WO2023275401A1

WO2023275401A1 - Simulation of road users with emotions

Info

Publication number: WO2023275401A1
Application number: PCT/EP2022/068398
Authority: WO
Inventors: Alexander Knorr; Lukas Brostek; Isabelle Garzorz
Original assignee: cogniBIT GmbH
Priority date: 2021-07-02
Filing date: 2022-07-04
Publication date: 2023-01-05
Also published as: DE102021117086A1

Abstract

Disclosed are techniques for simulating actions of road users taking into account dynamic values of one or more emotional states (161) of road users. Examples of emotional states that can be taken into account are aggressiveness or anxiety. A subjective perception of a situation in the surroundings of the road users can also be simulated on the basis of the dynamic values of the one or more emotional states.

Description

SIMULATION OF ROAD USER WITH EMOTIONS

TECHNICAL AREA

Various examples of the disclosure relate generally to simulating road user behavior taking into account road user emotions. Based on this, training or an evaluation of a driver assistance functionality of a motor vehicle can take place.

BACKGROUND

The use of automated or autonomous driver assistance functionalities (ADAS systems) is widespread and continues to increase. In order to develop ADAS systems safely, it can typically be helpful to consider a large number of different traffic situations. For example, different traffic situations can be taken into account when designing an ADAS system if appropriate decision logic is programmed or trained. However, different traffic situations can also be taken into account when validating an ADAS system if the programmed decision logic is already available and its performance is to be checked.

This can be made possible, for example, by using prototypes in field tests. Alternatively, techniques are also known for simulating the dynamic behavior of a large number of road users in a simulation world. User data can then be generated based on the simulation and it is possible to train and/or evaluate, ie validate or test, the ADAS system based on the user data.

In contrast to the field test, the use of a simulation has the advantage that the behavior of an ADAS system can be quickly checked in relation to a particularly large number of different traffic situations. More traffic situations can be detected. Critical traffic situations can also be simulated in a targeted manner. Therefore, the behavior of ADAS systems can be trained or evaluated more accurately and reliably on the basis of simulations. However, it has been observed that sometimes the simulations of the dynamic behavior of road users are not particularly accurate or may show systematic differences to the behavior of real road users. BRIEF SUMMARY OF THE INVENTION

Therefore, there is a need for improved techniques for simulating the dynamic behavior of a variety of road users. In particular, there is a need for techniques that can simulate the dynamic behavior of road users in a particularly accurate and realistic manner. This object is solved by the features of the independent patent claims. The features of the dependent claims define embodiments.

Techniques for simulating a time-varying, i.e. dynamic, behavior of a large number of road users in a simulation world described. Emotional states are taken into account for road users. These emotional states, such as aggression, anger, confusion, nervousness, fear, sadness, joy, tiredness, exhaustion, or stress, can affect the decision-making process of road users (sometimes referred to as agents) regarding actions to be taken to interact with the simulation world. This makes it possible to examine how the various road users interact with the environment and with each other if subjective (ie agent-specific) boundary conditions or influences on behavior in the form of emotional states are taken into account. Depending on their character and environmental situation, different road users can carry out actions that are largely guided by different emotional states.

A computer-implemented method includes simulating a dynamic behavior of a large number of road users in a simulation field and during a simulation period. In this case, the simulation for each road user of the multiplicity of road users includes the determination of dynamic values of one or more emotional states during the simulation period. The simulating also includes, for each road user of the plurality of road users, determining actions of the respective road user to interact with the simulation world based on the dynamic values of the one or more emotion states. The computer-implemented method also includes generating user data for training or for evaluating an automated or autonomously acting driver assistance functionality of a motor vehicle based on the dynamic behavior of the large number of road users.

A computer program or a computer program product or a computer-readable storage medium includes program code. The program code can be loaded and executed by a processor. This causes the Processor executes a computer-implemented method. The method includes simulating a dynamic behavior of a large number of road users in a simulation world and during a simulation period. In this case, the simulation for each road user of the large number of road users includes the determination of dynamic values of one or more emotional states during the simulation period. The simulation also includes, for each road user of the plurality of road users, determining actions of the respective road user to interact with the simulation world based on the dynamic values of the one or more emotion states. The computer-implemented method also includes generating user data for training or for evaluating an automated or autonomously acting driver assistance functionality of a motor vehicle based on the dynamic behavior of the large number of road users. A device includes a process and a memory. The processor can load and execute program code from memory. This causes the processor to simulate dynamic behavior of a large number of road users in a simulation world and during a simulation period. In this case, the simulation for each road user of the large number of road users includes the determination of dynamic values of one or more emotional states during the simulation period. The simulation also includes, for each road user of the plurality of road users, determining actions of the respective road user for interaction with the simulation world based on the dynamic values of the one or more emotion states. In addition, the processor is set up to generate user data for training or for evaluating an automated or autonomously acting driver assistance functionality of a motor vehicle based on the dynamic behavior of the large number of road users. The features set out above and features described below can be used not only in the corresponding combinations explicitly set out, but also in further combinations or in isolation without departing from the protective scope of the present invention.

BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 schematically illustrates an apparatus for data processing according to various examples.

FIG. 2 is a flowchart of an example method. FIG. 3 schematically illustrates an agent model of a road user with multiple modules according to various examples.

FIG. 4 schematically illustrates several exemplary transfer functions that describe a change in the value for an emotional state based on an environmental situation for various static physiological-psychological properties.

FIG. 5 schematically illustrates a change in a dynamic emotion state over time according to various examples.

FIG. 6 illustrates a traffic situation with several road users according to various examples. FIG. 7 illustrates aspects related to decision making depending on the value of an emotion state according to various examples.

FIG. 8 illustrates an exemplary statistical frequency for the occurrence of certain values of a physiological-psychological property. DETAILED DESCRIPTION OF EMBODIMENTS

The properties, features and advantages of this invention described above, and the manner in which they are achieved, will become clearer and more clearly understood in connection with the following description of the exemplary embodiments that are explained in more detail in connection with the drawings.

The present invention is based on preferred forms of execution explained in more detail with reference to the drawings. In the figures, the same reference symbols designate the same or similar elements. The figures are schematic representations of various embodiments of the invention. Elements depicted in the figures are not necessarily drawn to scale. Rather, the various elements shown in the figures are reproduced in such a way that their function and general purpose can be understood by those skilled in the art. Connections and couplings between functional units and elements shown in the figures can also be implemented as an indirect connection or coupling. A connection or coupling can be implemented wired or wireless. Functional units can be implemented in hardware, software, or a combination of hardware and software.

Techniques are described below for simulating the dynamic, i.e. time-varying, behavior of a large number of road users in a simulation world during a simulation period. This is formed from the large number of actions undertaken by the various road users during the simulation period.

For example, the behavior of road users driving passenger vehicles, trucks or bicycles can be simulated. It the behavior of pedestrians or playing children could also be simulated.

In particular, a trajectory, i.e. the movement path, of road users can be simulated through the simulation world. Other interactions of road users with other road users or elements of the simulation world can also be simulated.

The techniques described herein make it possible to determine payload data based on the simulation. The user data can be determined based on the dynamic behavior of the large number of road users and can, for example, describe the trajectories of the road users through the simulation world. The user data can therefore describe a sequence of positions of the road users. As an alternative or in addition, the user data can also contain other information obtained from the simulation, for example information on the internal status of the various road users, which, for example, makes the decision-making process of the road users plausible, which leads to a specific action.

Based on such user data, different application scenarios can be made possible. In particular, it would be possible for the user data to be used for training or for evaluation (ie testing and/or validating) of an ADAS system. The ADAS system can be an automated driver assistance functionality, for example. For example, such an automated driver assistance functionality could support the driver of a motor vehicle in certain specific driving situations, such as when driving on the freeway, when keeping a distance from a vehicle in front (adaptive cruise control), when turning (turning assistant), when driving through a narrow alley, etc. For example, the ADAS system could be designed for autonomous driving, for example according to the Society of Automotive Engineers classification "Level 1" to "Level 5". For example, it would be possible for the ADAS system to include a machine-learned algorithm, such as an artificial neural network. This can be trained based on the user data. The artificial neural network can, for example, receive sensor data from environment sensors of a motor vehicle as input and, based on this, generate control instructions for the operation of the motor vehicle. Such sensor data can be simulated based on the user data. For this purpose, the user data can be loaded, for example, from a rendering program and then sensor data can be simulated by the rendering program. Then a ground truth can be generated from control instructions by an expert using the simulated sensor data as input and based on a deviation between the control instructions output by the artificial neural network and the ground truth (a corresponding loss function can be defined ) weights of the artificial neural network can be adjusted with a gradient descent method. Such techniques can be helpful not only in connection with the training of an artificial neural network, but also, for example, in connection with the computer-assisted design (CAD) of manually parameterized algorithms.

Another application scenario relates to testing an ADAS system, for example. The ADAS system can be confronted with traffic situations that are mapped by the user data. It can then be checked whether control instructions generated by the ADAS system achieve a desired result. In such a case, too, an emulation of sensor data can be carried out as an input into the ADAS system based on the user data.

In addition to such testing of individual traffic situations, validation of the ADAS system based on the user data would also be conceivable. at a With such a validation, a degree of reliability can be determined. For example, many traffic situations of a specific class of traffic situations can be used to generate correspondingly simulated sensor data. It can then be checked within the class of traffic situations with what reliability (e.g. percentage) the ADAS system within a specific target responds.

The simulation can include a sequence of time steps. For each time step, one or more internal states can be determined for each road user and based on this, one or more actions for the interaction of the respective road user with the simulation world can be determined. The one or more internal states may be determined according to an agent model.

