Method for determining at least one perception requirement for a function of a vehicle system of a motor vehicle, computing device and computer program

Abstract

Method for determining at least one perception requirement for a function of a vehicle system (2) of a motor vehicle (1), which determines output data for controlling at least one component of the motor vehicle (1) from structured input data comprising a plurality of data elements and determined from the sensor data of at least one sensor of the motor vehicle (1), wherein the perception requirement describes a required quality and / or accuracy of the input data and / or the sensor data, wherein - for a given base value set (10) of input data, a plurality of variation value sets (11, 16) are determined by varying the value of at least one data element according to at least one variation rule, - the function for determining respective output data sets is applied to the variation value sets (11, 16), - the output data sets are evaluated to determine quality information regarding the fulfilment of at least one result requirement, and - at least one perception requirement is determined from the quality information.

Description

[0001] The invention relates to a method for determining at least one perception requirement for a function of a vehicle system of a motor vehicle. The method determines output data for controlling at least one component of the motor vehicle from structured input data comprising a plurality of data elements and determined from the sensor data of at least one sensor of the motor vehicle. The perception requirement describes a required quality and / or accuracy of the input data and / or the sensor data. The invention also relates to a computing device and a computer program.

[0002] Vehicle systems that allow fully automatic control of the motor vehicle within certain limits, known as system limits, are already available and are also the subject of research, particularly on the path to a fully autonomously operable motor vehicle. According to SAE J3016 of April 30, 2021, five levels of autonomous driving are distinguished, with fully automatic control being achieved from level 3 onwards. This involves a driving mode-specific execution of all aspects of the dynamic driving task by the vehicle system with the expectation that the driver will respond to a request for intervention (driver takeover request). At higher levels, this expectation no longer exists.

[0003] Autonomous driving functions designed for fully automatic control of the motor vehicle utilize structured input data, for example, the result of sensor data fusion. This process combines sensor data and, typically, situational information from other vehicle systems to obtain the most accurate possible representation of the current driving situation. With regard to the motor vehicle's surroundings, in particular relevant environmental objects, the result of the sensor data fusion can, for example, comprise a map of the surroundings and / or an object list. Attributes are typically assigned to the corresponding environmental objects as data elements, such as size and / or position and / or at least one class and / or other properties.

[0004] Autonomous driving functions, as well as other functions of vehicle systems in motor vehicles, such as driver assistance functions, usually have high performance requirements for the output data used to control vehicle components (e.g., for driving interventions). These performance requirements relate in particular to the safety of the vehicle and its occupants, for example, to the implementation of the levels of autonomous driving described above. However, performance requirements relating to the comfort of using the vehicle are also conceivable.

[0005] For functions of vehicle systems, especially autonomous driving functions, of this type, environmental perception, and in particular the environmental sensors used, plays a key role. In particular, perception in motor vehicles must be designed in such a way that the result requirements of the function can be met. The perception of the environment, in particular, is a complex field and subject to a certain susceptibility to errors. Therefore, perception requirements for functions of vehicle systems, especially for autonomous driving functions, are usually set extremely high to "play it safe." The specific configuration of sensors is often selected through tests. For example, a specific configuration of sensors can be implemented and test drives or similar can be used to check whether the result requirements of the function are met.However, such test drives cannot cover all conceivable practical cases, and the mechanisms that lead to errors in measurement, sensor data fusion, and the processing of input data into output data by the function are in many cases difficult or impossible to understand, and therefore cannot be theoretically modeled. This ultimately leads to the installation of expensive sensors that meet particularly high perception requirements, especially environmental sensors, in motor vehicles in order to be prepared for situations and errors that are incomprehensible, rare, or simply accidentally untested.

[0006] The invention is based on the object of determining perception requirements that are better tailored to the actual performance of a function of a vehicle system of a motor vehicle, in particular an autonomous driving function.

[0007] This object is achieved according to the invention by a computer-implemented method, a computing device, and a computer program according to the independent claims. Advantageous further developments are set out in the dependent claims.

[0008] In a method of the type mentioned above, the invention provides that - for a given base value set of input data, a plurality of variation value sets are determined by varying the value of at least one data element according to at least one variation rule, - the function for determining the respective initial data sets is applied to the variation value sets, - the output data sets are evaluated to determine quality information regarding the fulfilment of at least one result requirement, in particular a safety-related one, and - at least one perception requirement is determined from the quality information.

