METHOD FOR DETERMINING DRIVING BEHAVIOR AND METHOD FOR ADAPTING CONTROL ALGORITHMS OF AUTOMATED DRIVING SYSTEMS

Description

[0001] The invention relates to a method for determining a driver's driving behavior in relation to a driving behavior of an automated driving system of a vehicle.

[0002] The invention further relates to a method for adapting control algorithms of automated driving systems of vehicles in a vehicle fleet.

[0003] The invention further relates to a device for adapting control algorithms of automated driving systems of vehicles in a vehicle fleet.

[0004] DE 10 2015 218 361 A1 discloses a method for verifying a vehicle function intended to guide a vehicle autonomously in the longitudinal and transverse directions. The method comprises: Determining a test control instruction of the vehicle function to an actuator of the vehicle based on environmental data relating to an environment of the vehicle, determining that a driver of the vehicle issues an actual control instruction that differs from the test control instruction and is actually implemented by the actuator, simulating a fictitious traffic situation that would exist if the test control instruction had been implemented instead of the actual control instruction, based on environmental data, determining whether the fictitious traffic situation represents a relevant event for a road user in the environment of the vehicle or for the vehicle; and providing test data relating to the fictitious traffic situation if it has been determined that the fictitious traffic situation represents a relevant event for a road user in the environment of the vehicle or for the vehicle.

[0005] DE 10 2017 200 180 A1 discloses a method for verifying a vehicle function for at least temporary autonomous longitudinal and / or lateral guidance of a vehicle. This method involves generating a test control command for an actuator based on environmental data during passive operation of the vehicle function, which is not implemented by the actuator. In this case, the actuator implements an actual control command issued by a driver of the vehicle. A discrepancy between the unimplemented test control command of the vehicle function and the driver's actual control command is considered an indication of a fault in the vehicle function.

[0006] US 2015 / 175 168 A1 discloses a system for evaluating the autonomous driving of a vehicle. It involves calculating a score from driving information generated during a driver-controlled journey and from driver assistance information relating to actions that a driver assistance system would have generated if it had controlled the journey.

[0007] DE 10 2019 202 106 A1 discloses a method for validating automated driving functions of a vehicle. It provides that control signals supplied to an actuator for executing the driving function are generated but supplied to a recording device instead of the actuator, that the driving function is simultaneously performed by a driver using their own control signals, and that the recording device stores the control signals if they differ from one another.

[0008] From DE 10 2017 126 021 A1 a method for controlling a vehicle during an automated journey is known, in which it is provided that an optimal trajectory for the automated journey is determined by means of an optimization method based on a cost function.

[0009] EP 3 428 031 A1 discloses a device for designing a driver assistance system, with which potential design errors are identified by evaluating detected driving anomalies.

[0010] The invention is based on the object of providing a novel method for determining a driver's driving behavior in relation to the driving behavior of an automated driving system of a vehicle. The invention is further based on the object of providing a novel method for adapting control algorithms of automated driving systems of vehicles in a vehicle fleet. The invention is further based on the object of providing a novel device for adapting control algorithms of automated driving systems of vehicles in a vehicle fleet.

[0011] The object is achieved according to the invention by a method for determining a driving behavior, which has the features specified in claim 1, by a method for adapting control algorithms of automated driving systems, which has the features specified in claim 6, and by a device for adapting control algorithms of automated driving systems, which has the features specified in claim 7.

[0012] Advantageous embodiments of the invention are the subject of the subclaims.

[0013] In the method for determining a driver's driving behavior in relation to a driving behavior of an automated driving system of a vehicle, according to the invention, driver control commands and a manually driven trajectory of the vehicle are recorded during manual driving of the vehicle. In a first calculation path, it is determined which automated control commands the automated driving system would generate at a current actual position of the vehicle on the manually driven trajectory if the automated driving system were active. In a second calculation path, a trajectory of the vehicle is simulated that the vehicle would travel with the automated driving system active.Depending on the recorded control commands of the driver, the automated control commands, the manually driven trajectory and / or the simulated trajectory, a score value is determined as a measure of a discrepancy between the driving behavior of the driver and the driving behavior of the automated driving system, wherein the score value is determined as a function of a probability that the automated driving system would have determined a trajectory as the target trajectory at an earlier point in time that at least substantially corresponds to the manually driven trajectory.

