Method for learning a following distance, and system for an automated longitudinal guidance of a vehicle

Abstract

In a method (2) for learning a following distance for an automated longitudinal guidance of a vehicle, the following steps are carried out for at least one speed data point of a distance-speed characteristic curve during a learning phase: providing (21) a previous target following distance for the speed data point; providing (22) the actual following distance; determining (23) a new target following distance for the speed data point on the basis of the actual following distance; and replacing (24) the previous target following distance with the new target following distance. The step of determining (23) the new target following distance is carried out on the basis of one or more of the following variables: - a deviation between the actual following distance and the previous target following distance; - a speed difference factor, which is greater the closer the actual speed of the vehicle is to the speed data point; - a learning duration factor, which increases during the learning phase as long as the actual following distance lies within specified limits around an initial actual following distance and the actual speed lies within specified limits around an initial actual speed; - a plausability value, which is increased or reduced at multiple previous points in time on the basis of a difference between the actual following distance detected at the respective point in time and the target following distance that applied to said point in time; and - a freeway factor, which is greater the longer an override of the automated longitudinal guidance of the vehicle on a freeway or freeway-like road lasts.

Description

[0001] Method for learning a following distance and system for automated longitudinal guidance of a vehicle

[0002] The invention relates to a computer-implemented method for learning a following distance for automated longitudinal guidance of a vehicle, a processing device and a computer program for executing such a method, and a computer-readable storage medium on which such a computer program is stored. Furthermore, the invention relates to a system for automated longitudinal guidance of a vehicle using a following distance learned by means of the method and to a vehicle equipped with such a system.

[0003] Many modern motor vehicles are equipped with automated driving functions which can, in particular, enable automated longitudinal guidance of the vehicle. One example of a widely used automated driving function is automatic adaptive cruise control, which is often also referred to as adaptive cruise control or ACC (for “Adaptive Cruise Control”). If a vehicle in front is traveling slower than a target speed set for the vehicle, the following distance to the vehicle in front is regulated to a target following distance using automatic engine and braking interventions. The target following distance can, for example, be a specific interval in seconds, i.e. it can be specified as a specific time interval at which the vehicle equipped with ACC should follow the vehicle in front.

[0004] Since different drivers prefer different following distances, the target following distance is usually adjustable by the driver within certain limits. For example, it may be possible to set the following distance globally in a number of distance levels, with a respective distance-speed characteristic curve stored for each of the distance levels. However, such predefined distance levels only partially reflect individual needs and habits. While they reflect an average value for average drivers, they do not offer the option of accommodating specific driver preferences.

[0005] The course of the preset distance-speed characteristic curves in conventional systems follows a fixed pattern. At low speeds, the distance per second is usually higher. The higher the speed, the lower the corresponding distance becomes. However, this does not necessarily correspond to the driver's desired behavior. For example, a driver may like to follow closely behind in heavy city traffic and choose comparatively large distances on motorways and country roads in order to reach their destination in a relaxed manner. These preferences are not reflected in conventional distance levels, as this does not correspond to the average driver. Due to the limited range of distance levels, individual driver preferences can only be taken into account to a limited extent.

[0006] Furthermore, different driving situations, such as city traffic or highway driving, are not taken into account, requiring the driver to manually switch between the distance settings in different situations. The need for such manual operations can impair the comfort experience and potentially distract the driver from their driving tasks.

[0007] There are already known solutions in the state of the art for learning desired distances from a driver's driving behavior and taking these into account in automatic longitudinal guidance.

[0008] For example, DE 102015 016 993 A1 describes a method for determining a driving profile for a driver assistance system of a vehicle, comprising the steps: when the driver assistance system is inactive, recording distance measurement signals from a distance measurement system of the vehicle and driving dynamics measurement signals of the vehicle;

[0009] Storing the recorded distance measurement signals and driving dynamics measurement signals in a memory device; determining a driving behavior from the stored distance measurement signals and driving dynamics measurement signals, wherein a distance and / or temporal distance to a vehicle in front is determined from the distance measurement signals; and storing the determined driving behavior. Furthermore, a driver assistance control method, in particular an ACC control method using the method, and a driver assistance system for implementing the driver assistance control method are proposed.

[0010] Furthermore, it has already been proposed to learn individual driver behavior with regard to following distances using neural networks (see, for example, HUANG, Xiuling; SUN, Jie; SUN, Jian. A car-following model considering asymmetric driving behavior based on long short-term memory neural networks. Transportation research part C: emerging technologies, 2018, 95th vol., pp. 346-362). It is an object of the present invention to provide a resource-efficient, computer-implemented method for learning a following distance as accurately as possible based on recorded driver behavior, as well as a system for automated longitudinal guidance of a vehicle that makes use of the method.

[0011] The problem is solved by the subject matter of the independent patent claims. Advantageous embodiments are specified in the dependent patent claims.

[0012] A first aspect of the invention relates to a computer-implemented method for learning a following distance for automated longitudinal guidance of a vehicle.

[0013] The vehicle can, in particular, be a motor vehicle. The term "motor vehicle" is understood to mean, in particular, a land vehicle that is propelled by mechanical power without being tied to railway tracks. A motor vehicle in this sense can be, for example, a passenger car, motorcycle, or tractor.

[0014] The automated longitudinal guidance of the vehicle can be carried out as part of an automated driving function, such as an ACC function or a more comprehensive function that supports, for example, combined longitudinal and lateral guidance.

[0015] In this document, the term "automated driving function" generally refers to a vehicle function that enables automated driving. The term "automated driving" refers to driving with automated longitudinal and / or lateral guidance. This can, for example, involve extended driving on the highway or limited-time driving while parking. The term "automated driving" encompasses automated driving with any degree of automation. Examples of levels of automation include assisted, partially automated, conditionally automated, highly automated, and fully automated driving (each with an increasing degree of automation). The five levels of automation mentioned above correspond to SAE levels 1 to 5 of the SAE J3016 standard (SAE - Society of Automotive Engineering) as of April 30, 2021.In assisted driving (SAE Level 1), the system performs longitudinal or lateral guidance in certain driving situations, with the expectation that the driver will assume all remaining aspects of the dynamic driving tasks. In partially automated driving (SAE Level 2), the system assumes longitudinal and lateral guidance in certain driving situations, whereby the driver must continuously monitor the system, as in assisted driving. In conditionally automated driving (SAE Level 3), the system assumes longitudinal and lateral guidance in certain driving situations without the driver having to continuously monitor the system; however, the driver must be able to assume control of the vehicle within a certain period of time upon request from the system. In highly automated driving (SAE Level 4), the system assumes control of the vehicle in certain driving situations, even if the driver does not respond to a request for intervention, thus eliminating the driver as a fallback.In fully automated driving (SAE Level 5), the system can perform all aspects of the dynamic driving task under any road and environmental conditions that can also be mastered by a human driver.

[0016] In the method proposed here, several steps are carried out for at least one support point speed of a distance-speed characteristic curve during a learning phase, which steps are explained in detail below.

[0017] According to one embodiment variant, the steps can also be performed for multiple interpolation point speeds of the distance-speed characteristic curve. Alternatively or additionally, the steps for the interpolation point speed or for the multiple interpolation point speeds can be performed at multiple points in time during the learning phase.