[0009] In this case, the at least one perception requirement can be determined in particular such that, upon fulfillment of the perception requirement, each result requirement is fulfilled and / or at least fulfilled to a certain degree (minimum degree). Furthermore, embodiments can provide for the base value set itself to be used as an additional variation value set.

[0010] According to the invention, it is therefore proposed to start from a set of base values, in particular describing a real driving situation and / or recorded in a real driving situation, which can preferably be error-free, in order to be able to test, through systematic variation of data elements defined by at least one variation rule, in which constellations the technical result requirements of the function are met, and possibly how well, as described by the quality information. From this, the technical perception requirements for the function can thus be derived. In other words, data elements, for example fields, of the input data describing technical facts are specifically varied in order to find robustness limits of the function.In particular, as will be discussed in more detail below, technical errors can be deliberately incorporated to determine whether the function can handle these errors within the result requirements. In this way, technical perception requirements of a function controlling components in the motor vehicle can be defined more precisely, which, as has been shown within the scope of the invention, usually leads to a reduction in at least one perception requirement compared to current design mechanisms. In this way, oversizing of sensors, in particular environmental sensors, can be avoided, and the sensor technology can be designed more simply, in particular more cost-effectively and less complexly, in order to implement the vehicle system and the motor vehicle.In particular, systematic variation of data elements based on at least one variation rule makes it possible to cover a significantly larger number of, particularly technically relevant, cases and driving situations, especially error cases, than is possible with test drives. Furthermore, no knowledge of the weaknesses of individual sensors is required.

[0011] It is therefore possible to determine the at least one perception requirement with low effort and in an improved manner, which results in a reduction of the at least one perception requirement and allows a simpler design of the sensors used for the function in the motor vehicle.

[0012] In this case, a clear functional relationship exists since the variation occurs in the input data, in particular at the output of a sensor data fusion. Therefore, data processing operations in the sensor system and / or post-processing of the sensor data, in particular the sensor data fusion, are neither considered nor required for determining the at least one perception requirement.

[0013] The actual design or selection of the sensors takes place in a subsequent step using the at least one perception requirement. The at least one determined perception requirement can then be translated into at least one sensor requirement for the at least one sensor. The at least one sensor requirement can be used in the design of the vehicle system or motor vehicle and / or in the design of a test system for selecting the at least one sensor and / or for selecting its properties in order to implement the perception requirement in the motor vehicle. In summary, specifications for perception, in particular environmental perception, can be derived through extreme value-based robustness tests of the functional logic of the function, in particular of an autonomous driving function.

[0014] In particularly preferred embodiments of the present invention, it can therefore be provided that the function is an autonomous driving function designed for the fully automatic control of the motor vehicle. In this case, the autonomous driving function is at least a Level 3 driving function according to SAE J3016 of April 30, 2021, which means that the vehicle system completely assumes control of the vehicle and thus the dynamic driving task within the scope of the autonomous driving function. This is precisely where high performance requirements exist, which can be ideally met with the inventive approach without excessive over-dimensioning.

[0015] The perception requirement here relates in particular, as already explained, to the motor vehicle's perception of the surroundings. Thus, the at least one sensor can comprise at least one environmental sensor of the motor vehicle, in particular selected from the group comprising at least one radar sensor and / or at least one camera and / or at least one lidar sensor and / or at least one ultrasonic sensor.

[0016] The quality information can, for example, record a binary value for each result requirement, thus indicating whether the at least one result requirement is met. However, in preferred embodiments, it is also conceivable for the quality information to comprehensively determine a quality value that describes a degree of fulfillment of at least one of the at least one result requirement. In this case, it can expediently be provided that the at least one perception requirement is determined to establish a minimum value for the degree of fulfillment, a minimum degree of fulfillment, of at least one of the at least one result requirement. Such a degree of fulfillment is particularly expedient when the result requirement relates to the comfort of the occupants of the motor vehicle, since safety-related result requirements must often be met absolutely.However, it is also conceivable within the scope of the present invention to determine the at least one perception requirement for at least part of the at least one result requirement in an optimization process such that the highest possible degree of fulfillment is given for at least part of the at least one result requirement.