[0014] In the development of automated, particularly highly automated or autonomous vehicles, approaches to optimizing automated driving systems using a so-called imitation process are well known. This involves assuming a driver's driving style as ideal, and recording their behavior is used to train a behavior algorithm. The goal of this algorithm is to imitate the driver's driving behavior. In future generations of automated driving systems, the performance of the driving systems will exceed the performance of safe average drivers. Using this method, any resulting discrepancy between the driver's driving behavior and that of the automated driving system can be reliably and precisely determined.

[0015] This allows a distinction to be made between different functional scopes of the automated driving system, for example, an entry-level variant and a variant with a full range of functions. This allows a comprehensive correlation to be established between the driving behavior of automated driving systems with different functional scopes and the driving behavior of the driver. The method enables the driving behavior to be calculated in metrics in order to create a classification of the driver's driving behavior. A record can be created between the capabilities of the automated driving system and those of the driver, which allows the discrepancy to be calculated, particularly for each driving situation. This calculated discrepancy can be used to generate an overview for the driver.

[0016] The precise calculation and determination of the discrepancy between the driver's respective driving behavior and that of the automated driving system is required for the following applications. For example, if the driver only uses a reduced basic version or entry-level variant of the automated driving system, it is possible to calculate in which driving situations they could use a more fully-fledged driving system, i.e., a driving system with a wider range of functions. It is also possible to show in which traffic situations or on which routes the driving system would have driven more safely or equivalently, in order to strengthen confidence in the automated driving systems. Furthermore, the information about a driver's ability can be used to optimize so-called safety assistance systems for the respective driver. The complete information about the performance of the driver and the system can be accumulated and used for a scoring system.

[0017] In one possible embodiment of the method, the score value is determined based on the deviation between the driver's recorded control commands and the automated control commands. This allows for a simple and reliable determination of the score value.

[0018] In another possible embodiment of the method, the score value is determined based on the deviation between the manually driven trajectory and the simulated trajectory. This also enables a simple and reliable determination of the score value.

[0019] In another possible embodiment of the method, the score value is determined based on the time period or distance the vehicle can travel until the deviation between the manually driven trajectory and the simulated trajectory reaches a predetermined threshold. This also enables a simple and reliable determination of the score value, whereby a time point and / or a position of the deviation between the manually driven trajectory and the simulated trajectory is determined, so that positions at which a statistically relevant discrepancy between the driver's driving behavior and the driving behavior of the automated driving system frequently occurs can be identified.

[0020] In another possible embodiment of the method, the deviation between the manually driven trajectory and the simulated trajectory is determined using at least one cost function. Different cost functions can be used individually or in combination, with the cost functions taking into account, for example, a quadratic error, an absolute error, a speed-weighted absolute error, a cumulative speed-weighted absolute error, a quantized classification error, and / or a threshold-based relative error. Using the at least one cost function, the deviation and the resulting score value can be determined easily and reliably, and the respective cost function can be easily adapted to the complexity of the task by extension.

[0021] In another possible embodiment of the method, the score value is determined as a summed score value based on a summation, for example, a weighted summation, of several score values, each determined according to one of the aforementioned embodiments. This summed score value indicates, in particular, how much the driver's driving style deviates from the driving style of the automated driving system.

[0022] This information can be used to determine the driver's current driving performance, particularly to determine whether the driver is driving in a distracted and / or unsafe manner. This allows the vehicle's active safety systems to be adapted to a potentially reduced driver performance. The driver can also be offered the option of activating an automated driving system because it is better able to handle a current driving task than the driver himself.

[0023] In one possible embodiment, the method is applied to a plurality of vehicles in a fleet to determine a respective score value. The determined score values ​​are collected and used to adapt control algorithms of automated driving systems of vehicles in a fleet. The score values ​​are determined by the vehicles and each indicate a measure of a discrepancy between the driving behavior of a driver of the vehicle and the driving behavior of an automated driving system at a respective position in the vehicle. The collected score values ​​are statistically evaluated, and the statistical evaluation identifies positions at which a statistically relevant discrepancy between the driver's driving behavior and the driving behavior of the automated driving system occurs frequently.Based on a determined cluster of statistically relevant discrepancies, parameters of the automated driving system's control algorithms are adjusted to reduce the discrepancies, and the adjusted parameters are made available to the vehicles in the fleet. This enables simple and reliable adjustment of the automated driving system's control algorithms, thus improving the performance and reliability of the automated driving systems.

