Computer-implemented method and test unit for approximating test results and method for providing a trained artificial neural network

Description

[0001] The present invention relates to a computer-implemented method for approximating test results of a virtual test of a device for at least partially autonomously controlling a motor vehicle.

[0002] The present invention further relates to a computer-implemented method for providing a trained, artificial neural network for approximating test results of a virtual test of a device for at least partially autonomously controlling a motor vehicle.

[0003] The invention further relates to a test unit for approximating test results of a virtual test of a device for the at least partially autonomous control of a motor vehicle. The present invention further relates to a computer program and a computer-readable data carrier. State of the art

[0004] Driver assistance systems such as adaptive cruise control and / or functions for highly automated driving can be verified or validated using various testing methods. In particular, hardware-in-the-loop methods, software-in-the-loop methods, simulations, and / or test drives can be used.

[0005] The effort, in particular the time and / or cost, required to test such vehicle functions using the above-mentioned testing methods is typically very high, since a large number of potentially possible driving situations must be tested.

[0006] This can lead to high costs, especially for test drives and simulations.

[0007] The paper "Model predictive control for adaptive cruise control with multi-objectives: comfort, fuel-economy, safety and car-following" by LI-HUA LUO ET AL. describes an "Adaptive Cruise Control" (ACC) algorithm based on a Model Predictive Control (MPC) framework. While MPC provides approximations regarding the design of driving functions, it does not specify a method that specifically considers and reduces the time and / or cost required to test such vehicle functions.

[0008] The paper "Representation of 3-D mappings for automotive control applications using neural networks and fuzzy logic" by HOLZMANN H ET AL. shows the modeling and simulation of an engine using a multi-layer perceptron and a radial basis function network.

[0009] Furthermore, the paper "Model Predictive Multi-Objective Vehicular Adaptive Cruise Control" by SHENGBO LI ET AL. also discloses an "Adaptive Cruise Control" (ACC) algorithm based on a Model Predictive Control (MPC) framework.

[0010] DE 10 2017 200 180 A1 specifies a method for verifying and / or validating a vehicle function intended to guide a vehicle autonomously in the longitudinal and / or transverse direction.

[0011] The method comprises determining, on the basis of environmental data relating to an environment of the vehicle, a test control instruction of the vehicle function to an actuator of the vehicle, wherein the test control instruction is not implemented by the actuator.

[0012] The method further comprises simulating, on the basis of environmental data and using a road user model relating to at least one road user in the environment of the vehicle, a fictitious traffic situation that would exist if the test control instruction had been implemented.

[0013] The method further comprises providing test data relating to the fictitious traffic situation. The vehicle function is operated passively in the vehicle to determine the test control instruction.

[0014] The disadvantage of this method is that in order to verify and / or validate the vehicle function, actual operation of the vehicle is required to determine the required data.

[0015] Consequently, there is a need to improve existing procedures and test facilities so that so-called test cases can be determined efficiently within the framework of scenario-based testing for systems and system components in highly automated driving.

[0016] It is therefore an object of the invention to provide a method, a test unit, a computer program and a computer-readable data carrier which can efficiently determine test cases within the framework of scenario-based testing for systems and system components in highly automated driving. Disclosure of the invention

[0017] The object is achieved according to the invention by a computer-implemented method for approximating test results of a virtual test of a device for the at least partially autonomous control of a motor vehicle according to patent claim 1, a computer-implemented method for providing a trained, artificial neural network for approximating test results of a virtual test of a device for the at least partially autonomous control of a motor vehicle according to patent claim 9, a test unit for approximating test results of a virtual test of a device for the at least partially autonomous control of a motor vehicle according to patent claim 12, a computer program according to patent claim 14 and a computer-readable data carrier according to patent claim 15.

[0018] The invention relates to a computer-implemented method for approximating test results of a virtual test of a device for at least partially autonomous control of a motor vehicle.

[0019] The method comprises receiving a first data set comprising a plurality of driving situation parameters consisting of environmental parameters describing the surroundings of the motor vehicle and ego parameters describing the state of the motor vehicle, wherein each driving situation parameter has a predetermined definition range.

[0020] The method further comprises approximating a numerical value range of a target function for the definition domain of at least one driving situation parameter using a trained artificial neural network applied to the plurality of driving situation parameters, wherein the target function represents a driving maneuver of a motor vehicle in response to a control of the device generated by the plurality of driving situation parameters of the definition domain. The method further comprises providing a second data set consisting of the numerical value range of the target function and the associated definition domain of the at least one driving situation parameter.