[0017] In a particularly preferred development of the present invention, as already indicated, it can be provided that at least one variation rule provides for the targeted provision of at least one error with regard to the driving situation described by the base value set and / or at least one extreme value for at least one data element. By deliberately and systematically introducing errors based on the base value set, the robustness of the function against errors, i.e. in particular data elements whose value does not describe the actual given circumstance, is tested in a systematic, structured manner. Alternatively or additionally, rare or extreme situations (which can of course also include error cases) can be specifically provided as a set of variation values. An extreme value can, for example, lie at the edge of a value interval of possible values ​​for the data element.However, it is also possible to define the extreme value with respect to the driving situation, for example, as increasing / maximizing the risk of an accident, in order to specifically test accident avoidance strategies and / or accident consequence mitigation strategies and their robustness. Finally, it is also conceivable to define an extreme value based on the rarity of its occurrence in practice. Thus, when determining at least one perception requirement, cases that are rare in practice are also tested, especially those that are also particularly relevant with regard to the function. In relation to autonomous driving functions, such driving situations are often referred to as "corner cases."

[0018] In specific embodiments, it can be provided that at least one variation rule for a discrete data element assuming a finite number of possible values ​​comprises the creation of variation value sets with all possible values ​​as variation values, and / or at least one variation rule for a data element assuming continuous values ​​in a value interval is a selection rule for selecting several discrete, in particular equidistant, variation values ​​for respective variation value sets, particularly one that favors values ​​that occur less frequently in practice. This allows for a comprehensive test, for example, with regard to potentially occurring errors, which relates to the entire possible value range.Starting with a driving situation described by the baseline value set, this method can, for example, cover all conceivable errors related to the data element, and robustness can be tested across a broad distribution of error cases. Similarly, a wide variety of conceivable variants of a driving situation, including less frequently occurring "corner cases," can be reliably and systematically covered. It should be noted that the data element taking on discrete values ​​can, for example, be the result of a classification process, in which case all conceivable classifications can be covered.

[0019] A variation does not necessarily have to refer to a single data element; rather, at least one variation rule can also be used to simultaneously vary the values ​​of at least two data elements. The two data elements can be semantically linked, for example, at least partially dependent on each other, to allow for particularly consistent variations. However, it is also conceivable to generate simultaneously occurring, possibly even independent, errors in different data elements and perform a corresponding robustness test. Thus, the variation spectrum can be meaningfully expanded.

[0020] As already mentioned, the base value set can expediently be or represent a result of a sensor data fusion. For the robustness test of the function described here, in order to derive at least one technical perception requirement, the input data of the function, as it is usually the result of a sensor data fusion, can therefore be used directly, without requiring its details and / or the details of the sensors. Specifically, the base value set can comprise an object list and / or environmental map of environmental objects of a motor vehicle with attributes assigned to the environmental objects as data elements. The determination of such object lists and / or environmental maps using sensor data, in particular in the context of a sensor data fusion, is already largely known in the prior art.

[0021] In a practical, concrete embodiment of the method, it can be provided that the determination of the quality information includes a simulation of the progression of the driving situation described by the base value set and / or the corresponding variation value set using the respective initial data. For example, when deliberately predicting an error, the base value set is then expediently used as the starting point; if the variation value set also implies a variation of the driving situation from the base value set, this is expediently used as the starting point for the simulation. Both may be appropriate in some cases, allowing treatment both as an error and as a variation of the driving situation itself.Alternatively or additionally, plausibility functions and / or mechanisms can be used to determine the quality information and / or reliability values ​​that the function outputs with the output data can be used.

[0022] In addition to the method, the invention also relates to a computing device comprising at least one processor and at least one memory means, designed to carry out a method according to the invention. All statements regarding the method according to the invention can be applied analogously to the computing device according to the invention and vice versa, so that the aforementioned advantages can also be achieved with the computing device. The computing device can be or include a control unit for implementing the function.

[0023] Functional units can be formed in the computing device by hardware and / or software to carry out steps of the method according to the invention. For example, the computing device can comprise an interface for receiving the base value set, a variation unit for determining the variation value sets, an application unit for applying the function to the variation value sets, an evaluation unit for determining the quality information, a determination unit for determining the at least one perception requirement, and optionally an output unit for outputting the at least one determined perception requirement. Further functional units can of course also be provided.

[0024] A computer program according to the invention can be loaded directly into a storage means of a computing device and has program means such that, when the computer program is executed on the computing device, the device is prompted to perform the steps of a method according to the invention. The computer program can be stored on an electronically readable data carrier, which thus comprises control information stored thereon, which comprises at least one computer program according to the invention and is configured such that, when the data carrier is used in a computing device, the device is configured to perform a method according to the invention. The data carrier is, in particular, a non-transient data carrier, for example a CD-ROM.