[0024] Embodiments of the invention are explained in more detail below with reference to drawings.

[0025] Showing: Fig. 1 schematically shows a block diagram of an embodiment of a device for automated driving of a vehicle, Fig. 2 schematically shows a block diagram of a further embodiment of a device for automated driving of a vehicle, Fig. 3 schematically shows a block diagram of a further embodiment of a device for automated driving of a vehicle, Fig. 4 schematically shows a block diagram of a device for determining a driver's driving behavior in relation to a driving behavior of an automated driving system of a vehicle, Fig. 5 schematically shows a plan view of a traffic situation, Fig. 6 schematically shows a real trajectory of a vehicle during manual driving and a trajectory simulated during this driving, Fig. 7 schematically shows a real trajectory of a vehicle during manual driving and trajectories simulated during this driving, Fig.Fig. 8 schematically shows a real trajectory of a vehicle during manual driving and trajectories simulated during this driving operation, and Fig. 9 schematically shows a resimulation of a probability for a real trajectory of a vehicle.

[0026] Corresponding parts are provided with the same reference numerals in all figures.

[0027] In Figure 1 is a block diagram of a possible embodiment of a device 1 for automated, in particular highly automated or autonomous, driving operation of a Figure 5 shown vehicle 2.

[0028] Vehicle 2 is equipped with fully-fledged hardware for automated driving, although the hardware's performance can be limited by software at various levels. This can be achieved in at least two levels, for example, a basic level and a high level. These levels differ in their predefined performance, which is characterized by their ability to handle different use cases.

[0029] The device 1 comprises a plurality of sensors 3.1 to 3.x designed for environmental detection and at least two computing units 4, 5 designed at a so-called control unit level, each of which forms a redundancy for the other computing unit 4, 5. Furthermore, the device 1 comprises a comparator 6 and actuators 7.1 to 7.z.

[0030] Sensors 3.1 to 3.x are designed to detect the vehicle environment.

[0031] The computing units 4, 5 are independent of each other for determining and generating trajectory trees with several possible trajectories β, β' (shown in Figure 5) and select a target trajectory from a trajectory tree using an optimization algorithm. Furthermore, the computing units 4, 5 are designed independently of one another to determine and generate driving commands for controlling the vehicle 2 in automated driving mode based on sensor data acquired by the sensors 3.1 to 3.x along the selected target trajectory and, by creating redundancy, enable the implementation of specified safety requirements for the automated driving operation of the vehicle 2, in particular a vehicle 2 with a high degree of automation. The computing units 4, 5 are each located in a computing path R1, R2.

[0032] The actuators 7.1 to 7.z are designed to execute the travel commands generated by the computing units 4, 5.

[0033] The computing units 4, 5 are coupled to all sensors 3.1 to 3.x, and the comparator 6 is coupled to both computing units 4, 5. The travel commands calculated from the sensor data by both computing units 4, 5 can thus be made available to the comparator 6 for comparison with the calculations redundantly performed by the computing units 4, 5. Using generally known safety algorithms, the comparator 6 determines a safe travel command or a safe control signal from its input data and transmits this to the actuators 7.1 to 7.z. The comparator 6 is hardware-independent, for example, designed according to the ISO 26262 standard in ASIL-D (ASIL = Automotive Safety Integrity Level).

[0034] Figure 2 shows a block diagram of another possible embodiment of a device 1 for automated driving of the vehicle 2.

[0035] In contrast to the Figure 1 In the exemplary embodiment shown, the computing units 4, 5 and the comparator 6 are formed in a common control unit 8 and are formed from at least two redundantly operating, mutually independent computing cores at chip level, for example separate semiconductors, processors, systems on chip, etc.

[0036] In Figure 3 a block diagram of a further possible embodiment of a device 1 for automated driving of the vehicle 2 is shown.

[0037] In contrast to the Figure 2 In the exemplary embodiment shown, the computing units 4, 5 in the common control unit 8 are designed on a common computing core 9 as a redundant multi-core architecture as a system on chip with a common input / output 10, which receives the sensor data and outputs it to the computing units 4, 5, and a communication unit 11, also referred to as inter-fabric.