[0018] The fundamental approach used to adapt the adaptive cruise control to individual driving behavior involves learning specific distance characteristics. The goal is to memorize the driver's preferences and adjust the set distance to their needs. The adaptive cruise control system is thus adapted to the driver's personal driving profile to make it as comfortable as possible for the driver.

[0019] Specifically, this is to be achieved by automatically adapting a stored distance-speed characteristic curve, which is used within the framework of automated longitudinal guidance for speed-dependent adjustment of the distance to the vehicle in front, to the observed driving behavior of the driver. The distance-speed characteristic curve assigns a target following distance to a number of speed values, which are referred to as support point speeds in the context of this description. The target following distance can, for example, be a time interval (hereinafter also referred to as a second interval), which indicates the time offset with which the vehicle should follow the vehicle in front. If the second interval is, for example, 1.5 seconds, the longitudinal guidance of the vehicle is controlled such that the vehicle always reaches a considered point (i.e., for example,a longitudinal position along the common lane of the vehicle and the vehicle in front). Alternatively, the target following distance can also be specified as a real distance, i.e., with a unit of length.

[0020] The characteristic curve should be adapted in defined learning phases during a following journey controlled or at least influenced by the driver. To define the learning phases, it is necessary to recognize as accurately as possible when the driver has set his or her desired distance. Road traffic is dynamic. New vehicles often cut in, acceleration occurs, new speed limits appear, or sudden braking is necessary. In all of these situations, the driver deliberately does not drive at a constant distance behind a target object. Rather, the resulting distances can change dynamically very significantly and be temporarily larger or smaller than the actual desired distance. Within the framework of the method proposed here or a corresponding learning algorithm, such situations should therefore preferably be disregarded, i.e. the learning phases are scheduled outside of such driving situations.It is advantageous to limit the learning phases to situations of stable following with a comparatively constant following distance, in which it can be assumed that the driver has consciously adjusted his desired following distance.

[0021] With regard to the manner in which such meaningful following situations can be recognized, reference is made to the applicant's patent application "Method and System for Detecting a Following Situation," filed with the German Patent and Trademark Office on January 15, 2024, the content of which is hereby incorporated into this application. The technical solutions described therein can be advantageously used within the framework of the method described here to define the learning phase(s).

[0022] In one step of the method, a previous (i.e. previously applicable) target following distance is provided for the interpolation point speed of the distance-speed characteristic curve. This provision can, for example, comprise reading out the previous target following distance from a data memory in which various speed values ​​with associated target following distances are stored. In a further step, a recorded actual following distance is provided. The actual following distance can, for example, be recorded using a suitable sensor (e.g. radar) during the learning phase. As mentioned above, the learning phase can take place in a situation in which the driver consciously sets a desired following distance and drives behind a vehicle in front at a relatively constant desired following distance.

[0023] In a further step, a new (i.e., updated) target following distance for the interpolation point speed is determined based on the actual following distance. The new target following distance can also be determined based on the previous target following distance. Further possible influencing factors when determining the new target following distance are described below.

[0024] Once the new target following distance has been determined, the previous target following distance in the distance-speed characteristic curve is replaced with the new target following distance in a next step. In this process, for example, a correspondingly updated distance-speed characteristic curve that assigns the new target following distance to the interpolation point speed can be saved.

[0025] According to the invention, the new target following distance is determined depending on one or more additional influencing factors (in addition to the detected actual following distance and, if applicable, the previous target following distance). These possible influencing factors are explained in detail below. The influencing factors can appear, for example, within an algorithm in a formula for determining the new target following distance, such as in an equation that specifies the new target following distance. All or just some of the influencing factors can be combined when determining the new target following distance.

[0026] One of the possible influencing factors is a deviation between the actual following distance and the previous target following distance. For example, the deviation in the form of a difference between the recorded actual following distance and the previous target following distance can be included in a formula for determining the new target following distance. However, the term "deviation" should not be understood so narrowly here that it only encompasses a mathematical difference between the recorded actual following distance itself and the previous target following distance itself. It is also conceivable, for example, that a difference between a first term containing the actual following distance (or corresponding to the actual following distance itself) and a second term containing the previous target following distance (or corresponding to the previous target following distance itself) is included in a mathematical determination of the new target following distance.In particular, in the first term, the actual following distance can be provided with a situation-dependent factor, which opens up the possibility of taking external influences and special driving situations into account when learning the desired distance and of adapting the recorded actual distances accordingly for the purposes of learning.

[0027] According to one embodiment, the new target following distance is determined as a function of the deviation between the actual following distance and the previous target following distance, such that the greater the deviation between the actual following distance and the previous target following distance, the more the new target following distance deviates from the previous target following distance. For example, a difference between the actual following distance and the previous target following distance or a difference between the first and second terms (as described in the previous paragraph) can be linearly incorporated into an equation that specifies the new target following distance.

[0028] Another possible influencing factor when determining the new target following distance is a speed difference factor, which increases the closer the actual speed of the vehicle when the actual distance is recorded is to the considered reference point speed. For example, the speed difference factor can be inversely proportional to the absolute value of the difference between the actual speed and the considered reference point speed. The speed difference factor can be capped by a certain maximum value, such as 1, so that it does not assume any values ​​greater than the maximum value or is set to the maximum value if the calculation would result in a value greater than the maximum value depending on the absolute value of the difference between the actual speed and the reference point speed.

[0029] It should be noted that the term "speed difference factor," as used in this description, does not necessarily imply that this influencing variable must be included in a formula for determining the new target distance in the strictly mathematical sense as a factor of a product (although this can be the case according to preferred embodiments). Rather, the term "factor" is to be understood here in accordance with common usage as an influencing factor in the sense of an influencing variable. The same applies to the other influencing variables described below that contain the term "factor." These influencing variables can, but do not have to, be included in a formula as factors of a product in the mathematical sense.

[0030] As a further possible influencing factor, a learning duration factor can be taken into account when determining the new target following distance, which increases over the duration of the respective learning phase (e.g. with each calculation cycle) as long as the recorded actual following distance is within predetermined limits around an initial actual following distance recorded at the beginning of the learning phase and an actual speed that the vehicle has when recording the actual following distance is within predetermined limits around an initial actual speed at the beginning of the learning phase.

[0031] The statement that the actual following distance or the actual speed lies within predetermined limits around a respective initial value is intended to mean that the respective variable lies within a defined interval that encompasses the respective initial value. It is also conceivable that the initial value lies on the edge of the interval.

[0032] The learning duration factor is based on the idea that the longer the vehicle is driven at a certain speed and at a constant distance behind a target object, the more certain it is that this is the actual desired distance. By taking the learning duration into account, the speed of learning can be increased to best meet the driver's needs.

[0033] For example, the learning duration factor can increase linearly with the duration of the learning phase, as long as the above conditions are met.

[0034] For example, it is possible for the learning duration factor to start at zero or another minimum value at the beginning of each learning phase and to be incremented in each phase, provided that the actual speed and the actual following distance are within a respective predetermined interval around a respective speed or following distance value stored at the beginning of the learning phase. It can be provided that as soon as the actual speed or the actual following distance falls outside the respective interval, a new learning phase begins and the learning duration factor is reset to zero or the minimum value.

[0035] There are scenarios in which, based on the available data, everything points to a following sequence, but in reality, the driver has not actually consciously adjusted the distance behind a target object. In such situations, relearning the characteristic curve based on the actual following distance could potentially lead to a deterioration of the characteristic curve.