[0021] The present invention is thus advantageously capable of replacing the simulation of driving situations and / or driving situation parameters with the aid of an artificial neural network.

[0022] Within the scope of the present method, an artificial neural network is advantageously used, whose task is to approximate test results. These test results are test results of interest that are to be the subject of virtual tests of a device, e.g., a control unit, for the autonomous control of a motor vehicle.

[0023] In scenario-based testing of systems and system components for autonomous driving of a motor vehicle, scenarios are defined that can be described as an abstraction of a traffic situation. A logical scenario is the abstraction of a traffic situation with the road, driving behavior, and surrounding traffic without specific parameter values.

[0024] By selecting specific parameter values, the logical scenario becomes a concrete scenario. Each concrete scenario corresponds to a specific traffic situation.

[0025] An autonomous driving function is implemented by a system, such as a control unit. The control unit is typically tested in real traffic situations in the real vehicle or, alternatively, validated through virtual tests.

[0026] In this context, the present method approximates test results or traffic situations of interest within a predefined definition range of the driving situation parameters used. The definition range of the driving situation parameters is chosen to cover, for example, safety-relevant driving situations such as collisions or near-collisions between the vehicles involved.

[0027] The test results approximated in this way can then advantageously be validated within the framework of virtual tests of the control unit, so that the method according to the invention enables efficient virtual validation of control units for the autonomous control of motor vehicles.

[0028] Further embodiments of the present invention are the subject of the further subclaims and the following description with reference to the figures.

[0029] According to one aspect of the invention, the method according to the invention further comprises that the ego parameters comprise a speed of the motor vehicle and the environmental parameters comprise a speed of another motor vehicle and a distance between the motor vehicle and the other motor vehicle.

[0030] Using these parameters, a so-called cut-in scenario can be advantageously approximated. A cut-in scenario can be described as a traffic situation in which a highly automated or autonomous vehicle is driving in a given lane and another vehicle, traveling at a lower speed than the ego vehicle, merges from another lane into the ego vehicle's lane at a certain distance.

[0031] The speed of the ego vehicle and the other vehicle, also called the fellow vehicle, remains constant. Since the speed of the ego vehicle is higher than that of the fellow, the ego vehicle must be decelerated to avoid a collision between the two vehicles.

[0032] On the basis of the above-mentioned ego parameters and environmental parameters, the method according to the invention can thus approximate relevant traffic situations in the predetermined definition range of the above-mentioned parameters.

[0033] According to a further aspect of the invention, the method according to the invention further comprises that the objective function is a safety objective function which has a numerical value which has a minimum value at a safety distance between the motor vehicle and the further motor vehicle of ≥ V FELLOW x 0.55, has a maximum value in the event of a collision between the motor vehicle and the further motor vehicle, and has a value which is greater than the minimum value at a safety distance between the motor vehicle and the further motor vehicle of ≤ V FELLOW x 0.55.

[0034] The safety objective function indicates how safe the traffic situation is for the ego vehicle. It is specified as follows: If the distance between the ego vehicle and the fellow vehicle is always greater than or equal to the safety distance, the function value of the safety objective function is 0.

[0035] The safety distance can be defined as the distance at which, depending on the speed difference between the ego vehicle and the fellow vehicle as well as the distance between the ego vehicle and the fellow vehicle, a safe braking of the ego vehicle is always possible without the occurrence of a collision with the fellow vehicle.

[0036] In this example, such a distance is defined by a value in meters corresponding to the speed V FELLOW x 0.55.

[0037] As the distance between the ego vehicle and the fellow vehicle decreases, or as the safety distance is exceeded, the objective function value increasingly approaches 1. If a collision occurs between the ego vehicle and the fellow vehicle, the distance between the ego vehicle and the fellow vehicle is therefore less than or equal to 0, and the objective function value is 1.

[0038] According to a further aspect of the invention, the method according to the invention further comprises that the target function is a comfort target function or an energy consumption target function which has a numerical value which has a minimum value in the case of no change in the acceleration of the motor vehicle, has a maximum value in the case of a collision between the motor vehicle and the other motor vehicle, and has a value between the minimum value and the maximum value in the case of a change in the acceleration of the motor vehicle depending on an amount of the change in the acceleration.

[0039] Using the comfort target function, statements can be made about how comfortable a driving maneuver is for the driver of the ego vehicle. Sharp acceleration or deceleration, and frequent repetition of these maneuvers, are considered uncomfortable.