[0038] Each computing unit 4, 5 comprises in particular a CPU 4.1, 5.1, a GPU 4.2, 5.2, an NPU 4.3, 5.3 and a memory 4.4, 5.4.

[0039] To implement a method for determining a driver's driving behavior in relation to a driving behavior of an automated driving system of the vehicle 2, the redundancy present in the device 1 is utilized and modified according to the following explanations. The explanations apply to all previously mentioned and other suitable embodiments of the device 1.

[0040] The aim of the method is to determine and quantify the driver's performance related to a driving task using the device 1 for the automated driving operation of the vehicle 2.

[0041] Figure 4shows a block diagram of a device 12 for determining a driving behavior of a driver in relation to a driving behavior of an automated driving system of a vehicle 2, wherein the device 12 is a device 1 for automated driving operation of the vehicle 2 according to the Figures 1 to 3 includes.

[0042] The device 12 is designed such that, during manual driving of the vehicle 2 performed by the driver, an assisting system (basic expansion level) or highly automated system (high expansion level) implemented by the device 1 is fully executed in the background, with the exception of the control of the actuators 7.1 to 7.z. This means that all sensors 3.1 to 3.x and other hardware and software algorithms are executed as if they were controlling the vehicle 2 in automated operation. Only the control itself is not executed. Rather, a drive and the actuators 7.1 to 7.z are not controlled by the device 1 but by the driver. In such background or passive operation, the control of the vehicle 2 is thus subject to the driver, who transmits control commands to the actuators 7.1 to 7.z and the drive via a steering system, pedals, and possibly other input devices.In a sum of scenarios, the driver's driving style will differ from that of device 1.

[0043] In passive operation, the device 1 calculates in real time by means of the first computing unit 4 in the first computing path R1 in a signal formation S1 setting and steering commands as well as in Figure 5 trajectories β, β' shown in more detail. Input signals for this are provided by sensors 3.1 to 3.x, for example radar sensors, lidar sensors, cameras, inertial measuring units, position detection systems, etc. The device 1 localizes itself in real time and calculates the control commands for a steering system, a brake, and a drive system of the vehicle 2 and the planned trajectory β, β'.

[0044] This means that in the first calculation path R1 it is determined which automated control commands the automated driving system should execute at a current actual position of the vehicle 2 on a Figure 5manually driven trajectory r shown in more detail if the automated driving system were active.

[0045] Results of signal generation S1 are forwarded to comparator 6. This occurs at each time step from the beginning according to a permanent initialization.

[0046] Since the system cannot follow its planned trajectory β, β' due to passive operation, a difference exists between the positions actually driven by the driver and the positions expected by the system. This error would accumulate over time and lead to system failure. For this reason, the system corrects the real-time information at each time step in the first calculation path R1 and adapts the newly generated control commands according to the actual driving situation.

[0047] Because no redundancy is required in passive mode, the second computing unit 5 is used to simulate a trajectory β, β' of the vehicle 2 in the second computing path R2 in a simulation mode S2, as if the vehicle 2 had traveled the trajectory β, β' determined in the first computing path R1, not the driver's trajectory, starting from a starting point. The simulation results are also forwarded to the comparator 6.

[0048] This means that in the second calculation path R2, a trajectory β, β' of the vehicle 2 is simulated, which the vehicle 2 would travel if the automated driving system were active.

[0049] A summed error resulting from a difference between the actual trajectory r and the trajectory β, β' determined in the first calculation path R1 is compensated up to a threshold value by a simulated input and then compared with the driver's performance.

[0050] Within comparator 6, an algorithm calculates score values ​​SW1 to SW4 for the driver and returns the results to the two computing units 4, 5. A calculation S3 of the score values ​​SW1 to SW4 is performed depending on the recorded driver commands SB and navigation data ND, the automated commands, the manually driven trajectory r, and the simulated trajectory β, β', with the score values ​​SW1 to SW4 indicating a measure of a discrepancy between the driver's driving behavior and the driving behavior of the automated driving system. This means that, in particular, implemented manual commands and unimplemented automated commands, as well as the driven trajectory r and the simulated trajectory β, β', are evaluated to determine the extent to which the driver's driving behavior deviates from the driving behavior that would result during automated driving in the same environment.

