other vehicle in the vicinity of the vehicle. By taking into account the individual driver goals and at least one piece of contextual information, precise predictions about the driver's decisions and actions can be made, which can contribute to improved safety and efficiency of at least partially automated driving.

[0035] Advantageously, the invention may provide that the use includes:

[0036] Predicting a trajectory of the vehicle and / or a trajectory of at least one other vehicle in a vicinity of the vehicle based on the at least one generated driver model and preferably R.415561

[0037] - 6- furthermore based on the modeled human driving behavior of at least one driver in the current situation.

[0038] The trajectory of a vehicle, or of at least one other vehicle in its vicinity, can be predicted based on an analysis of sensor data from at least one sensor on the vehicle, such as a camera, LiDAR, radar, or ultrasonic sensor. This sensor can detect the vehicle's surroundings and determine information such as the position, speed, and direction of other vehicles or objects. This data can then be processed by an algorithm to predict the future movement of the vehicle, or of at least one other vehicle in its vicinity. Factors such as road conditions, the behavior of other road users, and traffic regulations can also be taken into account. The predicted trajectory can then be used to adjust the vehicle's behavior and avoid potential collisions.

[0039] It is also conceivable that the use could optionally include:

[0040] Determining a driving action to be performed for the vehicle based on the predicted trajectory of the vehicle and / or the trajectory of at least one other vehicle in the vicinity of the vehicle.

[0041] The driving action to be performed can be a maneuver such as emergency braking or an evasive maneuver. It is also conceivable that the driving action to be performed is the vehicle continuing its normal journey along a specific trajectory.

[0042] According to another advantage, it may be provided that the use also includes:

[0043] Initiating vehicle control based on the identified driving action to be performed.

[0044] This allows the vehicle to be controlled precisely and safely, as it is initiated directly taking into account human driving behavior.

[0045] It is possible that the method according to the invention is used in a vehicle. The vehicle can be, for example, a motor vehicle and / or a passenger vehicle. R.415561

[0046] - 7- The vehicle must be a motor vehicle and / or at least partially automated / autonomous. The vehicle may have vehicle equipment, e.g., for providing an autonomous driving function and / or a driver assistance system. The vehicle equipment may be designed to control the vehicle at least partially automatically and / or accelerate and / or brake and / or steer.

[0047] The machine learning system, particularly in the form of a machine learning model, which, within the framework of the method according to the invention, can be used, for example, to determine the at least one driver's destination based on the analysis of sensor data or to predict the trajectory of the vehicle and / or the trajectory of at least one other vehicle in the vicinity of the vehicle, is trained, in particular, for classification and, especially, for object detection. The training can be provided for training the machine learning system or the machine learning model using a training dataset for the classification, in particular for image classification, of image data such as digital images based on image points and / or pixels, in particular pixel values, preferably edges or pixel attributes (of the image data). The image data or digital images can, for example, be...The recording results from a capture by at least one sensor, preferably at least one camera, preferably a vehicle, and most preferably a camera and / or vehicle environment during a journey (of a vehicle). The capture is possible, for example, by at least one camera of the vehicle. The classification can be designed to recognize objects in an environment depicted by the image data or digital images and / or to capture a traffic scene.

[0048] The classification can be used for various technical applications. One example is its use in vehicles. Based on the classification, and in particular at least one classification result, at least one control action, preferably for a vehicle or another technical system, can be initiated and / or carried out.

[0049] A classification result may include at least one of the following results and / or be specific to at least one of the following results: R.415561

[0050] - 8- be: a category of objects, an identification of objects, a position of objects and / or obstacles (e.g. in the direction of travel or beside the direction of travel), a presence of obstacles, a description of a traffic scene, a hazard warning, a number of objects, a type and / or position of road markings and / or a road boundary, a position and / or state of traffic signal systems, a position of a roadway, or the like.

[0051] Based on the classification result, at least one control action can be initiated and / or carried out for the vehicle. The control action can include at least one of the following: braking, steering, accelerating, overtaking, emergency braking, activating an alarm system, activating hazard warning lights, activating a turn signal, controlling the lights, or the like.

[0052] Classification allows, for example, the detection of an obstacle, regardless of whether it is directly in the direction of travel or to the side. Depending on the location (e.g., based on the expected vehicle trajectory), an appropriate control action, such as braking or swerving, can be initiated.