According to various examples, the one or more internal states also include one or more dynamic emotion states. In the agent model, neuronal and cognitive steps of signal processing in humans when participating in road traffic (e.g. as a car driver) can be simulated.

In detail, the agent model can include several modules, e.g. for perception, cognition, and action (engl perceive-think-act). The dynamic emotion states can be formed and/or taken into account in one or more such modules of the agent model. This means that, for example, perception and/or processing/cognition and/or action can be influenced by dynamic emotional states.

Examples of such dynamic emotional states are eg aggression; Fury; confusion; Nervousness; Fear; Sadness; Joy; Fatigue; Exhaustion; and stress. If one or more such dynamic emotional states are used, they can each be assigned a value or level, where the value changes over time (e.g. if the driver becomes aggressive due to an external stimulus).

According to various examples, the one or more internal states also include physiological and/or psychological properties (PP properties). In particular, these can be assigned statically to the various road users (ie, in contrast to the dynamic emotional states, show no change over time). The PP properties can, for example, describe certain boundary conditions or limitations for the agent-specific signal processing in the agent model, such as restricted field of view, reaction and processing times, typical errors in distance or speed estimation, limited short-term memory, etc. The PP properties can eg be selected from: age, physical condition, influence of intoxicants, driving experience, compliance with rules, general driving style, comfort, distractibility. In order to map the entire range of human road users, the individual modules of the agent model can be parameterized individually for each road user. This means that the values for the PP properties can be chosen differently depending on the road user. Factors such as age, -, alcohol influence, driving style (aggressive vs. defensive) can be adjusted by the user for the different road users. It would also be possible to parameterize the PP properties for the large number of road users with a statistical frequency distribution. These factors are then converted into individual internal states that quantitatively influence the algorithms within the modules. In principle, different techniques are possible to determine current values for the one or more emotional states per time step. The dimensioning of the one or more emotion states (i.e. the setting of Values for the one or more emotion states) can be done in an emotion module of the agent model. For example, in each simulation time step, the emotion module evaluates the internal representation of the agent (his individual idea of the current traffic situation, which may deviate from reality) with regard to relevant variables of other road users (their trajectories, distances to each other and to the ego agent). , their light and warning signals). Your own status (current speed, current position/lane on the road, current driving manoeuvre, etc.) is constantly compared with a target status, with the difference influencing the emotional status. This therefore means that the one or more dynamic emotion states can be determined depending on an influencing of a respectively planned trajectory and/or speed by an environmental situation of the respective road user in the simulation world.

The determination of values for the one or more dynamic emotion states can take place, in particular, taking into account the one or more PP properties. For example, it would be conceivable that for each road user a subjective perception of a dynamic environment situation of the respective road user in the simulation world is determined as a function of the one or more PP properties of the respective road user. The one or more dynamic emotional states for the road users can then be determined based on the respective subjective perception of the dynamic situation in the environment. This means that it would be possible to dimension one or more emotional states depending on the perception of the surrounding situation. Next, a concrete example to illustrate such and other dependencies is outlined. For example, there can be two emotional states: aggression (cautious vs. risky behavior), and anxiety. Aggression and anxiety can have different values. The influence of the subjective Perception of the surrounding situation on the dimensioning of aggression and anxiety on the two emotional states can in turn be parameterized agent-specifically, so that, for example, the same traffic situation (e.g. hesitation from behind) leads to increasing aggressiveness in the first road user and increasing anxiety in the other. This influence of the dynamic environmental situation on the dimensioning of one or more emotional states can be described by an appropriately defined transfer function. The transfer function can in turn be defined subjectively, namely as a function of one or more PP properties. For example, the same subjectively perceived environmental situation can increase the level of aggression in a driver who is under the influence of alcohol, but the level of anxiety in a tired driver. Alternatively or additionally, different basic levels of aggression and anxiety could also be mapped depending on the PP properties. The values of the two dynamic emotional states, in turn, influence the agent's decision-making in each journal, ie the one or more actions performed. Depending on the one or more emotional states, decision thresholds can be shifted during the risk assessment step, so that, for example, higher aggression leads to riskier decisions and vice versa. Anxiety has an effect on the decision-making process, for example by shifting decision thresholds, but at the same time by giving greater weight to the uncertainty present in the internal representation.

FIG. 1 schematically illustrates an apparatus 90 for data processing that can be used for various techniques described herein. Device 90 includes a processor 92. This could be implemented as a CPU or GPU, for example. In addition, the device 90 also includes a memory 93 and the processor 92 can load and execute program code from the memory. When the processor 92 executes the program code, that is, processor 92 performs techniques as described herein, for example: simulating the dynamic behavior of a variety of road users; Parameterization of internal states of an agent model of the road users based, for example, on a user input received via an interface 91; generating user data 165 based on a dynamic behavior of the plurality of road users; Evaluation or training of an ADAS system based on the user data 165.