[0051] Furthermore, the score values ​​SW1 to SW4 determined by means of the comparator 6 are transmitted to a backend 13. In this backend, score values ​​SW1 to SW5 of a plurality of vehicles 2 of a vehicle fleet are collected, wherein the score values ​​SW1 to SW4 are determined in particular according to the previous description of the vehicles 2. Furthermore, the collected score values ​​SW1 to SW4 are statistically evaluated, wherein the statistical evaluation determines positions at which a statistically relevant discrepancy between the driver's driving behavior and the driving behavior of the automated driving system occurs frequently. Based on a determined accumulation of statistically relevant discrepancies, parameters P of the control algorithms of the automated driving systems are then adjusted to reduce the discrepancies, and the adjusted parameters P are made available to the vehicles 2 of the vehicle fleet.

[0052] In Figure 5a plan view of a traffic situation with a vehicle 2 approaching an obstacle 14 is shown.

[0053] Vehicle 2 will be described as Figure 3 in manual driving mode controlled by the driver and actually follows a trajectory r.

[0054] At the same time, the device 12, also analogous to the description according to Figure 3 , in the basic expansion stage a trajectory β' and in the high expansion stage, i.e. as a highly automated system, the trajectory β is calculated, whereby the highly automated system is capable of carrying out more complex maneuvers than a system in the basic expansion stage.

[0055] Figure 6 shows a real driven trajectory r of a vehicle 2 during manual driving of the same and a trajectory β 0 simulated during this driving operation at different times t 0 to t 2 .

[0056] To determine a first score value SW1, the first computing unit 4 in the first computing path R1 determines a trajectory ß 0 as a target trajectory at a first time t 0 , starting from a current vehicle position at this time t 0 . The trajectory r lies in the future and is not available as information. Furthermore, the computing unit 4 determines the control commands that must be generated at the current vehicle position for the actuators 7.1 to 7.z if the vehicle 2 is to be guided along the trajectory ß 0 .

[0057] The deviation between the manual positioning commands SB generated by the driver at the actual vehicle position and the automated positioning commands generated by the first computing unit 4 at the same position is then determined. Thus, the deviation between what the driver does at the actual vehicle position and what the first computing unit 4 would do at the actual vehicle position is determined.

[0058] In the simplest case, the deviation can be determined by calculating the difference. If there are multiple deviation values, an average deviation can be determined by calculating the mean. For example, it is possible to weight a deviation between control commands for a longitudinal movement of the vehicle 2 and a deviation between control commands for a lateral movement of the vehicle 2 differently when calculating the mean, in particular to weight them depending on the speed. The averaging can involve calculating an arithmetic mean or a root mean square.

[0059] The first score value SW1 is calculated for the determined deviation, for example, using a predefined look-up table. This first score value SW1 is larger the larger the determined deviation.

[0060] The first score value SW1 is thus a measure of the deviation of the manual control commands SB from the automated control commands that the automated driving system would execute if it were active.

[0061] This described procedure for determining the first score value SW1 is repeated cyclically.

[0062] In Figure 7 a real driven trajectory r of a vehicle 2 during manual driving of the same and simulated trajectories β 0 , β 1 , β 2 at different times t 0 to t 2 are shown.

[0063] The actual vehicle position at the first time t 0 is the starting position for the simulation performed in the second computing unit 5. At the first time t 0 , the trajectory ß 0 , formed as a target trajectory, along which the vehicle 2 is to move, is determined.

[0064] The simulation determines a simulated position at which vehicle 2 will be located one time step later, at time t 1 . At time t 1 , the deviation between the actual vehicle position at this time t 1 and the simulated position determined for this time t 1 is determined.

[0065] At time t 1 , a new trajectory ß 2 , designed as a target trajectory, is then calculated, which theoretically can correspond to the previously determined trajectory ß 0 . The calculation is based on the new sensor data recorded at this time t 1 and on the assumption that the vehicle 2 is actually at the simulated position at time t 1 .

[0066] These calculation steps are repeated for subsequent time steps until the determined deviation exceeds a specified threshold ξ. This is the case, for example, at a time tx. One time step later, at time t x+1 , the simulation is restarted with the then-current actual vehicle position as the new starting position. This is necessary because a drift between the simulated position and the real sensor data leads to a parallax that can no longer be compensated for beyond the threshold ξ.