[0036] In order to counteract this problem, a plausibility value can be taken into account which has been increased or decreased at several previous points in time during the learning phase and / or before the learning phase as a function of a difference between an actual following distance recorded at the respective point in time and a target following distance applicable at that point in time according to the distance-speed characteristic curve in comparison to the last considered point in time.

[0037] The plausibility value, which is explained in more detail below in the detailed description, can be used to determine how reliable the currently learned distance is. Based on this, the learning of incorrect distances that are far from the previously learned distance can be suppressed or slowed down.

[0038] The plausibility value can be calculated for the entire characteristic curve. However, it is also possible to perform the calculation separately for individual or all support points to make the distinction more precise.

[0039] Within the learning algorithm, a plausibility factor can be defined when determining the new target following interval depending on the plausibility value. The plausibility factor decreases with increasing plausibility value and, conversely, increases with decreasing plausibility value. If the plausibility value is very high, the adaptation of the characteristic curve to the respective actual following interval is reduced or slowed down. If, on the other hand, the plausibility level is low, more significant adjustments of the characteristic curve to the current actual values ​​can be made, so that the learned characteristic curve changes more quickly over several calculation steps.

[0040] Another possible influencing factor is a motorway factor, which is greater the longer the driver overrides the automated longitudinal guidance of the vehicle on a motorway or motorway-like road. Driving on a motorway or motorway-like road (such as a German

[0041] Roads (e.g., motorways) can be considered special situations in which it may be useful to accelerate the learning of the characteristic curve by incorporating an additional influencing factor to ensure that the characteristic curve is adapted even when driver assistance is active. This can be achieved by taking the motorway factor into account in the learning algorithm when determining the new target following distance.

[0042] According to one embodiment, the highway factor increases with an oversteer duration during which the automated longitudinal guidance is oversteered. The oversteer duration should not be understood as merely the duration of a specific accelerator pedal actuation by which the driver causes the automated longitudinal guidance to be oversteered. Rather, the oversteer duration refers to the entire duration during which, as a result of an oversteer (which can also be caused, for example, by multiple accelerator pedal actuations at intervals), a modified (in particular, reduced) following distance is set compared to the following distance specified by the automated longitudinal guidance.

[0043] Preferably, the motorway factor depends more than linearly on the override duration. For example, an at least quadratic (in particular a quadratic) dependence of the motorway factor on the override duration can be provided.

[0044] With regard to learning the automated longitudinal guidance by overriding, reference is also made to the applicant's patent application "System and method for learning a following distance for automated longitudinal guidance of a vehicle," filed with the German Patent and Trademark Office on January 15, 2024, the content of which is hereby incorporated into this application. The technical teaching described therein, according to which a learning operation following distance that is increased compared to a normal operation following distance is set in a situation-specific manner, can be advantageously used within the framework of the method described here. This gives the driver more opportunity to set a desired following distance that is reduced compared to the learning operation following distance by overriding the automated longitudinal guidance, thus enabling particularly effective learning of the distance behavior of the automated longitudinal guidance even when the distance control is activated.

[0045] According to one embodiment, the driver is informed by means of a suitable output device, such as a visual display or an acoustic output device, when a learning phase of the learning algorithm for learning a following distance is active. For example, the driver can additionally be informed "in which direction" a learned following distance is currently being adjusted, i.e., whether their current driving behavior leads to an increase or a decrease in a previously stored target following distance (e.g., for the current speed). A second aspect of the invention is a (data) processing device which has at least one processor and is configured to carry out the method according to the third aspect of the invention by means of the at least one processor.According to some embodiments, this may also be a spatially distributed processing device (for example, across multiple spaced-apart processors or microcontrollers). The processing device may, for example, be a control unit (or a part thereof), such as a control unit that controls the automated longitudinal guidance of the vehicle. However, embodiments are also possible in which the processing device is arranged outside the vehicle. It is conceivable, for example, that the processing device is part of a backend server to which the vehicle is connected via a wireless communication link.

[0046] A third aspect of the invention is a system for automated longitudinal guidance of a vehicle, wherein the system is configured to adjust, within the scope of the automated longitudinal guidance, a following distance learned using the method according to the first aspect of the invention. To the extent that the system relies on the method according to the first aspect of the invention, the above and following explanations of the method according to the invention and its possible embodiments can be understood analogously for the system according to the invention, and vice versa.

[0047] According to one embodiment, the system comprises a processing device according to the second aspect of the invention or is data-technically connected to such a processing device—e.g., one located on board the vehicle or in a backend. For example, it can be provided that the system or a control module of the system receives a distance-speed characteristic curve learned or updated by the method from the processing device and, during operation at a given speed, can use it as a basis for a corresponding target following distance for distance control.

[0048] A fourth aspect of the invention is a vehicle, in particular a motor vehicle, with a system according to the third aspect of the invention.

[0049] A fifth aspect of the invention is a computer program comprising instructions which, when executed by a processing device (such as, for example, a processing device according to the second aspect of the invention), cause the processing device to execute a method according to the first aspect of the invention. The computer program can be divided into several separate subprograms, each of which can be executed by different, possibly spatially separated, processing devices (such as, for example, by several separate processors). A processing device according to the second aspect of the invention can thus be configured, in particular programmed, to execute a computer program according to the fifth aspect of the invention.

[0050] A sixth aspect of the invention is a computer-readable storage medium comprising instructions that, when executed by a (possibly distributed) processing device, cause the device to perform a method according to the first aspect of the invention. In other words, a computer program according to the fifth aspect of the invention can be stored on the computer-readable storage medium.

[0051] The invention will now be explained in more detail using exemplary embodiments and with reference to the accompanying drawings.

[0052] Fig. 1 illustrates an example and schematically a system for the automated longitudinal guidance of a vehicle.

[0053] Fig. 2 illustrates exemplary and schematic steps of a method for learning a following distance for automated longitudinal guidance of a vehicle.

[0054] Fig. 3 illustrates an example of a distance-speed characteristic curve including an upper and a lower limit characteristic curve.

[0055] Fig. 4 illustrates an example of a respective temporal learning curve of a target following distance with two different approaches to taking into account a deviation between the actual following distance and the previous target following distance.

[0056] Fig. 5 illustrates an example of an effect of a learning data point on a stored distance-speed characteristic curve.

[0057] In the following, the system 1 shown schematically in Fig. 1 for automated

[0058] Longitudinal guidance of a vehicle is explained with reference to steps 21 to 25 of a method 2 for learning a following distance, illustrated in Fig. 2 in the form of a block diagram.

[0059] The system 1 comprises a distance sensor system 12 for detecting the actual following distance of the vehicle from a vehicle in front (vehicle in front). For example, the distance sensor system 12 can comprise one or more radar sensors. Using a radar sensor, in addition to the following distance, a relative speed between the vehicle and the vehicle in front can also be determined, for example.

[0060] Furthermore, the system 1 comprises a speed sensor system 13 for detecting the actual speed of the vehicle. The speed sensor system 13 can comprise, for example, one or more wheel speed sensors, one or more acceleration sensors (from whose measured values ​​an actual speed can be calculated), and / or a receiver for a global satellite navigation system (where an actual speed can be determined based on a change in the vehicle's position detected by the global satellite navigation system).