FIG. 2 is a flowchart of an example method. The method of FIG. 2 can be executed by one processor. For example, the method of FIG. 2 are executed by processor 92 of device 90 based on program code that processor 92 loads from memory 93. The method enables the dynamic behavior of road users to be simulated in a simulation world, taking into account emotional states. Time-varying values for one or more emotional states can be assigned to each road user. The movements and thus the behavior of road users depend on these time-varying values. The emotional states can thus be referred to as dynamic emotional states. First, in box 3005, the dynamic behavior of a large number of road users is simulated in a simulation world during a simulation period. For this purpose, the movements of road users on their way through the simulation world are determined.

In box 3010, user data is then generated based on the dynamic behavior of the road users. This user data can, for example, indicate the positions and location of road users as a function of time. Alternatively or additionally, this user data can also contain one or more internal States of an agent model of the various road users include sen. For example, it would be possible for the user data to specify the dynamic values for one or more emotional states of the various road users as a function of time. For example, the payload could indicate the level of aggression or anger for different road users as a function of time. For example, the user data could indicate whether a corresponding road user was distracted or not. The user data could, for example, indicate the line of sight of road users. Events such as collisions, high acceleration, strong braking/steering interventions can also be indicated.

By outputting such and/or other internal states of the agent model, a cause analysis for a specific behavior of the various road users can then be carried out in box 3015 .

In box 3015, the user data is then optionally used. In principle, a wide variety of uses are conceivable. For example, it would be possible for the user data to be used for a computer-assisted development process for an ADAS system. For example, specific operating parameters of the ADAS system could be set in such a way that the ADAS system also provides reliable assistance in specific situations that are shaped or influenced by the emotional behavior of other road users. It would also be conceivable for an ADAS system to be evaluated in box 3015 based on the user data. This means that, for example, the operation of the ADAS system can be tested, i.e. it could be checked whether the ADAS system also achieves a desirable result for situations that are significantly characterized by emotionally conditioned decisions and the squabbles of different road users are. A validation of the ADAS system could also take place; ie it it could be checked whether a desired result is achieved for a large number of traffic situations with a sufficient frequency. Optionally, following the computer-assisted development process for the ADAS system, the ADAS system could also be used to control a motor vehicle.

In particular, it would be conceivable in some scenarios that feedback 3020 of the use of the user data in box 3015 to the execution of the simulation box 3005 is implemented. For example, it would be conceivable that when an ADAS system is validated, a specific class of traffic situations is identified in which the ADAS system only has reduced reliability. Such a class of traffic situations can be characterized by a specific behavior of the various road users. Such a class of traffic situations could also be characterized by a certain static environment, for example a certain type of road, certain road layout, certain environmental conditions (e.g. heavy rain), special intersections, etc. Then it would be conceivable that the simulation in box 3005 again within of this class of traffic situations, for example with a higher time resolution, higher accuracy or higher sampling rate in relation to the corresponding parameter space. Slight variations in the simulation parameters could be made or the initial situation could be varied. Based on the user data obtained in this way, a more precise validation could then take place in box 3015, or it would even be conceivable for the design of the ADAS system to be adapted.

The simulation in box 3005 can be time-discrete for multiple time steps during a simulation period. For each time step, a series of internal states of each road user can be determined using an agent model and then the change in the position and / o the situation and / or speed of the respective road user in the Simulation world can be derived based on one or more corresponding actions. Details on the agent model are given below in connection with FIG. 3 described.

FIG. 3 illustrates aspects relating to an agent model 105. The agent model 105 is used to determine a series of internal states for each road user per time step 251. The agent model includes a module 120 for determining a subjective perception of an environment situation of the respective road user in the simulation world. In addition, the agent model 105 also includes a module 130 for determining actions by the respective road user based on their perception of the surrounding situation (dependency 902).

These are only exemplary modules. For example, it would be conceivable for a further module (not shown in FIG. 3 ) to be arranged between module 120 for perception and module 130 for action, which module enables cognitive processing of perceived information from module 120 .

In the scenario shown, the agent model 105 also includes a module 140 for determining dynamic values of one or more emotion states 161 during the simulation period. The actions are then determined in module 130 based on the dynamic values of the one or more emotion states from module 140 (dependency 925). The actions then describe the dynamic behavior of the road users and it is possible to generate the user data 165 based on this dynamic behavior.

In the example of FIG. 3, the action of the respective road user is determined based on the dynamic values of the one or more emotional states 161 (dependency 925). Alternatively or additionally it would be it is also conceivable that the subjective perception of the surrounding situation of the respective road user is determined based on the one or more emotional states from module 140 (dependency 906).

In FIG. 3 also shows that it is possible for the agent model 105 to take PP properties 162 into account. In principle, these can influence both the determination of the subjective perception in module 120 and the determination of values for one or more emotional states 161 in module 140, and/or the determination of actions in module 130.

