Vehicle speed control device and method for controlling the speed of a vehicle

Abstract

A driving control device (180) for an ego vehicle (100) that performs a distance-based cruise control function (ACC) to control the ego vehicle (100) in a following journey, depending on the journey of a preceding target vehicle (101), comprising a detection section (181) that detects the movement of the preceding target vehicle (101); a first motion prediction section (182) that determines a trajectory (108) of the ego vehicle (100); a loss prediction section (183) that predicts a target object loss (ZOV) of the preceding target vehicle (101) for a future time and categorizes the predicted target object loss (ZOV) as a trivial target object loss (t-ZOV) or a non-trivial target object loss (nt-ZOV);a second motion prediction section (184) that predetermines a virtual trajectory (109) of the target vehicle (101) moving ahead, deforms this trajectory in a way relevant for detection, and a speed control section (185) that regulates the speed of the ego vehicle (100).

Description

Technical area

[0001] The invention relates to vehicles in general, in particular to subunits that automatically control the vehicle speed and, in particular, to a method for controlling the speed by means of a corresponding cruise control device. background

[0002] Cruise control devices designed to automatically regulate vehicle speed or to prevent exceeding or maintaining an arbitrarily set speed are known, for example, as adaptive cruise control (ACC) systems. A corresponding distance and speed control function (ACC) is defined, for example, within an SAE Level 1 driver assistance function (Society of Automotive Engineers) as a function designed to automatically regulate the distance between a controlled vehicle, or the ego vehicle, and a vehicle traveling ahead, according to a predetermined time interval and the current speed. A corresponding cruise control device is also known as a cruise control system with longitudinal distance control.

[0003] Referring to Fig.1A - 1C, within the ACC function of an ego vehicle (100), input variables are processed that originate from the sensor system (103) installed in the respective vehicle (100) for detecting the movement of the ego vehicle (100) and its surroundings. The loss of detection of a preceding vehicle (101) or target vehicle is known as target object loss (TOV). In the event of target object loss (TOV), the requested vehicle speed is, for example, discontinuously regulated to an arbitrarily determined, higher target speed, thus initiating acceleration of the ego vehicle (100).Due to physical detection limits of the vehicle's internal detection sensors (103), for example, influenced by the distance of objects or obstacles (105) from the ego vehicle (100) or the angular / spatial coverage of the vehicle's surroundings, such as laterally offset driving or the geometry of the roadway (102) itself, a target object loss (ZOV) can occur within the vehicle control device, even though the preceding target vehicle (101) remains in the immediate vicinity and in the same lane of the roadway (102) of the ego vehicle (100). This occurs particularly when cornering and driving over crests or during laterally offset driving. If the preceding target vehicle (101) is detected (107) by the vehicle's internal detection sensors (103), this can lead to increased driving dynamics impairments due to decelerations or jerks, and thus to restrictions in driving comfort and driving safety for the vehicle occupants.Corresponding cruise control systems are therefore subject to continuous development to improve comfort and safety. Furthermore, briefly accelerating to a higher target speed and then braking in the event of a loss of target object (ZOV) with subsequent re-detection is uneconomical. State of the art

[0004] From the published patent application DE 10 2005 032 182 A1 a cruise control device is known, wherein in a ZOV the acceleration of the ego vehicle is prevented for a certain termination time, wherein the termination time depends on a lateral acceleration of the vehicle to be controlled.

[0005] From the published patent application DE 10 2007 031 544 A1 a cruise control device is known, wherein during a transition from a following journey, depending on the travel of a preceding target vehicle, to a free journey, without a preceding target vehicle, the target speed of the cruise control device is set to the value of the vehicle speed immediately before the transition.

[0006] From the published patent application DE 10 2007 036 417 A1 a cruise control device is known, wherein an event area is determined by means of predictive route data, wherein in the case of a ZOV within the event area the acceleration of the ego vehicle is prevented if a target speed of the vehicle specified by the driver as the desired speed is higher than the current actual speed and if the vehicle and the target object were within the event area immediately before the ZOV occurred.

[0007] From the published patent application DE 10 2016 211 208 A1 a cruise control device is known, wherein a ZOV is determined by predicting a lane change of the target vehicle and wherein, when a ZOV is predicted by a lane change, a time in the future is predetermined at which the lane change will be completed, wherein a further time in the future is predetermined which lies before the time at which the lane change will be completed and which determines when an acceleration of the ego vehicle is initiated.