[0061] A (data) processing device 10 of system 1 is signal-connected to the distance sensor system 12 and the speed sensor system 13 and is configured to receive a distance signal A comprising the actual following distance from the distance sensor system 12 and a speed signal G comprising the actual speed from the speed sensor system 13. The processing device 10 can also receive, for example, a relative speed signal V from the distance sensor system 12, from which, in combination with the speed signal G, a speed of the vehicle in front can be determined.

[0062] In addition, the processing device 10 is signal-connected to a data memory 11 of the system 1 and is configured to receive characteristic curve information K from it (such as speed values ​​and the target following distance values ​​assigned thereto) according to a distance-speed characteristic curve (hereinafter also referred to as characteristic curve) stored therein.

[0063] The processing device 10 is further configured to determine updated characteristic curve information K' in the manner described in detail below with reference to method steps 21-25 as a function of a detected driving behavior (in particular as a function of detected actual distances and actual speeds) and to output this to the data memory 11 for storage, so that an updated characteristic curve is stored in the data memory 11 as a result.

[0064] The data memory 11 is further connected via signaling to a control unit 14, which is configured to control the automated longitudinal guidance of the vehicle. The control unit 14 can, for example, read characteristic curve information K' from the data memory 11 according to the updated characteristic curve and, depending on the updated characteristic curve information K', generate control signals S for controlling the automated longitudinal guidance, in particular to adjust a target following distance from a vehicle ahead according to the updated characteristic curve information K'.

[0065] In the embodiment shown in Fig. 1, the processing device 10, which executes method 2 described below, is a separate processing device from the control unit 14, which controls the automated longitudinal guidance of the vehicle. The control unit 14 is connected to the processing device 10 via signals and receives from it, for example, the distance and speed information A, G, V as the basis for the longitudinal control of the vehicle.

[0066] However, embodiments are also possible in which one and the same processing device executes method 2 for updating a distance-speed characteristic curve and then controls the automated longitudinal guidance depending on the updated characteristic curve. In other words, with reference to Fig. 1, the processing device 10 could be identical to the control unit 14 or be a part of it (or vice versa).

[0067] The control unit 14 can output the control signals S to a longitudinal guidance actuator 15, in particular to a braking system and / or to a drive module of the vehicle. The longitudinal guidance actuator 15 can then control the automated longitudinal guidance depending on the control signals S and, in particular, regulate a target following distance from a vehicle in front. In Fig. 1, the longitudinal guidance actuator 15 is shown in dashed lines, since it can basically be a generic longitudinal guidance actuator known per se in the prior art, which need not be understood as part of the system 1 according to the invention.

[0068] The processing device 10 is further connected to an output device 16 via signaling. The processing device 10 can generate control signals H for the output device 16 so that it outputs instructions or information related to the learning algorithm to the driver. In Fig. 1, the output device 16 is illustrated schematically and by way of example as a loudspeaker, since an instruction can be output, for example, in acoustic form (e.g., as a voice output). Additionally or alternatively, the output device 16 can comprise, for example, a display for visual information display.

[0069] If a learning phase of a learning algorithm for learning a following distance is active, the driver can be informed of this fact via the output device 16. For example, the driver can also be informed about the "direction" in which a learned following distance is currently being adjusted, i.e., whether the driver's current driving behavior leads to an increase or decrease in a previously stored target following distance.

[0070] The processing device 10 of the system 1 is configured to carry out the method steps 21 to 25 according to Fig. 2 for a plurality of support point speeds of a distance-speed characteristic curve at a plurality of times in a learning phase.

[0071] Before the method steps 21 to 25 according to Fig. 2 are described in detail, a (distance-speed) characteristic curve shown there as an example will be briefly discussed with reference to Fig. 3.

[0072] Fig. 3 shows an example of a characteristic curve in the form of a diagram in which a second interval d is plotted against a speed v. The actual characteristic curve is shown in the diagram in the form of small circles connected by a solid line ("base" in the diagram legend). The data points represented as small circles each assign a second interval to a support point speed (hereinafter also referred to as "support point"). The characteristic curve between the support points results from a (here linear) interpolation.

[0073] In the example characteristic curve shown in Fig. 3, five sampling points are used. The larger the number of sampling points, the greater the customizability of the characteristic curve. Therefore, the use of significantly more than five sampling points would generally be sensible, provided that the available memory allows this in a specific application. Due to the interpolation that takes place anyway between the individual data points, it is not necessary to select steps of 1 km / h or even finer steps between the sampling points to achieve a meaningful result. The most common speed limits such as 30, 50, 80, 100 and 120 km / h, as well as some higher speeds, e.g. 140, 160 or even 180 km / h, are sufficient to have a noticeable effect for the driver. If one adds to this the speed limits of the system (i.e.especially the automated driving function), for example, you get 10 support points to be saved with 10 corresponding distance values ​​to be saved.

[0074] A wide variety of data types can be used to store and process the characteristic curve on a microcontroller. If a uint8 with a maximum value range of 255 is used for the sampling points, all speeds can be covered. A float32, a common data type in the Classic Autosar (CAS) environment, can be used for the second intervals themselves. This results in a memory requirement of 50 bytes for 10 sampling points. Since persistent memory is significantly more complex and expensive than normal stack memory, this memory requirement is comparatively large. However, since the resolution of the intervals is subjective for the driver and not clearly definable anyway, there is no need for the fine resolution of the float32. One approach to reducing the memory requirement is therefore to use a different data type. Here, the option of using an integer value with fixed decimal places is offered.For example, if you choose a uintS which has a value range of 0 to 255 and specify a fixed decimal point before the second decimal place, values ​​from 0 to 2.55 can be mapped. This means that all values ​​up to a theoretical second interval of 2.5 s can be saved with a resolution of one hundredth of a second. The deviations from the actual interval caused by rounding at second intervals, even at typical maximum speeds, e.g. of an ACC function (e.g. 210 km / h), are so small that they are barely noticeable to the driver and in any case do not lead to any noticeable deterioration in function. When using fixed sampling points, the persistent memory requirement can be reduced to 1 byte per sampling point.

[0075] Nevertheless, some information is lost due to rounding and the resolution of the learning algorithm to be used is reduced. Therefore, an internal floating-point variable, such as a float32, can be used for the actual calculation. When the processing device 10 is shut down or when the vehicle is no longer ready to drive, the currently learned value can be rounded and written to the persistent memory. When the system is restarted, this value is then retrieved and used as a basis. Thus, for example, there is only one rounding error per terminal cycle, which consequently has a much less noticeable effect for the driver. However, according to an advantageous embodiment, the individual second intervals may not be learned completely freely, but certain restrictions can be provided within which they must move. In the diagram in Fig.In Figure 3, this is illustrated by a lower and an upper limit characteristic curve (see legend: "Limit"). The learned characteristic curve must be located between the upper and lower limit characteristic curves.

[0076] The upper limit can be determined, for example, by the performance of the sensor technology. A classic radar sensor has a range of around 150 meters, with more advanced components up to 300 meters. If objects are further away, the target object can be lost, which can lead to longitudinal control behavior that the driver finds uncomfortable. If the distant control targets are repeatedly lost, this results in accelerations and braking that are incomprehensible to the driver. For example, in a following vehicle where the target object is traveling at 190 km / h, a following distance corresponding to the maximum visibility of the radar of 150 meters would correspond to a second interval of 2.84 s. To avoid incorrect braking or other undesirable effects, an additional buffer is preferably provided. Therefore, in the embodiment shown in Fig.3, a value of 2.5 seconds is assumed as the maximum interval, which represents the upper limit of the characteristic curve. Depending on the sensor performance, this can be further increased or decreased.