Examples of emotional states 161 and their possible influence on determining actions and/or perception 120 (dependencies 925 and 906) are in TAB below. 1 described.

TAB. 1 : Various examples of emotional states 161 . In addition, the influences of the respective emotional states on perception or action are also presented. The corresponding examples are illustrative and other dependencies are conceivable. Various techniques have been described above in which actions of the respective road user for interaction with the simulation world are determined as a function of dynamic values of one or more emotional states 181 and/or the subjective perception of the environmental situation of the respective road user in the simulation world is determined . Alternatively or additionally, other modules of the agent model 105, for example a module for determining the cognitive processing of perceived information relating to the environment from module 120, could operate as a function of the dynamic values of one or more emotional states. Generally formulated, one or more modules of an agent model can have a dependency on dynamic values of one or more emotional states.

In principle, it would be possible for static PP properties 162 to be taken into account in addition to such transient emotional states, ie emotional states with dynamic values, compare FIG. 3.

There are, as a general rule, different ways to account for static PP properties 162 in the context of the agent model 105. It would be conceivable, for example, for the subjective perception of the surrounding situation to be determined for each road user as a function of one or more corresponding static PP properties 162 of the respective road user (dependency 904). Alternatively or additionally, it would also be conceivable for the values for the one or more emotion states 161 to be determined in module 140 based on the PP properties 162 (dependency 903). Alternatively or additionally, it would also be possible for one or more PP properties 162 to be taken into account in module 130 (dependency 920) when determining actions by the respective road user for the interaction of the respective road user with the environmental situation of the simulation world. Such static PP properties could be formed, for example, from the following group: age; physical condition; intoxicant influence; driver experience; regulatory compliance; general driving style; Convenience; distractibility. Corresponding values for one or more PP properties can be initialized or set once at the beginning of the simulation and then remain constant during the simulation period.

Details on example implementations of the subjective perception dependency 904 of module 120 on one or more PP properties 162 are in TAB below. 2 described.

TAB. 2: Different examples of PP properties and their influence on the subjective perception, which is determined in module 120. The qualitative dependencies presented are purely illustrative and other qualitative dependencies would be conceivable. Techniques were described above as to how the subjective perception of the surrounding situation in module 120 can be influenced as a function of one or more PP properties 162 . This subjective perception of the surrounding situation can then have several influences on the agent model 105, as shown in FIG. 3 shown. For example, the subjective perception from module 120 may affect the dynamic values of the emotion states 161 (dependency 905). The subjective perception from module 120 may also influence decision making for determining actions, module 130 (dependency 902).

Next, an example of the dependency 905 is explained, ie an influence that the subjective perception of the environmental situation from module 120 can have on the determination of values for one or more emotional states 161 in module 140 .

For example, the subjective perception from module 120 could cause a road user to perceive the perceived environmental situation as an intrusion into an ego area of the respective road user. In particular, actions by other road users can be perceived as an intrusion into their ego. Such an event, meaning the intervention in the respective area of the ego, can then result in a change in the values of one or more specific emotional states, for example increased aggressiveness or increased anxiety. An example would be that other road users come comparatively close to the respective road user. A corresponding distance threshold could be determined depending on PP properties 162 . A corresponding distance threshold value could also be determined depending on the current values of the emotion states 161 . Another example of an intrusion into the ego area would be the use of a headlight flasher or horn by other road users. Here, too, the anticipation of a corresponding action by other road users can be determined as a function of PP properties 162 and/or emotional states 161 .

In general terms, such one or more subjective interventions in the ego area could be selected from the following group: audio signals from another road user; Light signals from another road user; sudden approach of another road user; changes in the performance of a lane change maneuver; inability to pursue a desired speed; Violation of traffic rules by another road user.

Next, further details for determining values of one or more dynamic emotion states 161 are provided in connection with the subsequent figures.

FIG. 4 illustrates aspects in connection with transfer functions 311, 312, which map a respective environmental situation of a road user or a subjective perception of the environmental situation of a road user to values of one or more dynamic emotional states 161. The use of the transfer functions 311, 312 can therefore have the dependency 905 describe. The respective transfer function can be selected in the emotion module 140, depending on the PP properties 162 and according to the dependency 903. In the example of FIG. 4 two transfer functions 311, 312 are shown (continuous and dashed lines), each of which translates a distance 301 to neighboring road users (or a subjective perception of the distance 301) into a value of aggressiveness 161. The two transfer functions 311, 312 are each selected as a function of the PP properties 162 . For example, the transfer function 311, which basically results in a higher level of aggressiveness, could be chosen for a road user who is under the influence of intoxicants; while the transfer function 312 could be chosen for a non-intoxicated road user. In this way, the PP properties 162 can therefore have an influence on the values of one or more emotional states 161 .

While in FIG. 4 shows a one-dimensional transfer function (which translates distance 301 into aggressiveness level), it would generally be possible to use higher-dimensional transfer functions that translate one or more properties of the environmental situation into one or more values of one or more emotional states 161.