[0053] For example, braking can also be initiated if the classification indicates that there are obstacles in the direction of travel and / or a collision is likely. It is also conceivable that a lane and / or lane boundary could be detected based on the classification in order to move the vehicle, at least partially automatically, along the lane through the control action.

[0054] The terms "classification" and "image classification" can also include "object detection" or "object detection in images." This refers specifically to classifying whether or not objects are present in certain areas of the image. Furthermore, the terms "classification" and "image classification" can also refer to "semantic segmentation," particularly in the form of pixel-level classification. R.415561

[0055] - 9-

[0056] Accordingly, the training can result in at least one trained machine learning model, which can be used for classification and / or object detection. Its use, and thus the inference, can be implemented, for example, in a vehicle. The input data points can be, for example, pixels from image data or be based on them, in order to perform the classification and / or object detection of the data points based on the pixels. The input data can include sensor and / or image data, which at least partially results from acquisition with a sensor, preferably a camera sensor, and / or which has been at least partially synthesized, i.e., in particular, replicates the real data of a sensor. Specifically, it can be provided that the values ​​of image points, preferably pixels, in the image data represent the environment of a sensor and / or a vehicle and / or a traffic scene.A classification, preferably image classification and / or object detection, based on these values ​​can be provided. This enables, for example, the detection of objects in the traffic scene. The image data can be, for example, images from a radar sensor and / or an ultrasonic sensor and / or a LiDAR sensor and / or a thermal imaging camera. Accordingly, the images can also be presented as radar images and / or ultrasonic images and / or thermal images and / or LiDAR images.

[0057] The invention also relates to a computer program, in particular a computer program product, comprising instructions which, when executed by at least one computer, cause it to execute the method according to the invention. Thus, the computer program according to the invention offers the same advantages as those described in detail with reference to a method according to the invention.

[0058] Also part of the invention is a device for data processing configured to execute the method according to the invention. The device can, for example, be at least one computer which executes the computer program according to the invention. The computer can have at least one processor for executing the computer program. A non-volatile data storage device can also be provided in which the computer program is stored and from which the computer program can be read by the processor for execution. R.415561

[0059] - 10-

[0060] The invention may also relate to a computer-readable storage medium which contains the computer program according to the invention and / or includes instructions which, when executed by at least one computer, cause it to execute the method according to the invention. The storage medium is, for example, designed as a data storage device such as a hard drive and / or non-volatile memory and / or a memory card. The storage medium can, for example, be integrated into the computer.

[0061] Furthermore, the method according to the invention can also be implemented as a computer-implemented method. Alternatively or additionally, at least one of the disclosed method steps can be computer-implemented and / or carried out automatically.

[0062] Further advantages, features, and details of the invention will become apparent from the following description, in which exemplary embodiments of the invention are described in detail with reference to the drawings. The features mentioned in the claims and in the description can each be essential to the invention individually or in any combination. The drawings show:

[0063] Fig. 1 shows a schematic visualization of a method, a device, a storage medium and a computer program according to exemplary embodiments of the invention.

[0064] Fig. 2 shows a schematic representation of a vehicle and another vehicle according to exemplary embodiments of the invention.

[0065] Fig. 3 shows a schematic representation of a graphic representation according to exemplary embodiments of the invention.

[0066] Figure 1 schematically depicts a method 100, a device 10, a storage medium 15, and a computer program 20 according to exemplary embodiments of the invention. R.415561

[0067] - 11 -

[0068] Fig. 1 shows, in particular, an embodiment of a method 100 for using at least one driver model in at least partially automated driving of a vehicle 1. In a first step 101, driver goals are provided, wherein the driver goals characterize human driving behavior in road traffic. In a second step 102, at least one driver model is generated based on the provided driver goals, wherein the at least one driver model is a graphical representation 11, in which nodes 12 of the graphical representation 11 represent the driver goals and edges 13 of the graphical representation 11 represent relationships between the driver goals.In a third step 103, the at least one generated driver model is used in the at least partially automated driving of the vehicle 1 in order to take into account the human driving behavior in road traffic.

[0069] Fig. 2 shows a schematic representation of a vehicle 1 with a sensor 2 and a driver 3 and of another vehicle 4 according to exemplary embodiments of the invention.