[0040] The changes in acceleration are referred to as jerk. The smaller the calculated value of the comfort objective function, the more comfortable the driving situation. The energy consumption in the event of a collision between the ego vehicle and the fellow vehicle is a fixed maximum value. The reason for this is that the fuel tank in the case of a motor vehicle with an internal combustion engine and / or a traction battery in the case of an electric or hybrid vehicle can usually no longer be used in an accident.

[0041] With regard to the cut-in scenario, possible safety-relevant test cases lie between collision and non-collision cases, which can be defined based on the respective objective functions, i.e. the safety objective function, the comfort objective function and the energy consumption objective function.

[0042] According to a further aspect of the invention, the method according to the invention further comprises that the plurality of driving situation parameters, in particular the speed of the motor vehicle and the speed of the further motor vehicle, are generated within the predetermined definition range by a random algorithm.

[0043] Thus, the majority of driving situation parameters that form the data set used to approximate the test results can be generated easily and time-efficiently.

[0044] According to a further aspect of the invention, the method according to the invention further comprises using a separate artificial neural network to approximate the numerical value range of each objective function, wherein individual hyperparameters of each artificial neural network are stored in a database.

[0045] By using a separate artificial neural network to approximate each objective function, the artificial neural network used can be specifically trained for the best possible approximation of the respective objective function and is thus advantageously able to provide a more accurate approximation result compared to an artificial neural network that approximates a plurality of objective functions using a common algorithm.

[0046] According to a further aspect of the invention, the method according to the invention further comprises the artificial neural network using, depending on the objective function, between 2 and 15 layers, between 6 and 2048 neurons per layer, and ReLU, LeakyReLU, or ELU as the activation function. Thus, the best possible approximation results can be achieved in this configuration.

[0047] According to a further aspect of the invention, the method according to the invention further comprises that the second data set consisting of the numerical value range of the target function and the associated definition range of the at least one driving situation parameter can be graphically represented two-dimensionally or three-dimensionally as a function of a number of driving situation parameters.

[0048] Thus, a functional curve of the approximated numerical value range of the objective function can be advantageously graphically represented for the specified definition range of the driving situation parameters used. This enables the graphical representation of non-collision cases, collision cases, and a boundary between non-collision cases and collision cases.

[0049] The invention further relates to a computer-implemented method for providing a trained, artificial neural network for approximating test results of a virtual test of a device for at least partially autonomously controlling a motor vehicle.

[0050] The method comprises receiving a first data set of input training data comprising a plurality of driving situation parameters consisting of environmental parameters describing the surroundings of the motor vehicle and ego parameters describing the state of the motor vehicle, wherein each driving situation parameter has a predetermined definition range.

[0051] The method further comprises receiving a second data set of output training data consisting of the numerical value range of the objective function and the associated definition range of the at least one driving situation parameter, wherein the output training data is related to the input training data.

[0052] Furthermore, the method comprises training the artificial neural network to approximate a numerical value range of the objective function for the definition domain of at least one driving situation parameter, wherein the objective function represents a driving maneuver of the motor vehicle in response to a control of the device generated by the plurality of driving situation parameters of the definition domain, based on the input training data and the output training data with a training calculation unit.

[0053] Furthermore, the method comprises providing the trained artificial neural network for approximating test results of the virtual test of the device for at least partially autonomous guidance of the motor vehicle.

[0054] The artificial neural network trained in this way is thus advantageously able to achieve an accurate approximation result of the target function to be approximated, depending on the number of training data sets and / or training episodes used and the optimization carried out.

[0055] According to a further aspect of the invention, the method according to the invention further comprises generating the plurality of driving situation parameters, in particular the speed of the motor vehicle and the speed of the further motor vehicle, within the predefined definition range by a random algorithm and / or by a simulation. The parameter sets of driving situation parameters can thus be generated in a simple manner, e.g., using an artificial neural network by applying a random function within a predefined definition range.

[0056] According to a further aspect of the invention, the method according to the invention further comprises that the artificial neural network is trained by weighting settings of a plurality of data sets of driving situation parameters using gradient descent backpropagation, in particular using the Adam optimization method.

[0057] When an artificial neural network receives input values, the neurons' outputs are calculated layer by layer and forwarded until they reach the end of the output layer. This is called a forward pass.

[0058] A comparison is made between the actual output and the desired target output to determine the error. This is called error determination. The target output is specified by the data set and is called the target value. In the simplest case, the error can be determined by the absolute value of the difference between the actual and target values. This is called the absolute error. If multiple data sets are considered, the average absolute error can be used. For this purpose, the individual absolute errors are summed and divided by the number of data sets considered.

[0059] In the case of classifications, accuracy is usually given as the percentage of how often the output matches the target value.