In FIG. 4 shows an example in which a certain value for the distance 301 causes a certain level of aggressiveness as an emotion state 161. It would also be possible in various examples as an alternative or in addition for the transfer functions 311, 312 to describe a rate of change of one or more values of subjective emotional states 161 based on a persistence period of the respective environmental situation (or a subjective perception of the respective environmental situation). This therefore means that if the distance 301 falls below a specific threshold value 350 for a longer (shorter) period of time, the aggressiveness level of the respective road user can increase to a greater (lesser) extent. In other words the transfer function could be implemented as first-order differential equations in the period.

In the example of FIG. 4, the transfer functions 311, 312 are designed as linear functions (that is, the aggressiveness increases linearly for decreasing distance 301). There is a threshold distance 350 from which the linear progression begins (“kink”). For example, it would be conceivable that the threshold distance 350 depends on the PP properties 162 . Alternatively or additionally, it would also be conceivable for a basic level 351 to depend on the PP properties. It would also be conceivable that the steepness of the linear course depends on the PP properties.

A time profile 165 of the dynamic emotion states results from such a consideration of the environmental situation—which changes over time for the road users—when determining dynamic values of one or more emotion states 161 . This is in FIG. 5 for the simulation period 250, for the example of aggressiveness.

FIG. 6 illustrates aspects related to determining values for one or more emotion states 161. FIG. 6 particularly illustrates how the accessibility of certain flanges can affect emotional states (see branch 910 in FIG. 3). FIG. 6 schematically illustrates an excerpt from a simulation world 210 with two road users 211, 212. The road users 211, 212 are vehicles that are driving side by side—in adjacent lanes—on a road in the same direction.

A respectively planned trajectory 221, 231 is assigned to both road users 211, 212. The trajectory 221 of the road user 211 includes a planned lane change 272. However, this lane change 272 is one Action that is not possible because otherwise there would be a collision with road user 212 or at least the other road user would be “cut”. An action that corresponds to driving straight ahead 271 along a deviating trajectory 222 (dashed arrow) is therefore primarily possible. This corresponds to an influencing of the trajectory 221 of the road user 211 by the road user 212.

In general terms, it would be possible for the dynamic values of the one or more dynamic emotional states 161 to be determined in the simulation world 210 as a function of the influences on the respectively planned trajectory by an environmental situation of the respective road user. It would therefore be possible for one or more desired lane-changing maneuvers to be determined for each road user based on the respectively planned trajectory, and one or more possible lane-changing maneuvers to be determined based on the dynamic environment situation. Then, the values of the one or more dynamic emotion states may be determined based on a deviation between the one or more desired lane change maneuvers from the one or more possible lane change maneuvers.

In addition to such monitoring of the deviation of possible lane-changing maneuvers from desired lane-changing maneuvers, it would also be conceivable, however, for other target variables to be taken into account as an alternative or in addition, for example a speed. An example would be that a road user (e.g. due to the PP properties 162) would like to drive at a certain speed on the freeway, but cannot do so because another road user is driving at a lower speed on the corresponding one track drives.

Techniques were described above, such as a subjective perception of the surrounding situation based on one or more PP properties 162 can be determined in module 120 (dependency 904 in FIG. 3). It has also been described how values of one or more emotional states 161 can be taken into account when determining the subjective perception of the surrounding situation (dependency 906 in FIG. 3). It has also been described how the values of the one or more emotional states 161 can be influenced as a function of the surrounding situation and in particular of a subjective perception of the surrounding situation (dependency 905 in FIG. 3).

A description is given below of how one or more behaviors of the respective road user can be determined in module 130, for example taking into account the values of one or more emotional states 161 (dependency 925 in FIG. 3) and/or taking into account the static PP-Ei - properties 162 (dependency 920 in FIG. 3).

In principle, a decision-making algorithm can be used to determine one or more changes in the module 130 . Different decision-making algorithms are conceivable. For example, it would be possible to perform an iterative numerical optimization that causes a particular objective function to be maximized or minimized. This target function could depend, for example, on a deviation of the actual trajectory from a planned trajectory. Such numerical iterative optimization algorithms could be implemented, for example, using a simplex method or a genetic optimization algorithm. A gradient descent method could be used.

Another example of a decision making algorithm would be using an artificial neural network. For example, a convolutional network could be used. The different input variables, for example values of one or more emotional states 161 and/or PP properties 162 and/or the perception of the surrounding situation, can be used as different channels are entered into the artificial neural network.