[0077] For the lower limit, legal requirements and standards can be considered as boundary conditions. For example, a distance resulting from the learned characteristic curve should not be less than the distance resulting from a characteristic curve that corresponds to the lowest distance level adjustable by the driver. For example, if an ACC function has three distance levels, the smallest distance level should form the lower limit. The lower limit characteristic curve in Fig. 3 corresponds to such a defined limit.

[0078] In a system with three selectable distance levels, for example, a characteristic curve stored for the middle distance level can be used as a starting point for learning the characteristic curve. This characteristic curve can serve as a familiar reference point for the driver, who may have already driven previous models of the same vehicle brand. Learning in both positive and negative directions (with regard to the following distance) is quickly possible from this basis, so that a positive effect for the driver is noticeable promptly.

[0079] As explained above, certain driving situations or time periods during a journey can be identified as suitable for learning the characteristic curve. This can particularly apply to a relatively constant following journey with regard to a following distance, which is controlled by the driver. In such a situation, a learning phase can be initiated, during which the steps described below are executed at several points in time, e.g., in consecutive calculation cycles.

[0080] In a step 21 of the method 2, a previous (ie previously valid) target following distance d" for an assigned support point speed v n the distance-speed characteristic curve is provided. This provision can, for example, involve reading out the previous target following distance d nfrom the data memory 11, in which various speed values ​​with associated target following distances are stored according to the characteristic curve.

[0081] In a further step 22, a detected actual following distance d ist The actual following distance d ist e.g., received by the processing device 10 with the distance signal A from the distance sensor 12 and made available for further data processing.

[0082] In a further step 23, depending on the actual following distance d ist and the previous following distance d n a new (ie updated) target following distance d n , for the support point velocity v n certainly.

[0083] Then, in a further step 24, the previous target following distance d n for the support point v n in the characteristic curve by the new target following distance d n, replaced. In this process, for example, an updated characteristic curve information K', according to which the support point v n the new target following distance d n , is assigned, are stored in the data memory 11, so that an updated characteristic curve is stored in the data memory 11 as a result.

[0084] In an optional further step 25, control signals S for controlling the automated longitudinal guidance can be generated depending on the updated characteristic curve, in particular to adjust a desired following distance to a preceding vehicle according to the updated characteristic curve information K'. The control signals S can then be output to the longitudinal guidance actuator system 15.

[0085] When determining the new target following distance d n, in step 23, according to an embodiment described in detail below, at each of the several times of the learning phase (e.g. in each calculation cycle) a respective new target following distance d n , based on the respective previously applicable (e.g. determined in a previous calculation cycle) target following distance d n and the current actual following distance d ist determined according to the following equation:

[0086] On the right side of the equation, there is a learning speed factor a, a highway factor f AB , a learning duration factor f t , a speed difference factor f Sv , a plausibility factor f k and a factor which is determined by a deviation between the actual following distance d ist and the previous target following distance d n depends on.

[0087] It should be noted that embodiments are also possible in which an expression for determining the new target following distance d n , - in addition to the previous target following distance d n - only one or some of the above-mentioned additional influencing factors are used.

[0088] In one possible embodiment, at least one of the above-mentioned influencing factors is taken into account, which is determined by a deviation between the actual following distance d ist and the previous target following distance d n depends, such as the last factor in the second term on the right-hand side of the equation above. Preferably, a speed difference factor f Sv and / or a learning duration factor f t and / or a plausibility factor f k According to an advantageous further development, an additional motorway factor f AB join.

[0089] The above-mentioned possible influencing factors are explained in detail below.

[0090] The learning speed factor a can be a constant value which is chosen so that a desired speed of adaptation of the characteristic curve to the detected following distances d ist The higher the learning speed factor a is chosen, the more a detected following distance d ist the support point distance d n , change in the time step under consideration and the faster a (potential) change in the characteristic curve occurs accordingly - viewed over several time steps.

[0091] The factor (d ist / A ext - d n ) takes into account a deviation between the actual following distance d ist and the previous target following distance d n in such a way that the new target following distance d n , compared to the previous following distance d nchanges the more the larger this deviation is. In this embodiment, however, the deviation does not simply flow in the form of a difference between the actual following distance d ist itself and the previous target following distance d n - which would also be a possible variant (ie the above formula with A ext = 1). Rather, the first term of the difference is the actual follow-up distance d ist by a situation-dependent factor A ext divided. The factor A ext opens up the possibility of taking external influences and special driving situations into account when learning the desired distance and of calculating the actual distances recorded accordingly. ist for the purposes of learning. This allows external influences that affect the actual distance d ist compared to the driver's normal desired distance, are taken out of the equation.

[0092] Without such an adjustment via factor A extexternal influences could have a negative impact on the learning success when adapting the characteristic curve. If, for example, a driver is driving in heavy rain and is controlling the longitudinal control manually (or using the accelerator and brake pedals), they will probably choose greater following distances to the vehicles in front than usual because of the rain. This would then be taken into account in the learning algorithm and saved in the characteristic curve. However, if the sun is shining in the vehicle's next terminal cycle and the driver wants to drive with active distance control again, the ACC system will have learned the time gaps for rainy weather that do not correspond to the driver's ideal in this situation. This can cause frustration for the driver because the second intervals are perceived as incorrect. This example shows that the desired following distance depends on environmental conditions such as the weather. The same applies to special driving or traffic situations, such as busy roads.For example, driving in traffic jams or generally situations where the probability of vehicles cutting in is high. Factor A. ext This allows the learning algorithm to differentiate between different driving or traffic situations and / or between different environmental conditions, and to design the distance-speed characteristic curve independently of such external influences so that it reflects standard driver behavior. For example, the factor can be selected to be greater than 1 during a learning phase in rain, such as A. ext= 1.2, to account for the fact that the driver is likely to choose a closer following distance than usual in rainy conditions. If the learning phase takes place while driving in a traffic jam or in a traffic situation where the probability of vehicles merging is high, the factor can be reduced and set to, for example, less than 1. This allows for the fact that the driver is likely to choose a closer following distance than usual (e.g., to prevent vehicles merging).

[0093] As a possible alternative to considering the factor (d ist / A ext - d n ~) or the simple difference (d ist - d n ) when determining the new target following distance d n , according to one embodiment, a deviation between the actual following distance d ist and the previous target following distance d ne.g. be taken into account in the form of a term which in each calculation cycle leads to the new target following distance d n a constant positive value g n added if the actual following distance d ist greater than the previous target following distance d n and the same constant value g n subtracted if the actual following distance d ist smaller than the previous target following distance d n If the actual following distance d ist (within the scope of measurement accuracy) equal to the previous target following distance d n However, the term can be 0, meaning it does not contribute to a change in the target following distance. With this approach, only the sign, but not the magnitude of the distance between the actual following distance d ist and the previous target following distance d n taken into account.