[0008] From the published patent application DE 10 2017 007 504 A1, a cruise control device for an ego vehicle is known, which is configured to perform a distance-based cruise control function (ACC) with respect to a preceding target vehicle, which can be replaced by a primary cruise control depending on detected weather and road conditions.

[0009] The cruise control devices known from the state of the art presented focus primarily on intervening in the cruise control system or in the driving dynamics of the ego vehicle as a reaction to a ZOV or a correspondingly detected driving situation. Due to this reactive behavior, the described control measures of the ego vehicle can only be executed from the time the corresponding driving situation is detected at the earliest and are therefore dependent on the technical properties of the sensor technology. Brief description

[0010] The object of the present invention is to improve the detection of a driving situation, in particular a loss of target object (ZOV) and the recognition of the target object, as well as to predict it early and speculatively, thereby reducing the reaction time for said reactive control and regulation measures by the cruise control device according to the invention and increasing driving comfort, driving safety, and driving economy. This object is achieved by a cruise control device according to claim 1, a method for controlling the speed of a vehicle according to claim 11, and a vehicle according to claim 14 in that the movement of a preceding target vehicle (101) traveling in front of the ego vehicle (100) in a direction of travel is detected, and wherein a first trajectory (108) of the ego vehicle (100),as a function of the cruise control function (ACC), and a target object loss (ZOV) of the preceding target vehicle (101) is predetermined for a future point in time as a function of the detected movement of the preceding target vehicle (101) and the calculated trajectory (108) of the ego vehicle (100), wherein the predicted target object loss (ZOV) is categorized, wherein a second, virtual trajectory (109) of the preceding target vehicle (101) is determined as a function of the detected movement of that preceding target vehicle (101), the calculated trajectory (108) of the ego vehicle (100) and the predetermined target object loss (ZOV), wherein the speed of the ego vehicle (100) is determined as a function of the predetermined target object loss (ZOV), its categorization and the determined virtual (109) trajectory of the preceding vehicle (101) is regulated., Brief description of the drawings Fig. 1A - 1C show a driving situation that leads to a non-trivial target object loss. Fig. 2A shows the schematic structure of a vehicle comprising the cruise control device according to the invention. Fig. 2B shows the schematic structure of the cruise control device according to the invention. Fig. Figure 2C shows the partial schematic structure of a computer system. Fig. Figure 3 shows the structure of the encoder-decoder LSTM model for route prediction. Fig. Figure 4 shows the structure of an RNN-LSTM model for detecting and categorizing target object losses. Fig. Figure 5 shows the velocity and acceleration curve of a virtual trajectory with a non-trivial target object loss. Fig.6 shows the process flow of the method according to the invention for controlling the speed of a vehicle. Fig. 7A - B shows an embodiment of the method according to the invention according to a schematically illustrated driving situation. Detailed description

[0011] First, the schematic structure of the ego vehicle (100) is described, which comprises the cruise control device (180) according to an embodiment. Fig.2A shows a block diagram illustrating an exemplary schematic structure of the ego vehicle (100) that includes the inventive cruise control device (180) according to one embodiment. The individual elements shown in the block diagram are not intended to be exhaustive; rather, the illustrated embodiment of the ego vehicle (100) according to the invention is limited to elements that are functionally connected to the inventive cruise control device (180) according to this embodiment.For example, an ego vehicle (100) according to the invention comprises one or more of the elements listed below, such as a drive unit (120), a power transmission (130), a chassis (140), a vehicle control device (150), comprising, for example, a drive control device (160), a power transmission control device (170), a cruise control device (180) according to the invention, and vehicle-internal sensors, such as a detection sensor system (103) and a vehicle sensor system (141).

[0012] According to some embodiments, a drive unit (120) may comprise technical components such as an internal combustion engine, an electric machine, a hybrid drive train, or other technical components designed to provide mechanical or electrical power for driving the ego vehicle (100) according to the invention. According to some embodiments, a power transmission (130) may comprise a clutch, a transmission, a cardan shaft, a differential, and a chassis (140), drive shafts, and wheels. The power transmission (130) and chassis (140) are essentially designed to convert the mechanical or electrical energy output by the drive unit (120) into driving the ego vehicle (100) or to transmit the power of the drive unit (120) to the wheels.