[0070] According to exemplary embodiments of the invention, an advanced human driver model (AHDM) is provided, encompassing a variety of goals, preferences, and capabilities. Furthermore, the driver model can be context-dependent with respect to features such as time, day of the week, location, or passengers in vehicle 1. The driver model (AHDM) according to the invention can be used in a number of different applications, such as trajectory prediction and / or planning of surrounding vehicles 4 in autonomous driving systems from levels 1 to 5. Moreover, the driver model according to the invention can be used for motion planning in autonomous driving systems from levels 3 to 5 or for simulation tools for the validation of autonomous driving systems.

[0071] The driver model according to the invention preferably comprises various driver objectives, which are divided into main objectives and sub-objectives, hard objectives and soft objectives. R.415561

[0072] - 12- can be subdivided, with each driver goal being assigned a specific importance or priority value. Some driver goals can be learned from observational data, such as sensor data with context-dependent information. The context of a driving situation can, in turn, be used to make the driver model according to the invention context-dependent. The use of this driver model according to the invention can advantageously lead to a more accurate modeling of the human driving behavior of surrounding road users and thus to a more accurate prediction of driving behavior. It can also contribute to optimized motion planning that takes into account driver and / or passenger information.

[0073] According to exemplary embodiments of the invention, a motion planning method for at least partially automated driving is further described, which uses the advanced human driver model (AHDM), or driver model for short, according to the invention.

[0074] The driver model according to the invention can be used to model a vehicle 1, which can also be referred to as an ego-vehicle within the scope of the present invention, and relevant surrounding vehicles 4. Much information is known, particularly about the ego-vehicle 1. This allows for a detailed construction of the driver model for the ego-vehicle 1. The driver models for the surrounding vehicles 4 can be estimated during driving, using pre-trained or known driver types. With the help of perceptual information about a history of past movements, an individual driver model according to the invention can be estimated for each relevant surrounding vehicle. This leads, in particular, to a more accurate modeling of the agents or vehicles. The driver model according to the invention preferably also models a temporal interaction between two vehicles (e.g.,(the ego vehicle 1 and a vehicle 4 in the vicinity) to determine their incremental prediction of the behavior of the other vehicle 4 and thus to derive their own movement plan. R.415561.

[0075] - 13-

[0076] According to the invention, a detailed driver model is provided that represents various factors influencing a human driver's driving behavior. This leads, for example, to improved trajectory prediction and thus to improved motion planning in automated vehicles. Another advantage is, for example, that significantly less training data is required for accurate modeling of driver behavior.

[0077] The driver model according to the invention can be used in various applications such as driver assistance systems and autonomous driving systems.

[0078] The driver model according to exemplary embodiments of the invention has, for example, the following structure, components, and / or functions: various driver objectives, e.g., safety, traffic rules, destination / driving goals, social values ​​(altruistic, egoistic, sadistic-masochistic), driving ability (reaction time with driver assistance systems), driving style (average speed, maximum acceleration / deceleration), and / or distractibility. Furthermore, the objectives can be subdivided into primary and secondary objectives and / or into hard and soft objectives. The importance of the driver objectives and the dependencies between them can also be specified.

[0079] The setting or definition of each driver's goal can be predefined (e.g., safety), user-defined (e.g., preferences, adherence to traffic rules), and / or learned from data (e.g., skills, social values). Furthermore, contextual information can also be taken into account, such as the time of day, day of the week / weekend, the number and type of passengers in the vehicle, and / or a location area.

[0080] Furthermore, different driver types can be learned, e.g. based on data or user-defined.

[0081] In this way, a variety of driver models can be generated that differ in their driver goals, meaning, and / or context. These modeled driver models can then be used for driver assistance, automated driving, driver simulation, and / or the validation of automated driving systems. R.415561

[0082] - 14-

[0083] Examples of driver objectives and driver sub-objectives include safety objectives. Safety objectives can include, for example, avoiding collisions with vehicles, pedestrians, or objects, maintaining a prescribed safety distance in longitudinal or lateral directions, or showing consideration for other vehicles.

[0084] Another driver goal might be to reach a destination, e.g., in the shortest possible time, via the shortest route, via the most fuel-efficient route, or via the most comfortable route, e.g., with few intersections or curves.