[0094] The diagram in Fig. 4 illustrates two different learning curves based on the two approaches described above for taking into account a distance between the actual following distance d ist and the previous target following distance d n . The upper graph illustrates a temporal learning curve of a second interval d[s] based on the approach described in the previous paragraph, in which the temporal course of the learned second interval d[s] results from the fact that in each calculation cycle the equation

[0095] (with g n = 0.001 s). The lower graph shows a temporal learning curve as a result of the approach described above, in which the temporal course of the learned second interval d[s] results from the fact that in each calculation cycle the equation (with f n= 0.007) is applied. The elapsed time in seconds is plotted on the x-axis, with the learning algorithm under consideration being active between the third and eighth second. The output variable in both cases is a time gap of 1.5 s. The distance to the vehicle in front, actively set by the driver, remains constant at 2.0 s. The learning algorithm is active until second 8, after which the value is reset to illustrate the change. The computing time of the control unit is set at 10 ms, which means that the corresponding formula for calculating the distance was applied once for each of these time steps.

[0096] The upper graph shows that the algorithm d n , = d n + g nIn the 5 seconds of learning, or the 500 calculation cycles, this results in a maximum change of 500 ■ 0.001 s, which corresponds to a time gap difference of 0.5 s. Thus, during this time, the exact second interval set by the driver is adapted. The adjustment occurs linearly over time. The value of 1.9 s, and thus a deviation of 5 percent from the driver's desired interval, is reached after 4 seconds. This corresponds to 80 percent of the learning time.

[0097] In contrast, the algorithm illustrated with the lower graph d n , = d n - d ist - d n ~) ■ f n after the learning period of 5 seconds to a following interval of 1.985 s. In this case, the five percent deviation is already reached after 2.29 seconds and thus less than half the learning time.

[0098] Comparing the two approaches using the graphs reveals several differences. The algorithm underlying the lower graph learns quickly, especially when the time gaps are large. The closer the two values ​​become, the lower the learning potential becomes. Conversely, the upper graph shows a constant change, regardless of whether the difference between the actual and target values ​​is large or small. Applying this relationship in practice, the lower approach can react more quickly to changes in the driver's behavior. The upper one, on the other hand, requires more time.

[0099] In the case of an actually desired learning phase, i.e., in a situation in which the following distance set by the driver is representative of his normal desired distance, the rapid reaction according to the lower graph is advantageous, as an approximation to the driver's desired distance is possible accordingly quickly. In the opposite situation of an incorrect learning trigger, however, the situation is reversed, as here the linear approach according to the upper graph takes longer, depending on the severity. To avoid false triggering at excessively large distances, the plausibility factor f k, which is described in more detail below. This ensures that correspondingly large deviations are not included in the calculation at all. This offsets the disadvantage of the algorithm corresponding to the lower graph, and due to its greater responsiveness when there is a large difference between the stored and current distance at the considered reference point, it can be used advantageously, especially in combination with the plausibility factor.

[0100] Another possible influencing factor when determining the new target following distance d n the speed difference factor f Sv be taken into account. The speed difference factor f Sv ensures a weighting of the influence of a given actual speed v ist recorded actual distance d ist to the various support points v n assigned new target following distances dn ,. On the one hand, this actual speed v ist normally not exactly at a support point velocity v n so that a regulation is needed for how the recorded actual distance d ist at all to the new target following distances d n , at the various support points v n On the other hand, in order to ensure a continuous and natural progression of the learned characteristic curve, it would not be sufficient to simply measure the distance value at the point corresponding to the actual speed v ist to the nearest sampling point. Rather, data points at other sampling points (such as neighboring sampling points) must also be influenced. Therefore, a rule must be found for each of these values ​​to perform the adjustment.

[0101] A simple approach is to influence all support points equally. This means that the calculation rule for the new target distance is applied to all support points as if the current actual distance d ist recorded at exactly these speeds. On the one hand, this would be very simple, as no additional logic is required. On the other hand, this approach has some weaknesses. The added value of the different support points is precisely that the driver can learn different time gaps at different speeds. This advantage would be lost with the simple approach, as after sufficient time, the same distance would apply to all support speeds. The characteristic curve would then be a horizontal straight line with the value of the uniform new target distance.

[0102] Such a constant distance does not correspond to the objective of the distance regulation.

[0103] Experience from industrial applications has shown that the desired behavior of the average driver differs between different speed ranges. Furthermore, a value adjusted to standstill is advantageous in low-speed ranges.

[0104] Consequently, it is advantageous if the target following distances at nearby sampling points (i.e., at sampling point speeds that are close to the actual speed at which the actual distance is recorded) are strongly influenced, whereas target following distances belonging to more distant sampling points are only slightly adjusted to the actual speed. In other words, the influence of the recorded actual distance on the target distances at the various sampling points should vary depending on the distance of the respective sampling point from the current vehicle speed. The closer these two are to each other, the greater the corresponding effect. This can be achieved algorithmically by adjusting the speed difference factor f Sv is defined so that it is greater the closer the actual speed v ist , which the vehicle uses to detect the actual distance d istat the considered support point velocity v n lies.

[0105] For example, the speed difference factor f Sv inversely proportional to an amount of a difference from the actual speed v ist and the respective considered support point velocity v n be:

[0106] The speed difference factor f Sv be capped by a predetermined maximum value, such as 1 , so that it does not take on values ​​greater than the maximum value or is set to the maximum value if, according to the calculation depending on the amount of the difference between the actual speed v ist and the support point velocity v n would result in a value greater than the maximum value.

[0107] The diagram in Fig. 5 shows an example of the effect of a new learning data point (here: d ist = 1.7 s at v ist= 30 m / s), which is marked in the diagram by a single cross, to a stored distance-speed characteristic curve. The initial characteristic curve is the lower graph, which is referred to as "Base" in the legend. The upper graph shows the new resulting characteristic curve ("Result"), which was created after the learning phase. Using the graphical representation, all support points can be viewed and their respective effects evaluated: The value at 20 m / s is closest to the 30 m / s driven and thus also has the largest upward change in value. If one compares this with the next higher data point at 50 m / s, the change is smaller here, despite the same initial learning value. This was the goal to be achieved, since the closest data point should be adjusted in particular. The point at 10 m / s shows an even smaller change than the value at 50 m / s, despite the same distance. This is not due to the speed difference factor f Sv, but rather by the fact that the original value is already closer to the learned value. This becomes even clearer at the value at 0 m / s, which changes only minimally in comparison due to the distance and the already small difference in seconds. Basically, the graph shows that the adjustments are having the desired effect. Reference points that are close to the corresponding value learn faster than those that are further away. This fulfills the requirement that close speed ranges should be learned comparatively more quickly. However, the reference points that are further away are also influenced so that a meaningful progression is achieved. Excessive jumps, which would otherwise occur, would not reflect real driver behavior.Rather, finding a suitable following distance according to the algorithm is a continuous process, as would be the case with a real driver.

[0108] Another possible influencing factor when calculating the new target following distance d n , using the learning duration factor f t The time a learning phase has already lasted must be taken into account. The longer the vehicle drives at a constant distance behind a target object, the more certain it is that this is the actual distance desired by the driver. By taking the learning duration into account, the speed of learning can be increased to best meet the driver's needs.

[0109] For example, it can be provided that the learning duration factor starting from 1 f t increases linearly with the duration of the learning phase:

[0110] The size t twith each journal (ie in each calculation cycle) by a constant increment a t increased, so that the value t in the next journal is:

[0111] In this case, t t start at 0 and be capped at a maximum value, such as 1 . In this

[0112] In this case, the learning duration factor f t between 1 and 2. If the driver drives for a long time within the corresponding detection limits of the following vehicle, the actual distance d recorded will be ist with a maximum change rate for adapting the characteristic curve. The learning duration factor f t a maximum of doubling the adjustment in each journal.