[0025] Further advantages and details of the present invention will become apparent from the exemplary embodiments described below and from the drawings. In the drawings: Fig. 1 a schematic diagram of a motor vehicle, Fig. 2 a general flow chart of an embodiment of the method according to the invention, Fig. 3 a possible variation in discrete values ​​of a data element, Fig. 4 a possible variation in continuous values ​​of another data element, and Fig. 5 the functional structure of a computing device according to the invention.

[0026] In the following, an embodiment of the method according to the invention for determining perception requirements, i.e. requirements for the input data with regard to perception, in particular environmental perception, which can, for example, describe a tolerance with regard to errors, for an autonomous driving function which is designed for the fully automatic control of the motor vehicle, will be described. Fig. 1 a schematic diagram of a motor vehicle 1 in which the autonomous driving function can be used.

[0027] The motor vehicle 1 comprises a vehicle system 2 with a control device 3, in particular a control unit that provides the autonomous driving function. The environmental perception required for this is provided by various environmental sensors 4, of which a camera 5, radar sensors 6, and a lidar sensor 7 are shown purely as examples. In order to be able to provide structured input data for the autonomous driving function from the sensor data of the environmental sensors 4, among other things, the motor vehicle 1 further has a sensor data fusion control device 8, in which sensor data fusion takes place, which can also take into account situation information from other vehicle systems. In the present case, the sensor data fusion provides, in particular, an object list of environmental objects and / or an environmental map that also contains the environmental objects. Attributes are assigned to the environmental objects as data elements.

[0028] In the embodiment of the method according to Fig. 2, a base value set of the structured input data is first provided in a step S1, which is preferably recorded in a real driving situation or at least describes a real driving situation. The base value set can comprise input data for several consecutive points in time within a driving situation. This can be a so-called "corner case." The base value set can preferably be error-free, meaning that the environmental information in the structured input data or the structured input data in general is correct.

[0029] Starting from the base value set, a plurality of variation value sets are generated in step S2 by applying variation rules. The variation rules are generally chosen to systematically generate conceivable errors and / or provide extreme values ​​for data elements. For example, certain attributes assigned to environmental objects can be varied to "artificially" introduce errors and / or vary the driving situation, particularly towards a "corner case," in order to test the robustness of the autonomous driving function using the corresponding variation value set.

[0030] This can be illustrated by two schematic examples using the Fig. 3 and Fig. 4 is explained in more detail. Fig. Figure 3 schematically shows the value range 9 of a data element, which can assume a finite number of discrete values. For example, the data element can be the result of a classification process, e.g., describing an object class of an environment object. The base value set 10 contains the hatched value. For all other, non-hatched values, a variation value set 11 is then determined, in which the hatched value is replaced by the respective non-hatched value. It should be noted at this point that the base value set 10 can also, in principle, be used as an additional variation value set 11.

[0031] Fig. Figure 4 illustrates the case of a continuous value range, i.e., value interval 12. The base value set 10 specifies a relatively central value, indicated by arrow 13. The value interval 12 is limited by a minimum value 14 and a maximum value 15 as extreme values. Interpolation points are then selected between them according to the corresponding variation rule. These can also correspond to extreme values ​​defined in other ways, for example, extremely rare values, values ​​from "corner cases," and the like. If a uniform robustness check is desired, the interpolation points can also be selected equidistantly. A variation data set 16 with the corresponding variation value for the data element is then generated for each of the minimum value 14 and the maximum value 15, as well as for the other variation values ​​of the interpolation points.Due to the selection of variation values ​​(minimum value 14, maximum value 15, support points), this variation rule can also be called a selection rule.

[0032] Variation rules can also affect multiple data elements, i.e. their joint variation.

[0033] In a step S3, the autonomous driving function is then applied to all variation value sets 11, 16 as respective input data in order to determine associated output data sets containing control information for components of the motor vehicle 1, for example to allow driving interventions.