[0013] According to some embodiments, the detection sensor system (103) may comprise technical components such as a camera, radar or LiDAR sensors, but also infrared or ultrasonic sensors, or other electromagnetic or acoustic sensors configured to detect the surroundings of the ego vehicle (100). The detection sensor system (103) is understood here as the entirety of technical components that realize the detection of the surroundings of a vehicle when applying a conventional ACC function.

[0014] According to some embodiments, the vehicle sensor system (141) may comprise technical components such as a rotational speed sensor, a speed sensor, an acceleration sensor, a wheel speed sensor and others, which are configured to detect state variables of the ego vehicle (100).

[0015] The drive control device (160) and the power transmission control device (170) comprise technical components, which are also known as "ECU" ("electronic control unit"), which are essentially designed to receive signals from further components of the ego vehicle (100) and to process them into further signals, control or regulation variables or other status information.

[0016] In one embodiment, the drive control device (160) receives, for example, information from the cruise control device (180), comprising a value for the target speed of the ego vehicle (100), wherein the drive control device (160) in turn outputs control signals to the drive unit (120) in order to control it to output a corresponding power required to achieve the desired target speed. In a further embodiment, the power transmission control device (170), for example, also receives corresponding information about a desired target speed and outputs control data to the power transmission (130) in order to control it such that the power output by the drive unit (120) is converted into a required wheel speed or a corresponding wheel torque in accordance with the desired target speed of the vehicle.The exemplary embodiments described here are not to be understood as exhaustive; rather, they serve to illustrate the functional interrelationship of the technical components of the ego vehicle (100) according to the invention, which are shown in extracts.

[0017] The cruise control device (180) according to the invention of the ego vehicle (100) comprises all the technical components required to execute a distance-based cruise control (ACC) function. Different input signals from other technical components of the ego vehicle (100), such as the vehicle sensor system (141) and the detection sensor system (103), are received and further processed into output signals, which are transmitted, for example, to the drive control unit (160) and the power transmission control unit (170) in order to execute the distance-based cruise control (ACC) function. One aspect of the cruise control function (ACC) comprises controlling the speed of the ego vehicle (100) in a following journey, depending on the movement of a preceding target vehicle (101). A fundamental distinction is made here between cruise control in a following journey and cruise control in a free-floating journey.A following drive involves controlling the speed of the ego vehicle (100) to a value equal to the speed of a detected, preceding target vehicle (101) while maintaining a predefined distance from that target vehicle (101). A free drive involves accelerating the ego vehicle (100) to a preset target speed and maintaining this speed. Restrictions arise, for example, when cornering or in other detected situations in which acceleration can be blocked or a current speed reduced.

[0018] Fig.2B shows a block diagram illustrating an exemplary schematic structure of the cruise control device (180) according to the invention. According to one aspect of the invention, the cruise control device (180) comprises a detection section (181), a first movement prediction section (182), a loss prediction section (183), a second movement prediction section (184), and a speed control section (185). According to some embodiments of the cruise control device (180), these sections can comprise both software and hardware components, such as sensors or other measurement technology configured to generate measurement signals, but also processors or other computing technology to process corresponding measurement signals and other information, or even computer-implemented programs executed within the individual sections.Furthermore, computer-implemented programs or other software-implemented elements of the individual sections may be part of the same hardware or software components.

[0019] Referring to Fig.2C, the cruise control device (180) according to the invention and / or individual sections thereof, in one aspect of the invention, each comprise completely or partially elements of a computer system (200). Such a computer system (200) can comprise one or more of the following elements, such as an internal communication device (201), a computing device (202), a memory device (204) for transient and persistent data storage, such as a ring buffer, and a network interface (206). The internal communication device (201) can comprise a bus or a bus system, such as a peripheral PCI architecture bus (“peripheral component interconnect”). The internal communication device (201) is configured to exchange data and information between the respective elements of the computer system (200).The computing unit (202) can comprise a processor, which is designed, for example, as a CPU ("central processing unit") or any other type of processor, such as a microprocessor, a digital signal processor, or a microcontroller, which is suitable for carrying out calculations associated with the method according to the invention. The storage device (204) can comprise one or more working memories, such as RAM ("random access memory"), data memories, such as ROM ("read only memory"), and mass storage devices. For example, instructions such as arithmetic operations, general functional sequences, or algorithms contained in the computer program can be loaded into the RAM memory of the storage device (204) of the computer system (200), which a processor of the computing device (202) executes.The storage device (204) may further comprise any type of mass storage device intended for storing all types of information and data, such as multiple hard disks, floppy disks, optical disks, semiconductor memories, non-volatile memories, EPROM, EEPROM, flash memories, magnetic disks, removable storage devices, or other storage media. The network interface (206) may comprise various communication protocols, such as "Transmission Control Protocol / Internet Protocol" (TCP / IP) or "Hyper Text Transfer Protocol" (HTTP), or others suitable for providing external communication connections to the computer system (200) via remote access and / or computer networks.