[0113] However, there may be special cases which can be addressed by the implementation of the learning duration factor f described above. tnot be taken into account. For example, if the driver drives behind the same target object for a long time and slightly increases the distance d ist , the learning phase will remain active. However, the same distance d will not be maintained over the entire period. ist learned, but rather a change occurs over the duration of the learning phase. The system therefore does not learn a constant distance. It is therefore advantageous to store a logic that increments the value of t t only permitted within a limited framework. This can be achieved by temporarily storing an initial-actual following distance d t0 and an initial actual speed v driven at the beginning of the learning phase t0 Only as long as the actual distance d ist and the actual speed v ist in a certain delta (öd max or 8v max ) um the initial actual follow-up distance d t0or the initial actual speed v t0 stay around, the size t t and thus the learning duration factor f t continues to increment. The following constraints apply to the increment:

[0114] \ is ~ ^to l < ^max

[0115] I ^act ^t0 | < public transport max

[0116] Only as long as these conditions are met, t h e.g. according to the equation t = ti + a h further increased. If this is no longer the case, a new learning phase begins and t t is set back to 0 (and accordingly the learning duration factor f t to its minimum value 1). The current values ​​of distance and speed are set as initial values ​​for the beginning of the learning phase, so at the beginning of the new learning phase:

[0117] You can then start counting up again.

[0118] There are scenarios in which, based on the data, everything points to a following, but in reality, the driver has not actually consciously adjusted the distance behind a target object. This usually occurs due to apparent following, which deviates significantly from the driver's actual desired distance.

[0119] In an example scenario, the driver is driving in an urban area at 50 km / h, in accordance with the speed limit. Another vehicle is traveling further away, also adhering to the speed limit. Neither deviates significantly from this and thus selects the same desired speed. However, the driver of the learning vehicle is not deliberately following the vehicle in front at a specific distance; rather, the first priority here is adhering to the speed limit. Since both vehicles are constantly following one another, the accelerations are minimal, and the speeds of the vehicles do not differ from each other, this situation initially appears to the learning algorithm as a relevant following journey.

[0120] Another example scenario is following behind a tractor that has dropped parts of its load. The driver doesn't want to come into contact with these parts and therefore maintains a significantly greater distance. The vehicle's sensors cannot detect such a situation, and there is no other way to identify the cause of this increase in distance.

[0121] In both example scenarios, the following distance does not correspond to the driver's desired distance, as recorded in other situations. This can quickly lead to significant errors in the stored characteristic curve. Since the driver is far away from the vehicle in front, the learned characteristic curve is also adapted quickly. These examples clearly demonstrate the trade-off between quickly learning the characteristic curve when preferences are unknown and a rather slow adaptation to avoid errors.

[0122] At the beginning of a learning cycle, the desired distances are unknown, and the stored characteristic curve must be adjusted quickly. The longer the driver is on the road, the more information is already known about the driver's request. On this basis, the plausibility or credibility of the previously learned value can be determined. If the current distance deviates too significantly from the previously learned value, the learning process can be prevented. It would therefore be advantageous to exclude values ​​that are too far removed from the learning process based on the previously learned empirical values ​​and the duration of the corresponding learning process. This can prevent situations in which the driver does not follow the vehicle in front, as well as other special situations, from distorting the learning of the learned desired distance behavior. For this purpose, a plausibility value k can be introduced. This can be, for example,be defined as a value between 0 and 1 with a fixed resolution, where 0 indicates that no learning has taken place and 1 represents maximum plausibility. The plausibility value k should not only depend on how long the learning took, but also on the extent to which the actual intervals during learning correspond to the time gaps already learned. The plausibility value k can increase if the current intervals are very similar to the time gaps already learned, or decrease if new current intervals deviate significantly from what has been learned so far. The plausibility value k therefore does not merely represent a learning duration, but can rather be interpreted as a statement about the reliability of the currently stored characteristic curve.

[0123] The increment or decrement of the plausibility value k can be determined depending on a deviation between a recorded actual following distance d istand a characteristic curve for the speed v learned up to this point ist applicable target following distance d gelernt be determined. If this difference is small during an active learning phase, the plausibility value k is increased more significantly. If the difference is large, the plausibility value k is increased by a smaller value or even decreased. If the geiernt then due to the learning algorithm to d ist then k increases again.

[0124] In addition, a range of values ​​around d gelernt The value in which the plausibility k increases or decreases can be determined by specifying a fixed distance value d a where the plausibility k should decrease if the actual following distance d ist by more than the distance value d a from the previously learned following distance d geiernt deviates.

[0125] Algorithmically, a calculation of an updated plausibility value k' based on a previous plausibility value k (e.g. determined in a previous calculation cycle) can be realized as follows:

[0126] The rear part describes a function which depends on the difference between the learned distance d geiernt and the current actual distance d ist This function has its zero exactly at the set distance value d a By choosing the factor a dIt is possible to specify how quickly the value of k should actually change. As mentioned above, the plausibility value k can be limited, for example, to the value range between 0 and 1 to allow for further learning. Preferably, the plausibility value k is stored persistently across multiple learning phases (and possibly across multiple driver journeys) and continuously increased or decreased accordingly.

[0127] However, it should be noted that the driver's distance preferences are not identical on every trip. Emotions and stress can influence the current perception of distance. A plausibility k that may have been at its maximum on the last trip may not necessarily be high on the next trip.

[0128] Therefore, according to one embodiment, the stored plausibility value k is limited. This will be illustrated by an example scenario: During the last trip, the driver largely stayed within the same value range close to a following interval of 2 seconds, so that the plausibility value k is close to 1 at the end of the trip, which rules out larger distance deviations. If the driver now wants to drive a smaller interval of approximately 1.3 seconds during a new trip, this lies outside the limit range and cannot be learned. This problem can be solved by reducing the plausibility value k to, for example, 0.7 after each trip end (or at the start of a new trip). In the example scenario, the current interval in seconds would then be back within the range that enables learning.This solution gives the driver greater flexibility at the start of the journey, allowing them to respond to changing distance requirements. However, the limitation or resetting of the plausibility value k after the end of the journey (or at the start of a new journey) should be set at a level that prevents significantly incorrect scenarios from being learned.

[0129] Overall, the plausibility value k can be used to determine how reliable the currently learned distance is. Based on this, the learning of incorrect distances that are far from the previously learned distance can be suppressed. Furthermore, rapid learning at the beginning of a learning cycle, for example, at the start of a new journey, is possible.

[0130] The plausibility value k can be calculated for the entire characteristic curve. However, it is also possible to perform the calculation separately for individual or all support points to make the distinction more precise. Within the learning algorithm, the new target following distance d can be used to determine the new target following distance. n , a plausibility factor should be included, e.g. as factor f k in the determination equation introduced above. To define the plausibility factor f k The plausibility value k described above can be used. This provides information about whether the previously stored value d n meaningful or negligible in individual cases. This information can be used to influence the learning rate.

[0131] If the plausibility value k is low, the previously stored value is usually not very reliable. This could be because the system has been newly set up and no learning data is available so far. In this situation, it is sensible to learn the characteristic curve relatively quickly. For example, the driver logs into a new vehicle with his user profile. By default, the distance characteristic curve can be reset and with it the value of k, which would then be 0, for example. If the driver subsequently drives in traffic, the system's aim is to store a meaningful value in a short space of time in order to recognise his driving behaviour as quickly as possible. If the driver wants to use the distance control after a short time, the set distance should then be at least approximately within the desired range.