[0067] The deviations determined for the times t 1 to tx are evaluated with a cost function Ψ according to the following table, for example with the mean square error function: Table 1 Metric name Parameters Metric definition Squared error - 1 V ∑ i ∈ V a i − a ^ i 2 Absolute error - 1 V ∑ i ∈ V a i − a ^ i 1 Speed-weighted absolute error - 1 V ∑ i ∈ V a i − a ^ i 1 v i Cumulative speed-weighted absolute error T 1 V ∑ i ∈ V ∑ t = 0 T a i + t − a ^ i + t v i + t 1 Quantized classification error σ 1 V ∑ i ∈ V 1 − δ Q a i σ , Q a ^ i σ Thresholded relative error α 1 V ∑ i ∈ V θ a ^ i − a i − α a i

[0068] For the result obtained, a second score value SW2 is determined, for example using a look-up table.

[0069] The second score value SW2 is thus a measure of the deviation between the actually driven trajectory r and the trajectory β 0 , β 1 , β 2 simulated in the second computing unit 5.

[0070] The second score value SW2 is stored, for example, in a matrix, where all trajectories β x to r over all tx and the dependent cost function Ψ are the entries. The summation of all matrices to the second score value SW2 according to Ψ β 0 ∑ Ψ ≥ ξ t i ⋮ ⋮ Ψ β i ∑ Ψ ≥ ξ t i is used to determine the overall performance and transmitted to the backend 13. The more divergent the trajectories β x are from the actual trajectory r, i.e. the more the cost function Ψ increases, the worse the second score value SW2 becomes for the driver.

[0071] In Figure 8a real driven trajectory r of a vehicle 2 during manual driving of the same and simulated trajectories β 0 , β 1 , β 2 at different times t 0 to t 2 are shown.

[0072] If the threshold ξ exceeds the cost function Ψ for the trajectory β 0 , a current simulation cycle in the second calculation path R2 is terminated. The reached time step duration is incorporated into a third score value SW3. The longer the trajectory β 0 remains below the threshold ξ, the better the third score value SW3 is for the driver.

[0073] This means that it is determined how long or how far vehicle 2 can travel from the first time t0 until the deviation between the actual vehicle position and the position simulated by the second computing unit 5 exceeds the specified threshold ξ. Thus, the time elapsed between times t0 and tx is determined, or the distance traveled by vehicle 2 during this time period. The third score value SW3 is determined for the determined time period or distance, for example, using a look-up table.

[0074] The third score value SW3 is thus a measure of the time or distance that vehicle 2 can travel until the deviation between the manually driven trajectory r and the simulated trajectory ß x reaches the specified threshold ξ.

[0075] Before the simulation restarts in the second calculation path R2, a re-simulation takes place at the time tx at which the specified threshold ξ for the simulated trajectory ß x was exceeded according to Figure 9 . Here, the real vehicle position and a real vehicle state at time tx , in this case at time t 2 , are used to calculate back which conditions would have had to be met for the system to have moved from a real state at time t 2 to a state in automated operation at the same time t 2 .

[0076] In the decision space at time t 0 , all potentially possible trajectories β x receive a probability weight. The system selects the trajectory β x with the best probability, which is output as trajectory β 0 .

[0077] All potentially viable trajectories β x span a decision space of the system. Statistically improbable trajectories β x are rejected.

[0078] To calculate a fourth score value (SW4), the probability weighting of the actual trajectory r in the decision space is calculated retrospectively. This means the system originally calculated the probability of the driver's trajectory r before it was driven. The worse the weighting of the actual trajectory r, the worse the fourth score value (SW4) for the driver.

[0079] This means that if the specified threshold ξ is exceeded at time tx, the probability with which the first computing unit 4 would have selected a trajectory βx from the trajectory tree as the target trajectory at an earlier time t0, t1, t2, which essentially corresponds to the actually traveled trajectory r, is retrospectively determined. The fourth score value SW4 is determined for the determined probability, for example, using a look-up table.