[0060] The error can be propagated back through the entire network to the input layer. Since the goal is to minimize the error, the gradient descent method can be used to update the weights toward a local minimum. This is called a backward pass.

[0061] By adjusting the weights in this way, the neural network is able to solve the specific problem better and improve its prediction accuracy. The process of forward passing, error determination, and backward passing is called backpropagation. A single backpropagation sequence is called a training step.

[0062] The Adam optimization method uses different learning rates for different parameters. These can also increase. Furthermore, these learning rates are influenced by momentum. This means that successive weight updates in the same gradient direction result in a higher learning rate. In contrast, changing the gradient direction during successive updates decreases the learning rate.

[0063] The invention further relates to a test unit for approximating test results of a virtual test of a device for at least partially autonomously driving a motor vehicle.

[0064] The test unit comprises means for receiving a first data set comprising a plurality of driving situation parameters consisting of environmental parameters describing the surroundings of the motor vehicle and ego parameters describing the state of the motor vehicle, wherein each driving situation parameter has a predetermined definition range.

[0065] Furthermore, the test unit comprises means for approximating a numerical value range of a target function for the definition domain of at least one driving situation parameter using a trained, artificial neural network which is applied to the plurality of driving situation parameters, wherein the target function represents a driving maneuver of the motor vehicle in response to a control of the device generated by the plurality of driving situation parameters of the definition domain.

[0066] The test unit further comprises means for providing a second data set consisting of the numerical value range of the target function and the associated definition range of the at least one driving situation parameter.

[0067] In the context of the present test unit, an artificial neural network is advantageously used, which has the task of approximating test results depending on the given definition range of the driving situation parameters of interest.

[0068] The test results approximated in this way can then advantageously be validated within the framework of virtual tests of the device, so that the test unit according to the invention enables efficient virtual validation of the device for autonomous driving of motor vehicles.

[0069] According to a further aspect of the invention, it is provided that one of the driving situations underlying the approximation of the test results of the virtual test of the device, which is designed in particular as a control unit, is a lane change of another motor vehicle into a lane of the motor vehicle using the plurality of driving situation parameters.

[0070] The test unit is thus advantageously able to approximate corresponding test results of the virtual test with regard to, for example, a cut-in scenario.

[0071] The invention further relates to a computer program with program code for implementing the approximation method according to the invention when the computer program is executed on a computer. Furthermore, the invention relates to a computer-readable data carrier with program code of a computer program for implementing the approximation method according to the invention when the computer program is executed on a computer.

[0072] The features of the method described herein can be used to approximate test results from a variety of different scenarios or driving situations. The test unit according to the invention is also designed to test a variety of different devices or control units of, for example, automobiles, utility vehicles and / or commercial vehicles, ships, or aircraft in order to approximate test results. Short description of the drawings

[0073] For a better understanding of the present invention and its advantages, reference is now made to the following description in conjunction with the accompanying drawings. The invention is explained in more detail below using exemplary embodiments shown in the schematic illustrations of the drawings.

[0074] They show: Fig. 1 shows a flowchart of a method for approximating test results of a virtual test of a device for at least partially autonomous guidance of a motor vehicle according to a preferred embodiment of the invention; Fig. 2 shows a three-dimensional representation of an inventive target function according to the preferred embodiment of the invention; Fig. 3 shows a three-dimensional representation of a further inventive target function according to the preferred embodiment of the invention; Fig. 4 shows a three-dimensional representation of a further inventive target function according to the preferred embodiment of the invention; Fig. 5 shows a flowchart of the method for approximating the subset of test results of the virtual test of the device for at least partially autonomous guidance of the motor vehicle according to a further preferred embodiment of the invention;6 shows a flowchart of the method for approximating the subset of test results of the virtual test of the device for at least partially autonomous guidance of the motor vehicle according to another preferred embodiment of the invention; 7 shows a flowchart of a method for providing a trained, artificial neural network for approximating test results of a virtual test of a device for at least partially autonomous guidance of a motor vehicle; and 8 shows a test unit for approximating test results of a virtual test of a device for at least partially autonomous guidance of a motor vehicle.

[0075] Unless otherwise indicated, like reference numerals refer to like elements in the drawings. Detailed description of the embodiments

[0076] Fig. 1shows a flowchart of a method for approximating test results of a virtual test of a device for at least partially autonomously guiding a motor vehicle according to a preferred embodiment of the invention.

[0077] The method comprises receiving S1 a first data set DS1a, DS1b, DS1c comprising a plurality of driving situation parameters consisting of environmental parameters FP1, FP2 describing the surroundings of the motor vehicle and ego parameters FP3 describing the state of the motor vehicle.