[0132] If the plausibility factor k is high, the values ​​stored in the characteristic curve are already meaningful because they have been learned over a longer period of time and have also proven to be accurate. The speed at which the characteristic curve is adapted within the learning algorithm can therefore be reduced. For example, if the driver has been using the driving profile for a long time and has covered a long distance, relatively high plausibility values ​​are stored. On the trip under consideration, the driver confirmed these again, and the value of k was accordingly increased to the maximum value of 1. In this case, in the case of slightly different new distances, learning should not be carried out directly with full extent, but only for longer deviations, in order to avoid unnecessarily falsifying the already valid stored result.

[0133] It is therefore useful to use the plausibility factor f kSelect it so that it decreases with increasing plausibility value k and, conversely, increases with decreasing plausibility value k. If the plausibility level k is very high, the adaptation of the characteristic curve to the respective actual following interval is reduced. If, on the other hand, the plausibility level k is low, more significant adjustments of the characteristic curve to the current actual values ​​can be made, so that the learned characteristic curve changes more quickly over several calculation steps.

[0134] This behavior can be explained with a definition of f k be achieved, after which f k decreases linearly with the plausibility value k, as in the following equation: f k = 2 - k, ke [0,1]

[0135] This means that the plausibility factor f k As a result of a minimum plausibility value (k = 0), a learning speed twice as high as with maximum plausibility (k = 1) can be set.

[0136] Currently, the main application area for automated driving is the highway. Assisted driving is frequently used for long periods of time. However, since the driver always has the option to override the function (e.g., by oversteering) and temporarily take over the driving task, the current time lag can be learned even when the driver assistance function is activated.

[0137] If a driver activates the adaptive cruise control system and exceeds the threshold several times without noticing a real change in the distance to the vehicle in front, the driver could prematurely deactivate the system or switch to a standard distance level. Therefore, it is sensible to provide a logic for this special case that enables significantly faster adaptation to the learning target. This can be achieved by using the learning algorithm to determine the new target following distance d n, a motorway factor f AB which is greater the longer the automated longitudinal guidance of the vehicle is oversteered by a driver of the vehicle on a motorway or motorway-like road.

[0138] For a concrete design of the motorway factor f AB Several criteria need to be considered. On the one hand, this additional function should preferably only be active when the vehicle is on a highway or a highway-like road. In urban traffic, periods of exceeding the speed limit may frequently occur, which do not necessarily apply to the distance target. Furthermore, the usage rate in urban traffic is not as high as on other road types.

[0139] Furthermore, this decision should preferably not be binary. That is, the highway factor f ABshould not only be effective based on the location and the activity of the violation, but also with an override duration t ueb , during which the automated longitudinal guidance is overridden. Since a large impact is necessary in long learning phases, a quadratic relationship is chosen, for example, which increases significantly more after a certain time, compared to the learning period t t For example, the motorway factor f AB can be determined using the following equation: fAB O-AB ' ueb' t ueb G [0, t max ]

[0140] Here, a AB a fixed value to normalize the time and t ueb is the duration of the violation, which for the purposes of algorithmic processing is included in the motorway factor f AB at a maximum value t max can be capped. The duration of the violation t ueb can be calculated analogously to the above formula for the learning duration t tbe calculated, ie here too, at the beginning of the learning phase, a d t0 and a v t0be set, but preferably with an additional time period at the beginning. This is helpful in order to actually set the initial value to the correct one and to avoid any incorrect learning maneuvers. Otherwise, the additional factor may subsequently abort the adaptation acceleration despite only slight deviations. During this additional time period, the driver can adjust his or her desired distance by exceeding the limit. This takes a certain amount of time, while the vehicle must still accelerate accordingly in order to maintain the driver's constant desired distance. As soon as all conditions for learning are met, the driver actively overrides the accelerator pedal, and a determination of the road type actually results in a motorway or motorway-like road, a corresponding timer is started. Only when this timer expires will t ueb to 0, as well as d t0 and v t0set to the current values. The calculation of the motorway factor f then begins AB analogous to the formula given above for the learning duration t ( .

Claims

Patent claims 1. A computer-implemented method (2) for learning a following distance for automated longitudinal guidance of a vehicle, wherein the following steps are carried out for at least one support point speed of a distance-speed characteristic curve during a learning phase: Providing (21) a previous target following distance for the Support point speed; Providing (22) a detected actual following distance; Determining (23) a new target following distance for the Support point speed as a function of the actual following distance; and replacing (24) the previous target following distance in the distance-speed characteristic curve with the new target following distance; wherein the determination (23) of the new target following distance is carried out as a function of one or more elements from the list, which includes the following: a deviation between the actual following distance and the previous target following distance; a speed difference factor that is greater the closer an actual speed that the vehicle has when detecting the actual following distance is to the support point speed;a learning duration factor which increases over the duration of the learning phase as long as o the recorded actual following distance lies within predetermined limits around an initial actual following distance recorded at the start of the learning phase and o an actual speed which the vehicle has when recording the actual following distance lies within predetermined limits around an initial actual speed at the start of the learning phase; a plausibility value which has been increased or decreased at several previous points in time during the learning phase and / or before the learning phase as a function of a difference between an actual following distance recorded at the respective point in time and a target following distance applicable at the respective point in time according to the distance-speed characteristic curve compared to the last point in time considered;a motorway factor which is greater the longer the driver of the vehicle oversteers the automated longitudinal guidance of the vehicle on a motorway or motorway-like road; 2. Method (2) according to claim 1, wherein the determination (23) of the new target following distance is carried out as a function of the deviation between the actual following distance and the previous target following distance, in such a way that the new target following distance deviates more from the previous target following distance, the greater the deviation between the actual following distance and the previous target following distance.

3. Method (2) according to one of the preceding claims, wherein the determination (23) of the new target following distance is carried out as a function of the learning duration factor, wherein the learning duration factor increases linearly with the duration of the learning phase.

4. Method (2) according to one of the preceding claims, wherein the determination (23) of the new desired following distance is carried out as a function of the speed difference factor, wherein the speed difference factor is inversely proportional to an amount of a difference between the actual speed and the support point speed.

5. Method (2) according to one of the preceding claims, wherein the determination (23) of the new desired following distance is carried out as a function of the motorway factor, wherein the motorway factor increases with an override duration during which the automated longitudinal guidance is overridden.

6. Method (2) according to claim 5, wherein the motorway factor depends more than linearly on the override duration.

7. Processing device (10) with one or more processors, wherein the processing device is configured to carry out the method (2) according to one of the preceding claims by means of the processor or by means of the plurality of processors.

8. System (1) for the automated longitudinal guidance of a vehicle, wherein the system is configured to set a following distance learned by means of the method (2) according to one of claims 1 to 6 within the scope of the automated longitudinal guidance.

9. Vehicle with a system (1) according to claim 8.

10. A computer program comprising instructions which, when executed by a processing device (10), cause the processing device (10) to carry out a method (2) according to one of claims 1 to 6.

11. A computer-readable storage medium comprising instructions which, when executed by a processing device (10), cause the processing device (10) to carry out a method (2) according to one of claims 1 to 6.

Images