[0085] Another driving objective can be compliance with traffic regulations, such as adhering to speed limits or exceeding them by a certain amount. Furthermore, drivers may observe prohibited lane dividers (solid, double solid) and red lights. Driving must also be permitted only in the designated direction, e.g., prohibited turns and U-turns. Staying within the lane, i.e., not driving on the hard shoulder, sidewalk, or parking area, can also be a sub-objective. Another sub-objective of complying with traffic regulations can be observing overtaking prohibitions.

[0086] Social value orientations can also be considered driver goals, particularly in dimensions such as altruism, egoism, sadism, or masochism. This can manifest in interactions with other road users, for example, braking to allow other vehicles to merge into the driver's lane or overtaking only when other vehicles are not obstructed. The degree to which these behaviors are expressed can vary depending on the individual and the context.

[0087] The factors of the human driver can also be taken into account. The driver in question may be experienced or inexperienced. An experienced driver can understand traffic situations and risks well and make predictions. The driver in question may also be risk-prone or cautious. Risk-prone behavior can be characterized by high speed, a short distance to other vehicles, etc.

[0088] - 15- towards a vehicle ahead and / or by accelerating and / or braking sharply. The driver in question may also be aggressive or defensive, which may manifest itself in not braking for others, not letting others pass, not pulling forward, and / or running a red light. The driver in question may also be distracted or inattentive. The distracted driver may lead to risky situations and / or incorrect decisions. The driver in question may also be impaired or healthy. This may be noticeable in reaction time, coordination, and / or judgment. The driver in question may also have high or low cognitive ability and / or visual acuity, which may be noticeable in early and rapid reactions and quick speed adjustments to traffic.

[0089] According to exemplary embodiments of the invention, the driver model can be defined as AHDM = (G, E), where G = {Gh UG S Let} be a set of driver goals {gi , g2, ..., gn} and E represent a set of relationships or dependencies between the driver goals. Each driver goal gh e Gh is, in particular, a hard goal that must be met for the system, for example, vehicle 1, to be considered safe, and each driver goal g s eG sA driver goal is, in particular, a soft goal, the fulfillment of which is desirable but which can be violated to a certain extent without resulting in a serious incident. Furthermore, driver goals can also be linked to a specific driver de D, where D is the set of drivers. The subset of driver goals associated with a specific driver is thus denoted, for example, as follows: G' = { ge G | 3d e D : f(d, g)}, where f(d, g) specifically represents an assignment of a specific driver d to a driver goal g.

[0090] Each driver-goal G can be formally represented as an entity in a graphical representation such as a knowledge graph (KG), including its respective attributes and relationships. The use of a knowledge graph offers, for example, the following advantages: It allows for the easy mapping and integration of other models of human driving behavior, as well as the integration of rules and prior knowledge about human driving behavior.

[0091] - 16- driving behavior. The relationships can be enriched with semantic axioms to encode similarity, hierarchy, symmetry, and transitivity. Furthermore, relationship qualifiers can be added to provide information origin and development.

[0092] A key task of motion planning according to the invention is, in particular, to plan the next action of the vehicle as accurately as possible in accordance with the specified driver objectives and the dynamics of the environment.

[0093] In any given specific situation, vehicle 1 aims to optimize the achievement of the driver's objectives. To optimize this process for each action, the dependencies between the driver's objectives can also be represented in the knowledge graph. Thus, for each subset of objectives G', an objective dependency graph DG' = (G', E) can be extracted in the form of a directed acyclic graph (DAG). Here, E c G' x G' ​​is, in particular, a set of directed edges representing dependencies between driver objectives. If (gi,gj) ∈ E, then the driver objective gj depends, in particular, on the driver objective gi. Furthermore, gj preferably fails whenever the evaluation of gi fails.

[0094] Using DG' as a predefined input can enable a prioritization of the evaluation of driver objectives, whereby in the case of a violation of the high-priority driver objectives (i.e., the parent nodes in the graph), the dependent driver objectives are effectively excluded from further consideration, i.e., the subsequent subgraph is specifically excluded from further evaluation.