[0078] Each driving situation parameter has a predefined domain. The method further comprises approximating S2 a numerical value range WB1, WB2, WB3 of a target function F1, F2, F3 for the domain of at least one driving situation parameter using a trained artificial neural network K1.

[0079] The artificial neural network K1 is applied to the majority of driving situation parameters. The objective function F1, F2, F3 represents a driving maneuver of the motor vehicle in response to a control of the device generated by the majority of driving situation parameters of the definition domain.

[0080] Furthermore, the method comprises providing a second data set DS2a, DS2b, DS2c consisting of the numerical value range WB1, WB2, WB3 of the objective function F1, F2, F3 and the associated definition range of the at least one driving situation parameter.

[0081] The ego parameter(s) FP3 comprise(s) a speed V EGO of the motor vehicle and the environmental parameters FP1, FP2 comprise(s) a speed V FELLOW of another motor vehicle and a distance d SPUR between the motor vehicle and the other motor vehicle.

[0082] The majority of driving situation parameters, in particular the speed V EGO of the motor vehicle and the speed V FELLOW of the other motor vehicle, are generated within the specified definition range by a random algorithm.

[0083] Depending on the objective function F1, the artificial neural network K1 uses between 2 and 15 layers, between 6 and 2048 neurons per layer, and ReLU as the activation function. Alternatively, LeakyReLU or ELU can be used as the activation function.

[0084] The second data set DS2a consists of the numerical value range of the objective function F1 and the associated definition domain of at least one driving situation parameter. Depending on the number of driving situation parameters, the aforementioned value range can be graphically represented, for example, two-dimensionally or three-dimensionally.

[0085] Fig. 2shows a three-dimensional image of an inventive target function according to the preferred embodiment of the invention.

[0086] The objective function F1 shown is a safety objective function. The safety objective function has a value determined by the (in Fig. 1 shown) artificial neural network K1 approximated numerical value or function value.

[0087] The function value has a minimum value at a safety distance between the motor vehicle and the other motor vehicle of ≥ V FELLOW x 0.55. Furthermore, the function value has a maximum value in the event of a collision between the motor vehicle and the other motor vehicle. Furthermore, the function value has a value that is greater than the minimum value at a safety distance between the motor vehicle and the other motor vehicle of ≤ V FELLOW x 0.55.

[0088] The Fig. 2The value range WB1 of the second data set DS2a is shown three-dimensionally in this illustration. The value range thus clearly shows the individual approximation results of the respective parameter combinations of the driving situation parameters used, represented by respective points. Likewise, a boundary between collision and non-collision cases can be identified, for example, based on the displayed function curve.

[0089] Fig. 3 shows a three-dimensional image of another objective function according to the invention according to the preferred embodiment of the invention.

[0090] The Fig. 3The objective function F2 shown is a comfort objective function which has a numerical value which has a minimum value in the case of no change in the acceleration of the motor vehicle, a maximum value in the case of a collision between the motor vehicle and the other motor vehicle, and a value between the minimum value and the maximum value in the case of a change in the acceleration of the motor vehicle depending on an amount of the change in the acceleration.

[0091] The Fig. 3 The numerical value range of the comfort target function shown from 0 to 140 is arbitrarily selected and can alternatively have a different value range.

[0092] Based on the displayed value range, it can thus be determined whether the displayed values lie in a range in which a collision between the vehicle and the other vehicle occurs, in which no collision occurs between the vehicle and the other vehicle, or whether the numerical function value lies in an intermediate range.

[0093] Fig. 4 shows a three-dimensional image of another objective function according to the invention according to the preferred embodiment of the invention.

[0094] The Fig. 4The objective function F3 shown is an energy consumption objective function which has a numerical value which has a minimum value in the case of no change in the acceleration of the motor vehicle, a maximum value in the case of a collision between the motor vehicle and the other motor vehicle, and a value between the minimum value and the maximum value in the case of a change in the acceleration of the motor vehicle depending on an amount of the change in the acceleration.

[0095] The Fig. 4 The numerical value range of the comfort objective function shown from 0.1 to 0.2 is arbitrarily selected and can alternatively have a different value range.

[0096] Based on the displayed value range, it can thus be determined whether the displayed values lie in a range in which a collision between the vehicle and the other vehicle occurs, in which no collision occurs between the vehicle and the other vehicle, or whether the numerical function value lies in an intermediate range.