A rule-based procedure would also be possible, for example a benefit-risk assessment. For various possible actions, it can be checked how great the benefit of the respective action is and how great the risk of the respective action is. This can also be referred to as a cost-based formalism. By comparing the benefit with the risk, for example against a predetermined benefit-risk threshold value, it can then be determined which of a number of candidate actions are possible. In principle, it would be possible for the decision-making algorithm to determine the one or more actions based on the values of the one or more dynamic emotion states 161 as boundary conditions, regardless of the specific implementation of the decision-making algorithm. Such boundary conditions can, for example, limit the possible actions actually chosen. Alternatively or additionally, it would also be conceivable for such boundary conditions to meet certain specifications for the basic initial situation under which the decision-making algorithm operates (bias). An example of this is shown in FIG. 7 explained in more detail.

FIG. 7 illustrates an example of a decision making algorithm. In the example of FIG. 7 the decision-making algorithm is implemented rule-based. In the example of FIG. 7, a benefit-risk assessment is carried out. This means that different actions can each be evaluated in terms of benefit. For example, it could be checked whether the planned trajectory can be followed by the respective action or can be followed as well as possible, and based on this check the benefit of the respective action could be determined. A risk assessment can also be carried out for each of the various actions. This means that, for example, it is possible to check the risk with which the respective action will lead to an accident (criticality). The assessment of the criticality of an action could depend on the subjective perception of the surrounding situation by the respective road user, as discussed above. For example, alternatively or additionally, it could be checked with what probability the respective action is perceived by another road user as an intrusion into the ego area of the respective road user. It is then possible for the benefit to be set in relation to the risk. This means that it can be determined, for example, whether a particularly large benefit justifies a particularly large risk. A corresponding threshold value 505, from which the benefit-risk assessment recognizes a specific action as permissible, is shown in FIG. 7 plotted as a function of the aggressiveness level as an example of a dynamic emotional state 161. From FIG. 7 it can be seen that more actions are accepted for a higher level of aggressiveness, i.e. fewer requirements are placed on a risk-benefit ratio.

In more general terms, a risk that is accepted depending on the benefit can depend on the values of the one or more dynamic emotional states 161 of the respective road user.

This could, for example, be concrete using the example of FIG. 6 will be explained. There it could be assessed, for example, that the risk of a lane change maneuver 272 to follow the planned trajectory 221 is comparatively high, because the road user 212 would then be cut off by the road user 211 and under certain circumstances a collision could even occur , for example if road user 212 continues to accelerate. For a traffic participant 211 with a comparatively high level of aggressiveness because of the high benefit and the high risk, a correspondingly predetermined cost-benefit threshold value 505 is nevertheless exceeded, and the lane change maneuver 272 is therefore carried out. The corresponding criticality of such an action can in turn be determined as a function of the subjective perception of the road user in the simulation world 210 .

Various techniques have been described above which make it possible to determine time-varying, ie dynamic, values for emotional states 161 for different road users. In principle, it would be possible for these values to be initialized randomly, e.g. for different road users to be randomly selected. It would also be conceivable that these values are initialized based on the PP properties at the beginning of the simulation period (cf. FIG. 3: dependency 903). For example, a corresponding basic level can be selected. In such a case in particular, it would then be possible for the PP properties for the various road users to be initialized at the start of the simulation as a function of a predefined statistical frequency distribution function 601 (cf. FIG. 8). A corresponding statistical distribution function 601 could be based on empirical values for the frequency of the occurrence of, for example, decisive drivers or drivers under the influence of alcohol.

Of course, the features of the embodiment described above can be combined with one another and aspects of the invention. In particular, the features can be used not only in the combinations described, but also in other combinations or taken on their own, without departing from the field of the invention.

Claims

P AT E N T A N S P R U C H E

1. Computer-implemented method that includes:

- Simulating a dynamic behavior of a large number of road users (211, 212) in a simulation world (210) during a simulation period (250), the simulation comprising:

- for each road user (211, 212) of the plurality of road users (211, 212): determining (140) dynamic values of one or more emotional states (161) during the simulation period (250), and

- For each road user (211, 212) of the plurality of road users (211, 212): determining actions (241, 271, 272) of the respective road user (211, 212) to interact with the simulation world (210) based on ( 925) the dynamic values of the one or more emotion states (161), and

- Based on the dynamic behavior of the large number of road users (211, 212): generating user data (165) for training or for evaluating an automated or autonomous driver assistance functionality of a motor vehicle.

2. Computer-implemented method according to claim 1, wherein the method further comprises:

- Carrying out (3015) the training of the automated or autonomously acting driver assistance functionality of the motor vehicle.

3. A computer-implemented method according to claim 1 or 2, the method further comprising:

- Carrying out (3015) the evaluation of the automated or autonomously acting driver assistance functionality of the motor vehicle. 4. Computer-implemented method according to one of the preceding claims, wherein the simulating further comprises:

- For each road user (211, 212) of the plurality of road users (211, 212): determining (120) a subjective perception of an environmental situation of the respective road user (211, 212) in the simulati onswelt (210) based on (906) the one or more emotional states (161).