[0095] Information about a driver 3 of vehicle 1 can be obtained from various sources, either during the journey, if information about the driver 3 is available, such as a driving history, or directly from sensors 2. Alternatively or additionally, the information can be provided manually by the driver 3. This allows an accurate driver model according to the invention to be created for the driver 3 of vehicle 1. R.415561

[0096] - 17-

[0097] It can be assumed that the hard targets Gh, e.g., safety-relevant ones, are the same for all surrounding drivers or vehicles 4. It is possible that some of the soft targets G s Drivers can observe various aspects, such as their driving style, maneuver intentions, whether they are in a hurry, or whether they are cooperative. However, other "soft" goals are not easily observed. Therefore, a range of driver types can be learned from data. Observations can be used to create a more detailed driver model that models driver goals and their relationships or dependencies. These relationships or dependencies can also be learned from the data. A more accurate and detailed driver model with learned relationships or dependencies is particularly well-suited to describing human driving behavior even in cases where no observations are available during the training period.

[0098] The driver model can be learned using, for example, a machine learning model such as a deep neural network, gated recurrent units (GRUs), or transformers, as shown in the examples.

[0099] When applying the driver model, preferably the most probable driver type is used, depending on the statistics and context.

[0100] Furthermore, at least one piece of contextual information can be taken into account, for example through a context model, to make the driver model context-dependent. A context C can include, but is not limited to, the following factors: a location, e.g., a country; a driving area, e.g., city or country; a road type, e.g., highway, country road, city street, expressway; a time of day; a day of the week; traffic density; and / or an individual vehicle context, e.g., which lane, adjacent vehicles / pedestrians, distance to traffic elements (traffic light, traffic sign, intersection, etc.).

[0101] The driver behavior DB at time t can be described as a function of the dependency graph DG', which represents the goals of a particular driver and the context C at time t: R.415561

[0102] - 18-

[0103] DB'(t) = f(DG'(t), C(t))

[0104] The driver model according to the invention can be used to predict a trajectory, including a raster-based representation where the map and the agents are represented in a raster with different layers that take into account different types of information. It can also be used in a graph- or vector-based representation where the map elements are represented as vectors or connected nodes and the agents as nodes mapped onto the map. The driver model according to the invention can be used, for example, in a phase where possible maneuvers of a vehicle are estimated. While in classical trajectory prediction systems possible target trajectories are learned, the AHDM according to the invention is used in particular to construct a number of possible trajectories of a vehicle based on the selected driver model DBi(t), wherein the driver type is preferably determined based on the context.The map representation, agents, and other information can be entered into the AHDM and encoded using deep learning, GRU, attention, or other methods.

[0105] A decoder for predicting a trajectory can be based on a generative, regressive, autoregressive, hybrid, or other method.

[0106] Based on observed human driving behavior, the driver type can be estimated and the likely behavior predicted, particularly as a prediction of the trajectory. The method according to the invention preferably uses an AHDM of the surrounding road users to accurately predict their driving paths.

[0107] The priority of the various goals, which in the AHDM are modeled, for example, as a hierarchical graph, enables, in particular, an efficient evaluation of the goals in different traffic situations. It is possible, if a higher-level driver goal fails, to truncate the hierarchical graph and to continue the evaluation of all dependent nodes, i.e., driver goals. (R.415561)

[0108] - 19- den. If, for example, the goal of being safe is violated, then preferably all other goals, such as goals related to driver preferences, will not be evaluated further.

[0109] According to exemplary embodiments of the invention, a raster-based representation can be used, in which the map and the agents are displayed in a raster with different layers representing different types of information. Alternatively, a graph- or vector-based representation can be used, in which map elements are represented as vectors or connected nodes, and agents are represented as nodes mapped onto the map.

[0110] The driver model according to the invention can be modeled by a graphical representation 11, in particular a heterogeneous graph, as shown in Fig. 3. Here, the nodes 12 can represent the driver goals and the edges 13 the relationships or dependencies between the driver goals. Furthermore, a heterogeneous neural graph network or a heterogeneous graph transformer can be encoded. The representation of the map, the agents, and other information can be input into the AH DM and encoded using deep learning, GRU, attention, or another method.

[0111] Motion planning can be learned, for example, using a deep reinforcement learning method. For scenarios in which an environment agent might interact with vehicle 1, the action space can be defined, for example, as a Markov decision process. The state space of the environment vehicle 4 and vehicle 1 includes, for example, the AHDM, position, speed or direction of travel, as well as the map and the context. The action space of vehicle 1 includes, for example, acceleration and direction of travel.