[0097] Fig. 5 shows a flowchart of the method for approximating the subset of test results of the virtual test of the device for at least partially autonomous guidance of the motor vehicle according to a further preferred embodiment of the invention;

[0098] The Fig. 5 The artificial neural network K1 shown has an input layer L1, a plurality of hidden layers L2, and an output layer L3. The output of the artificial neural network K1 is a first function value F1, a second function value F2, and a third function value F3.

[0099] The first function value F1 concerns a safety objective function, the second function value F2 concerns a comfort objective function and the third function value F3 concerns an energy consumption objective function.

[0100] The artificial neural network K1 is thus capable of approximating the function value F1 based on a first data set DS1a of the safety objective function. Based on the function value of the objective function F1, the second data set DS2a can then be generated, i.e., a graphical representation of the numerical value range WB1 of the safety objective function can be created.

[0101] The artificial neural network K1 receives a first data set DS1a concerning the safety objective function, a second data set DS2a concerning the comfort objective function and a third data set concerning the energy consumption objective function.

[0102] The data sets DS1a, DS1b, DS1c each have a plurality of driving situation parameters consisting of environmental parameters FP1, FP2 describing the environment of the motor vehicle and ego parameters FP3 describing the state of the motor vehicle.

[0103] Based on the function values K1 approximated by the artificial neural network, a second data set DS2a, DS2b, DS2c consisting of the numerical value range WB1, WB2, WB3 of the objective function F1, F2, F3 and the associated definition range of at least one driving situation parameter can then be generated.

[0104] Fig. 6 shows a flowchart of the method for approximating the subset of test results of the virtual test of the device for at least partially autonomous guidance of the motor vehicle according to a further preferred embodiment of the invention.

[0105] In contrast to the Fig. 5In the embodiment shown, a separate artificial neural network K1, K2, K3 is used in the present embodiment to approximate the numerical value range WB1, WB2, WB3 of each objective function F1, F2, F3. Individual hyperparameters of each artificial neural network K1, K2, K3 are stored in a database.

[0106] The first neural network K1 has an input layer L1a, a plurality of hidden layers L2a and an output layer L3a, which outputs the function value of the security objective function F1.

[0107] The first artificial neural network K1 further receives the first data set DS1a. Based on the approximated function value of the first artificial neural network K1, the second data set DS2a can then be created, which consists of the numerical value range of the target function F1 and the associated definition domain of the at least one driving situation parameter.

[0108] The second neural network K2 has an input layer L1b, a plurality of hidden layers L2b and an output layer L3b, which outputs the function value of the security objective function F2.

[0109] The second artificial neural network K2 further receives the second data set DS1b. Based on the approximated function value of the second artificial neural network K2, the second data set DS2b can then be created, which consists of the numerical value range of the target function F2 and the associated definition domain of the at least one driving situation parameter.

[0110] The third neural network K3 has an input layer L1c, a plurality of hidden layers L2c and an output layer L3c, which outputs the function value of the security objective function F3.

[0111] The third artificial neural network K3 further receives the third data set DS1c. Based on the approximated function value of the third artificial neural network K3, the second data set DS2c can then be created, which consists of the numerical value range of the objective function F3 and the associated definition domain of the at least one driving situation parameter.

[0112] Fig. 7 shows a flowchart of a method for providing a trained, artificial neural network for approximating test results of a virtual test of a device for at least partially autonomously driving a motor vehicle.

[0113] The method comprises receiving S1' a first data set DS1a', DS1b', DS1c' of input training data comprising a plurality of driving situation parameters consisting of environmental parameters FP1, FP2 describing the surroundings of the motor vehicle and ego parameters FP3 describing the state of the motor vehicle. Each driving situation parameter has a predetermined definition range.

[0114] Furthermore, the method comprises receiving S2' a second data set DS2a', DS2b', DS2c' of output training data consisting of the numerical value range WB1, WB2, WB3 of the objective function F1, F2, F3 and the associated definition domain of the at least one driving situation parameter. The output training data are related to the input training data.

[0115] Furthermore, the method comprises the step of training S3' the artificial neural network to approximate a numerical value range WB1, WB2, WB3 of the objective function F1, F2, F3 for the definition range of at least one driving situation parameter.

[0116] The objective function F1, F2, F3 represents a driving maneuver of the motor vehicle in response to a control of the device generated by the plurality of driving situation parameters of the definition domain based on the input training data and the output training data. The approximation is performed using a training calculation unit 20.

[0117] Furthermore, the method comprises providing S4' the trained artificial neural network K1, K2, K3 for approximating test results of the virtual test of the device for at least partially autonomous guidance of the motor vehicle.