5. Computer-implemented method according to any one of the preceding claims, wherein the simulating further comprises:

- For each road user (211, 212) of the plurality of road users (211, 212): determining a subjective perception of an environmental situation of the respective road user (211, 212) in the simulation world (210) as a function of (904). or several corresponding static physiological-psychological properties (162) of the respective road user (211, 212), wherein the dynamic values of the one or more emotional states for each road user are determined based on the respective subjective perception of the surrounding situation.

6. Computer-implemented method according to claim 5, wherein the one or more static physiological-psychological properties (161) are selected from the following group: age, physical condition, influence of intoxicants, driving experience, compliance with rules, general driving style, comfort, distractibility.

7. Computer-implemented method according to one of claims 4 to 6, wherein the dynamic values of the one or more emotional states

(161) in response to one or more subjective interventions by other road users (211, 212) of the plurality of road users (211, 212) in an ego area of the respective road user (211, 212) are determined based on the subjective perception of the surrounding situation.

8. Computer-implemented method according to claim 7, wherein the one or more subjective interventions are selected from the following group: audio signals of another road user; Light signals from another road user; sudden approach of another road user; preventing lane changes from being made; preventing you from driving at the desired speed; Disobeying traffic rules.

9. Computer-implemented method according to any one of the preceding claims, wherein the dynamic values of the one or more emotional states (161) based on (905) a respective environmental situation of the traffic participants (211, 212) in the simulation world and using one or meh reren subjective transfer functions (311, 312), which map the respective dynamic environmental situation (301) of the road users (211, 212) to the dynamic values of the respective one or more emotional states (161), are determined, the one or more subjective transfer functions (311, 312) are determined as a function of one or more static physiological-psychological properties (162) of the respective road user (211, 212).

10. Computer-implemented method according to claim 9, wherein the one or more transfer functions (311, 312) describe a change in the dynamic values of the one or more emotional states (161) based on a persistence period of the respective environmental situation. 11. Computer-implemented method according to one of the preceding claims, wherein each road user (211, 212) is assigned one or more static physiological-psychological properties (162), wherein one or more basic levels (351) of the one or more emotional states (161 ) depend on the one or more static physiological-psychological properties (162). 12. Computer-implemented method according to any one of the preceding claims, wherein each road user (211, 212) of the plurality of road users (211, 212) is assigned a respective planned trajectory (221) in the simulati onswelt (210), wherein the dynamic Values of the one or more emotional states are determined in the simulation world (210) as a function of an influence (222) on the respective planned trajectory (221) by a situation surrounding the respective road user (211, 212). 13. The computer-implemented method of claim 12, wherein the simulating further comprises:

- For each road user (211, 212): determining one or more desired lane-changing maneuvers (272) and/or speeds based on the respective planned trajectory (221) and determining one or more possible lane-changing maneuvers (271) and/or speeds based on the environmental situation, the dynamic values of the one or more emotional states (161) based on a deviation between the one or more desired lane-changing maneuvers (272) and/or speeds of the one or more possible lane change maneuvers (271) and/or speeds are determined.

14. Computer-implemented method according to any one of the preceding claims, wherein each road user (211, 212) is assigned one or more static physiological-psychological properties (162), wherein the determination of the respective actions depends on the corresponding one or more static physiological - psychological properties (162).

15. Computer-implemented method according to one of the preceding claims, wherein the actions of the respective road user (211, 212) are determined based on a decision-making algorithm that determines boundary conditions for the actions based on the dynamic values of the one or more emotional states (161). .

16. Computer-implemented method according to any one of the preceding claims, wherein each road user (211, 212) of the plurality of road users (211, 212) is assigned a respective planned trajectory (221, 231) in the simulation world (210), wherein the determination of the actions (271, 272) for the road users (211, 212) each includes a benefit/risk assessment, the benefit of the benefit/risk assessment depending on a deviation from the planned trajectory (221, 231 ) is determined, whereby the risk of the benefit-risk assessment is determined depending on the criticality of the various actions (271, 272), wherein a risk (505) accepted as a function of the benefit depends on the dynamic values of the one or more emotional states (161) of the respective road user (211, 212).

17. Computer-implemented method according to claim 16, wherein the criticality of the various actions is determined as a function of a subjective perception of the respective road user in the simulation world (210).

18. Computer-implemented method according to one of the preceding claims, wherein the one or more emotional states (161) are selected from the following group: aggression; Fury; confusion; Nervousness; Fear; sadness; Joy; Fatigue; Exhaustion; and stress.

19. Computer-implemented method according to one of the preceding claims, wherein each road user (211, 212) is assigned one or more static physiological-psychological properties (162), the method further comprising:

- Parameterizing the one or more static physiological-psychological properties (162) for the large number of road users (161, 162) as a function of a predetermined statistical frequency distribution function (601).

20. Computer-implemented method according to one of the preceding claims, wherein the useful data (165) include the positions and location of the road users (211, 212). 21. Computer-implemented method according to one of the preceding claims, wherein the user data (165) comprises one or more internal states of an agent model (105) of the various road users (211, 212).