[0080] The four determined score values ​​SW1 to SW4 are then summed. A weighted summation is also conceivable. The summed score value indicates how much the driver's driving style differs from that of the automated driving system. This information can be used to determine the driver's current driving performance, in particular to determine whether the driver is driving in a distracted and unsafe manner, and to adapt active safety systems to a potentially reduced driver performance, or to offer the driver the option of activating the automated driving system because the system is currently driving more safely than the driver.

[0081] The summed score value can also be sent to backend 13 along with the position at which it was determined. Backend 13 can then use statistical analysis to determine whether there is a cluster of discrepancies between the driving styles of many drivers and the driving style of the automated driving system at a specific position, in order to improve the algorithms of the automated driving system if necessary.

[0082] For example, if it is determined that the driving behavior of drivers at a roundabout often differs significantly from the driving behavior of the automated driving system, then this information can be used as an indication that optimizing the driving system's algorithms for operation at this roundabout would be useful.

[0083] If the driver is competent, the score values ​​SW1 to SW4 can be used to determine whether the automated driving system performs as well as the competent driver or worse than the competent driver. The score values ​​SW1 to SW4 can then be used to optimize the algorithms of the automated driving system so that the driving behavior of the automated driving system is aligned with that of the competent driver.

Claims

1. Method for determining a driving behavior of a driver in relation to a driving behavior of an automated driving system of a vehicle (2), characterized in that during a manual driving operation of the vehicle (2) - actuating commands (SB) from the driver and a manually driven trajectory (r) of the vehicle (2) are detected, - in a first calculation path (R1), it is determined which automated actuating commands the automated driving system would generate at a current actual position of the vehicle (2) on the manually driven trajectory (r) if the automated driving system were active, - in a second calculation path (R2), a trajectory (β, β', β0 to β2) of the vehicle (2), which the vehicle (2) would cover when the automated driving system is active, is simulated, and - depending on the detected actuating commands (SB) from the driver as well as the automated actuating commands and / or the manually driven trajectory (r) as well as the simulated trajectory (β, β', β0 to β2), at least one score value (SW1 to SW4) is determined as a measure of a discrepancy between the driving behavior of the driver and the driving behavior of the automated driving system, the score value (SW1 to SW4) being determined based on a probability that the automated driving system would have, at an earlier point in time (t0, t1, t2), specified a trajectory (β, β', β0 to β2) which at least substantially corresponds to the manually driven trajectory (r) as the target trajectory.

2. Method according to claim 1, characterized in that the score value (SW1 to SW4) is determined depending on a deviation between the detected actuating commands (SB) from the driver and the automated actuating commands.

3. Method according to either claim 1 or claim 2, characterized in that the score value (SW1 to SW4) is determined depending on a deviation between the manually driven trajectory (r) and the simulated trajectory (β, β', β0 to β2).

4. Method according to claim 3, characterized in that the score value (SW1 to SW4) is determined depending on a period of time or distance that the vehicle (2) can travel until the deviation between the manually driven trajectory (r) and the simulated trajectory (β, β', β0 to β2) reaches a specified threshold (ξ).

5. Method according to either claim 3 or claim 4, characterized in that the deviation between the manually driven trajectory (r) and the simulated trajectory (β, β', β0 to β2) is determined using at least one cost function (Ψ).

6. Method according to any of the preceding claims, characterized in that it is used in a plurality of vehicles (2) of a vehicle fleet to determine a corresponding score value (SW1 to SW4) and in that the determined score values (SW1 to SW4) are used to adjust control algorithms of automated driving systems of the vehicles (2) of the vehicle fleet by - collecting the score values (SW1 to SW4) of the plurality of vehicles (2) of the vehicle fleet, the score values (SW1 to SW4) being determined by the vehicles (2) and each indicating a measure of a discrepancy between a driving behavior of a driver of the vehicle (2) and a driving behavior of an automated driving system at a corresponding position of the vehicle (2), - statistically evaluating the collected score values (SW1 to SW4) and determining, in the statistical evaluation, positions at which there is frequently a statistically relevant discrepancy between the driving behavior of the driver and the driving behavior of the automated driving system, - adjusting parameters (P) of the control algorithms of the automated driving systems to reduce the discrepancies based on a determined accumulation of statistically relevant discrepancies, and - making the adjusted parameters (P) available to the vehicles (2) of the vehicle fleet.

7. Device (12) designed to carry out a method according to claim 6.