[0118] The majority of driving situation parameters, in particular the speed V EGO of the motor vehicle and the speed V FELLOW of the other motor vehicle, are generated within the specified definition range by a random algorithm. Alternatively, they can be generated, for example, by a simulation.

[0119] The artificial neural network K1, K2, K3 is trained by weighting adjustments of a plurality of data sets of driving situation parameters using a gradient descent backpropagation S3a', in particular using the Adam optimization method.

[0120] Fig. 8 shows a test unit for approximating test results of a virtual test of a device for at least partially autonomous control of a motor vehicle.

[0121] The test unit comprises means 10 for receiving a first data set DS1a, DS1b, DS1c comprising a plurality of driving situation parameters consisting of environmental parameters describing the surroundings of the motor vehicle and ego parameters describing the state of the motor vehicle. Each driving situation parameter has a predetermined definition range.

[0122] Furthermore, the test unit 1 comprises means 12 for approximating a numerical value range WB1, WB2, WB3 of a target function F1, F2, F3 for the definition domain of at least one driving situation parameter using a trained artificial neural network K1, K2, K3. The artificial neural network is applied to the plurality of driving situation parameters.

[0123] The target function F1, F2, F3 represents a driving maneuver of the motor vehicle in response to a control of the device generated by the plurality of driving situation parameters of the definition domain. Furthermore, the test unit 1 has means 14 for providing a second data set DS2a, DS2b, DS2c consisting of the numerical value range WB1, WB2, WB3 of the target function F1, F2, F3 and the associated definition domain of the at least one driving situation parameter.

[0124] In the present exemplary embodiment, a driving situation underlying the approximation of the test results of the virtual test of the device, which is designed in particular as a control unit, is a lane change of another motor vehicle into a lane of the motor vehicle using the plurality of driving situation parameters.

[0125] Alternatively, an underlying driving situation can be another driving situation such as approaching a traffic light with and / or without a lane change and one or more vehicles involved.

[0126] Although specific embodiments have been illustrated and described herein, it will be understood by those skilled in the art that numerous alternative and / or equivalent implementations exist. It should be noted that the exemplary embodiment or exemplary embodiments are merely examples and are not intended to limit the scope, applicability, or configuration in any way.

[0127] Rather, the foregoing summary and detailed description will provide one skilled in the art with a convenient road map for implementing at least one exemplary embodiment, it being understood that various changes in the functionality and arrangement of elements may be made without departing from the scope of the appended claims and their legal equivalents.

[0128] In general, this application is intended to cover modifications, adaptations, or variations of the embodiments presented herein.

Claims

1. A computer-implemented method for approximating test results of a virtual test of a device for at least partially autonomously guiding a motor vehicle, having the steps: receiving (S1) a first data set (DS1a, DS1b, DS1c) comprising a plurality of driving situation parameters consisting of environmental parameters (FP1, FP2) describing the environment of the motor vehicle and ego parameters (FP3) describing the state of the motor vehicle, each driving situation parameter comprising a specified definition range; approximating (S2) a numerical value range (WB1, WB2, WB3) of a target function (F1, F2, F3) for the definition range of at least one driving situation parameter using a trained artificial neural network (K1, K2, K3) applied to the plurality of driving situation parameters, the target function (F1, F2, F3) representing a driving maneuver of the motor vehicle in response to a control of the device generated by the plurality of driving situation parameters of the definition range; and providing (S3) a second data set (DS2a, DS2b, DS2c) consisting of the numerical value range (WB1, WB2, WB3) of the target function (F1, F2, F3) and the associated definition range of the at least one driving situation parameter.

2. The computer-implemented method according to claim 1, characterized in that the ego parameters (FP3) are a speed (VEGO) of the motor vehicle and the environmental parameters (FP1, FP2) are a speed (VFELLOW) of another motor vehicle and a distance (dSPUR) between the motor vehicle and the other motor vehicle.

3. The computer-implemented method according to claim 2, characterized in that the target function (F1) is a safety target function having a numerical value having a minimum value at a safety distance between the motor vehicle and the other motor vehicle of ≥ VFELLOW x 0.55, having a maximum value in the event of a collision between the motor vehicle and the other motor vehicle, and having a value greater than the minimum value at a safety distance between the motor vehicle and the other motor vehicle of ≤ VFELLOW x 0.55.

4. The computer-implemented method according to claim 2 or 3, characterized in that the target function (F2, F3) is a comfort target function or an energy consumption target function having a numerical value having a minimum value in the event of no change in the acceleration of the motor vehicle, a maximum value in the event of a collision between the motor vehicle and the other motor vehicle, and a value between the minimum value and the maximum value in the event of a change in the acceleration of the motor vehicle depending on the amount of the change in acceleration.