[0034] In step S4, the obtained output data sets are further evaluated for all variation value sets to determine quality information for each output data set. This information describes whether and to what extent the result requirements for the autonomous driving function are met. Safety-related result requirements can relate to the safe, reliable operation of the autonomous driving function, while comfort-related result requirements relate to the comfort of occupants. To determine the quality information, a simulation is carried out. Variation value sets 11, 16 describing errors are based on the base value set 10, and variation value sets 11, 16 describing varied driving situations are based on the respective variation value set 11, 16. This simulation describes the further development of the driving situation, if applicable, using the output data sets.With variation value sets 11, 16 describing an error, the error can be retained throughout the simulation. Depending on how the driving situation evolves, it can be assessed whether the respective result requirements are met or to what degree they are met. Quality values ​​can indicate in binary form whether a result requirement is met or can also describe a degree of fulfillment.

[0035] In step S5, the quality information is used to determine at least one perception requirement for the autonomous driving function. For binary quality values, fulfillment of the respective result requirement can be required; for quality values ​​indicating a degree of fulfillment, a minimum degree of fulfillment can be required, at least in part. An optimization process is also conceivable, within the framework of which, for example, the degrees of fulfillment for at least some of the result requirements, for example, comfort-related result requirements, are maximized.

[0036] The perception requirements thus determined can then be used in the further course to specifically design the vehicle system 2 or the motor vehicle 1, specifically the environmental sensors 4.

[0037] Fig.5 finally shows a schematic diagram of a computing device 17 according to the invention, which comprises at least one processor (not shown in detail here) and a storage means 18. The base value set 10 can be received via an interface 19. A variation unit 20 is designed to determine the variation value sets 11, 16 according to step S2. In an application unit 21, the autonomous driving function is then applied to the variation value sets 11, 16 according to step S3. The application unit 21, or else the entire computing device, can be provided by a control unit. In an evaluation unit 22, the quality information is determined according to step S4 by assessing the output data sets. A determination unit 23 determines the at least one perception requirement according to step S5.

[0038] In addition, an output unit 24 can also be provided for outputting the at least one determined perception requirement.

Claims

[1] Method for determining at least one perception requirement for a function of a vehicle system (2) of a motor vehicle (1), which determines output data for controlling at least one component of the motor vehicle (1) from structured input data comprising a plurality of data elements and determined from the sensor data of at least one sensor of the motor vehicle (1), wherein the perception requirement describes a required quality and / or accuracy of the input data and / or the sensor data, characterized by , that - for a given base value set (10) of input data, a plurality of variation value sets (11, 16) are determined by varying the value of at least one data element according to at least one variation rule, - the function for determining respective output data sets is applied to the variation value sets (11, 16), - the output data sets are evaluated to determine quality information regarding the fulfilment of at least one result requirement, and - at least one perception requirement is determined from the quality information. [2] Method according to claim 1, characterized by that the function is an autonomous driving function which is designed for the completely automatic guidance of the motor vehicle (1), and / or the at least one sensor comprises at least one environmental sensor (4) of the motor vehicle (1), in particular selected from the group comprising at least one radar sensor (6) and / or at least one camera (5) and / or at least one lidar sensor (7) and / or at least one ultrasonic sensor. [3] Method according to one of the preceding claims, characterized bythat the quality information is determined comprehensively as a quality value that describes a degree of fulfillment of at least one of the at least one result requirement. [4] Method according to claim 3, characterized by that the at least one perception requirement is determined to produce a minimum value for the degree of fulfillment of at least one of the at least one of the at least one result requirement. [5] Method according to one of the preceding claims, characterized by that at least one variation rule provides for the targeted provision of at least one error with regard to the driving situation described by the set of basic values ​​(10) and / or at least one extreme value for at least one data element. [6] Method according to one of the preceding claims, characterized bythat at least one variation rule for a discrete data element assuming a finite number of possible values ​​comprises the creation of variation value sets (11) with all possible values ​​as variation values ​​and / or at least one variation rule for a data element assuming continuous values ​​in a value interval (12) is a selection rule for selecting a plurality of discrete, in particular equidistant, variation values ​​for respective variation value sets (16), in particular preferring values ​​that occur less frequently in practice. [7] Method according to one of the preceding claims, characterized by that at least one variation rule concerns the simultaneous variation of the values ​​of at least two data elements. [8] Method according to one of the preceding claims, characterized bythat the base value set (10) is a result of a sensor data fusion and / or that the base value set (10) comprises an object list and / or environment map of environmental objects of a motor vehicle (1) with attributes assigned to the environmental objects as data elements. [9] Computing device (17) comprising at least one processor and at least one memory means (18), designed to carry out a method according to one of the preceding claims. [10] Computer program which, when executed on a computing device (17), causes the computing device (17) to carry out a method according to one of claims 1 to 8.

Images