[0020] According to one aspect of the invention, the cruise control device (180) comprises a detection section (181) that detects the movement of the preceding target vehicle (101). The detection of the preceding target vehicle (101), or also referred to below as the target object (ZO), is carried out by conventionally known technical peripherals in order to execute existing ACC functions. In one aspect of the invention, the detection section (181) comprises technical components of the detection sensor system (103), such as one or more of the components mentioned below, such as cameras, RADAR or LIDAR sensors, or other optical or acoustic measuring technology, which is configured to detect the surroundings of the ego vehicle (100), and of the vehicle sensor system (141), which is configured to detect operating parameters of the ego vehicle (100).In some embodiments, the detection section (181) is configured to detect the movement of the ego vehicle (100) and the movement of one or more preceding target vehicles (101) and to identify a target object (ZO) therefrom.

[0021] In one aspect of the invention, the detection of the movement of a preceding target vehicle (101) comprises the identification of the target object (ZO) as such and its tracking by the detection sensor system (103). For this purpose, calculations are performed in the detection section (181) to identify the target object and determine the position of the ego vehicle relative to it. The position of the ego vehicle (100) is essentially described by its position, for example within a global coordinate system or, after map comparison, in a local North-East Down (NED) coordinate system, as well as by its orientation. The position of the ego vehicle (100) influences the possible target objects (ZO) that can be detected by the detection sensor system (103).

[0022] In one aspect of the invention, the detection of the movement of a target vehicle (101) comprises determining the lateral offset of potential target objects (ZO) relative to the ego vehicle (100). For this purpose, the azimuth angle Φ max During the application of the radar, the radar signal continuously covers an angular range in front of the ego vehicle (100), whereby the azimuth angle Φ max and from this, the lateral offset of the reflecting target objects (ZO) is determined via the distance and geometric conditions.

[0023] In one aspect of the invention, the vehicle sensor system (141) comprises one or more of the following sensors, such as a wheel speed sensor, which is designed to determine the wheel speed and, in conjunction with this and the dynamic tire radius, the speed of the vehicle, a steering wheel angle sensor, in order to determine the steering angle δ H to determine, a rotation rate sensor to determine the yaw rate ψ of the ego vehicle (100) and a lateral acceleration sensor to determine the lateral acceleration a y,max to determine.

[0024] In one aspect of the invention, detecting the movement of the ego vehicle (100) comprises determining the current course curvature κ of the roadway within which the ego vehicle (100) is moving. The course curvature κ influences the detection range (104) of the detection sensor system (103) of the ego vehicle (100) when cornering. The course curvature κ corresponds to the reciprocal of the curve radius R and, in one aspect of the invention, is determined according to equation 1a. κ=1R

[0025] In an alternative embodiment, the course curvature κ(κ S ) via the steering angle δ H , determined according to equation 1b. Here, the course curvature κ S the steering angle δ H , divided by the product of the steering ratio i SG with the wheelbase I, multiplied by the addition of the square of the current speed v x , divided by the square of the characteristic velocity vchar and 1. The characteristic speed of the ego vehicle (100) characterizes the self-steering behavior at low lateral accelerations. κS=δH(iSG∗l)∗(1+vx2vcar2)

[0026] In an alternative embodiment, the course curvature κ(κ ψ ) via the yaw rate ψ of the ego vehicle (100), according to equation 1c. Here, the course curvature κ ψ the quotient of the first time derivative of the yaw rate ψ and the square of the current speed v x . κψ=ψ˙vx2

[0027] In an alternative embodiment, the course curvature κ(κ ay ) is determined via the lateral acceleration of the ego vehicle (100), according to equation 1d. Here, the course curvature κ ay the quotient of the lateral acceleration a y and the square of the current speed v x . κay=ayvx2