5. The computer-implemented method according to any one of claims 2 to 4, characterized in that the plurality of driving situation parameters, in particular the speed (VEGO) of the motor vehicle and the speed (VFELLOW) of the other motor vehicle, are generated within the specified definition range using a random algorithm.

6. The computer-implemented method according to any one of the preceding claims, characterized in that a separate artificial neural network (K1, K2, K3) is used to approximate the numerical value range (WB1, WB2, WB3) of each target function (F1, F2, F3), wherein individual hyperparameters of each artificial neural network (K1, K2, K3) are stored in a database.

7. The computer-implemented method according to any one of the preceding claims, characterized in that the artificial neural network (K1, K2, K3) comprises between 2 and 15 layers and between 6 and 2048 neurons per layer depending on the target function (F1, F2, F3), and ReLU, LeakyReLU or ELU is used as an activation function.

8. The computer-implemented method according to any one of the preceding claims, characterized in that the second data set (DS2a, DS2b, DS2c) consisting of the numerical value range (WB1, WB2, WB3) of the target function (F1, F2, F3) and the associated definition range of the at least one driving situation parameter can be displayed graphically in two or three dimensions depending on a number of driving situation parameters.

9. A computer-implemented method for providing a trained, artificial neural network (K1, K2, K3) for approximating test results of a virtual test of a device for at least partially autonomously guiding a motor vehicle, having the steps: receiving (S1') a first data set (DS1a', DS1b', DS1c') of input training data comprising a plurality of driving situation parameters consisting of environmental parameters (FP1, FP2) describing the environment of the motor vehicle and ego parameters (FP3) describing the state of the motor vehicle, each driving situation parameter having a specified definition range; receiving (S2') a second data set (DS2a', DS2b', DS2c') of output training data consisting of the numerical value range (WB1, WB2, WB3) of the target function (F1, F2, F3) and the associated definition range of the at least one driving situation parameter, the output training data being related to the input training data; training (S3') the artificial neural network for approximating a numerical value range (WB1, WB2, WB3) of the target function (F1, F2, F3) for the definition range of at least one driving situation parameter, the target function (F1, F2, F3) representing a driving maneuver of the motor vehicle in response to a control of the device generated by the plurality of driving situation parameters of the definition range, based on the input training data and the output training data, by means of a training calculation unit (20); and providing (S4') the trained, artificial neural network (K1, K2, K3) for approximating test results of the virtual test of the device for at least partially autonomously guiding the motor vehicle.

10. The computer-implemented method according to claim 9, characterized in that the plurality of driving situation parameters, in particular the speed (VEGO) of the motor vehicle and the speed (VFELLOW) of the other motor vehicle, are generated within the specified definition range by a random algorithm and / or by a simulation.

11. The computer-implemented method according to claim 9 or 10, characterized in that the artificial neural network (K1, K2, K3) is trained by weighting settings of a plurality of data sets of driving situation parameters using a gradient descent back propagation (S3a'), in particular using the Adam optimization method.

12. A test unit (1) for approximating test results of a virtual test of a device for at least partially autonomously guiding a motor vehicle, comprising: means (10) for receiving a first data set (DS1a, DS1b, DS1c) comprising a plurality of driving situation parameters consisting of environmental parameters describing the environment of the motor vehicle and ego parameters describing the state of the motor vehicle, each driving situation parameter having a specified definition range; means (12) for approximating a numerical value range (WB1, WB2, WB3) of a target function (F1, F2, F3) for the definition range of at least one driving situation parameter using a trained artificial neural network (K1, K2, K3) applied to the plurality of driving situation parameters, the target function (F1, F2, F3) representing a driving maneuver of the motor vehicle in response to a control of the device generated by the plurality of driving situation parameters of the definition range; and means (14) for providing a second data set (DS2a, DS2b, DS2c) consisting of the numerical value range (WB1, WB2, WB3) of the target function (F1, F2, F3) and the associated definition range of the at least one driving situation parameter.

13. The test unit according to claim 12, characterized in that a driving situation underlying the approximating of the test results of the virtual test of the device particularly implemented as a control unit is a lane change of another motor vehicle into a driving lane of the motor vehicle using the plurality of driving situation parameters.

14. A computer program having program code for performing the method according to any one of the claims 1 through 8 when the computer program is executed on a computer.

15. A computer-readable data storage medium having program code of a computer program for performing the method according to any one of the claims 1 through 8 when the computer program is executed on a computer.