[0112] The reward or loss function includes, for example, driver models of relevant surrounding vehicles 4, a driver model of vehicle 1 itself, and an interaction score. The interaction score is, in particular, a joint prediction of an interaction between the two vehicles 1,4. The interaction score can predict the temporal evolution of both agents and R.415561

[0113] - 20- take into account the probability of an accident. Furthermore, the probability that the surrounding vehicle either accelerates and risks an accident or restrains itself and avoids an accident can be taken into account. The preceding explanation of the embodiments describes the present invention exclusively by way of examples. Of course, individual features of the embodiments can be freely combined with one another, provided this is technically feasible, without departing from the scope of the present invention.

Claims

1. R.415561 - 21 - Claims 1. Method (100) for using at least one driver model in at least partially automated driving of a vehicle (1), comprising: Providing (101) driver objectives, wherein the driver objectives characterize human driving behavior in road traffic, Generating (102) at least one driver model based on the provided driver goals, wherein the at least one driver model is a graphical representation (11) in which nodes (12) of the graphical representation (11) represent the driver goals and edges (13) of the graphical representation (11) represent relationships between the driver goals, Using (103) the at least one generated driver model in the at least partially automated driving of the vehicle (1) to take into account human driving behavior in road traffic when driving the vehicle (1) at least partially automated. R.415561 - 22- 2. Method (100) according to claim 1, characterized in that the provision (101) comprises at least one of the following steps: Determine at least one of the driver's objectives based on an analysis of sensor data, wherein the sensor data result from the acquisition of at least one sensor (2), in particular at least one sensor (2) of the vehicle (1), Define at least one of the driver objectives based on a safety requirement, Automated or manual provision or modification of at least one of the driver's destinations based on an individual preference or driving characteristic of a driver (3) of the vehicle (1) and / or a driver (3) of at least another vehicle (4) in a vicinity of the vehicle (1).

3. Method (100) according to one of the preceding claims, characterized in that the driver objectives relate to the safety of road users, traffic rules, a driving goal of the vehicle (1), social values ​​of a respective driver, driving ability of a respective driver, driving style of a respective driver and / or attention ability of a respective driver.

4. Method (100) according to one of the preceding claims, characterized in that the driver objectives are subdivided into main driver objectives and sub-driver objectives subordinate to the main driver objectives and / or wherein the driver objectives are characterized as hard and soft driver objectives, wherein the hard driver objectives must be prioritized and fulfilled within the framework of at least partially automated driving and the soft driver objectives are only taken into account if the hard driver objectives are fulfilled. R.415561 - 23- 5. Method (100) according to one of the preceding claims, characterized in that the method (100) further comprises: Providing or determining at least one piece of contextual information regarding the environment of the vehicle (1), wherein the at least one piece of contextual information is selected from the location of the vehicle (1), the condition of a road, the current time of day, the weather condition and / or the current traffic, wherein, in the context of at least partially automated driving of the vehicle (1), the at least one piece of contextual information is also taken into account.

6. Method (100) according to one of the preceding claims, characterized in that the use (103) comprises: Modeling the human driving behavior of at least one driver (3) in a current situation based on the at least one generated driver model and preferably further based on the at least one context information, wherein the at least one driver (3) is a driver of the vehicle (1) and / or a driver of at least one other vehicle (4) in a vicinity of the vehicle (1).

7. Method (100) according to one of the preceding claims, characterized in that the use (103) comprises: Predicting a trajectory of the vehicle (1) and / or a trajectory of at least one other vehicle (4) in an environment of the vehicle (1) based on the at least one generated driver model and preferably further based on the modeled human driving behavior of the at least one driver (3) in the current situation. R.415561 - 24- 8. Method (100) according to claim 7, characterized in that the use (103) further comprises: Determining a driving action to be performed for the vehicle (1) based on the predicted trajectory of the vehicle (1) and / or the trajectory of at least one other vehicle (4) in the vicinity of the vehicle (1).

9. Method (100) according to claim 8, characterized in that the use (103) further comprises: Initiating control of the vehicle (1) based on the identified driving action to be performed.

10. Computer program (20), comprising instructions which, when the computer program (20) is executed by at least one computer (10), cause it to execute the method (100) according to one of the preceding claims.

11. Device (10) for data processing, which is configured to carry out the method (100) according to any one of claims 1 to 9.

12. Computer-readable storage medium (15) comprising instructions which, when executed by at least one computer (10), cause it to execute the steps of the method (100) according to any one of claims 1 to 9.

Images