[0028] In an alternative embodiment, the course curvature κ(κ v ) is determined from the difference in wheel speed of the left and right wheels Δv of the ego vehicle (100), according to equation 1e. Here, the course curvature κ v the quotient of the difference in wheel speed of the wheels on the left and right side Δv and the product of the current speed v x and the track width b of the ego vehicle (100). κv=Δvvx∗b

[0029] In some embodiments, at least one of the aforementioned calculation methods is used to determine the course curvature κ, whereby the selection of the corresponding calculation method may be influenced by one or more of the following influencing factors, such as crosswind, road slope, radar radius, measurement sensitivity at low speeds, measurement sensitivity at high speeds, or offset drift. In an alternative embodiment, information from a global or local coordinate system is used to determine the course curvature, such as GPS data from the ego vehicle (100).

[0030] In a further aspect of the invention, the detection of the movement of the ego vehicle (100) by the detection section (181) comprises the detection of the characteristic properties of the cruise control by the ACC function. In one embodiment, the vehicle acceleration x of the ego vehicle (100) x¨i+1 determined according to equation 2a. Here, the vehicle acceleration x of the ego vehicle (100) i+1, depending on time t, x¨i+1(t) the temporal translation of the vehicle acceleration x¨ of the target vehicle (101) x¨i from the difference between time t and a preset time gap τ set . x¨i+1(t)=x¨i(t−τset)

[0031] In an alternative embodiment, the vehicle acceleration of the ego vehicle (100) x¨i+1 determined according to equation 2b. Here, the vehicle acceleration of the ego vehicle (100), depending on time t, x¨i+1(t) the quotient of the difference in the speed of the target vehicle (101) i, as a function of time t, x˙i(t) and the speed of the ego vehicle (100), depending on the time t, x˙i+1(t), and the time period τ v , where the time span τ v corresponds to the time within which the speed difference v rel , between the preceding target vehicle (101) and the ego vehicle (100) is to be compensated. x¨i+1(t)=x˙i(t)−x˙i+1(t)τv=vrelτv

[0032] In an alternative embodiment, the vehicle acceleration of the ego vehicle of the (100) x¨i+1 determined according to equation 2c. Here, the vehicle acceleration of the ego vehicle (100), depending on time t, x¨i+1(t) the difference between the speed difference v rel and the quotient of the difference of the target distance d set , of the ego vehicle (100) to the preceding target vehicle (101), and the actual distance d, and the time constant for controlling the distance τ d , divided by the time span τ v . x¨i+1(t)=vrel−dset−dτdτv

[0033] In a further aspect of the invention, the detection of the movement of the ego vehicle (100) by the detection section (181) comprises the detection of the curve speed during free travel, i.e. without a target object detection (ZOE) and depending on the radar visibility, according to equation 3a. Here, the curve speed v c,p (τ pre , κ, Φ max) the quotient of twice the azimuth angle Φ max and the product of the curve curvature κ and the time gap τ pre , where the azimuth angle Φ max corresponds to the angle between the direction of travel and the target object and the time gap τ pre the time for a potential target object approach is taken into account. vc,p(τpre,κ,ϕmax)=2∗ϕmaxκ∗τpre

[0034] Here, equation 3b corresponds to the maximum range of the radar system d max,eff . dmax,eff=2∗ϕmaxκ

[0035] In an alternative embodiment, the curve speed during free travel, i.e. without target object detection, is determined as a function of the maximum lateral acceleration a y,max , determined according to equation 3c. Here, the curve speed v c,ay (κ, a y,max ) the square root of the lateral acceleration a y,max and the course curvature κ. vc,ay(κ,ay,max)=ay,maxκ

[0036] In one aspect of the invention, the identification of the target object (ZO) by the detection section (181) comprises the assignment of a driving corridor within which the control of the ACC function takes place. In some embodiments, the driving corridor comprises the determined course curvature κ for the course prediction of the ego vehicle (100) and a corridor width b corr .

[0037] In an alternative embodiment, the assignment of a driving corridor comprises the use of a variable driving corridor, wherein the detection device (181) detects whether or not one or more secondary lanes are present next to the driving corridor of the ego vehicle (100). Secondary lanes can be identified, for example, by stationary, but also oncoming vehicles, or by those vehicles that are moving with a lateral offset but in the same direction of travel as the ego vehicle (100). By detecting the additional vehicles, the lateral offset of those vehicles can be continuously statistically evaluated and, depending on this, the corridor width b corr be adjusted.

[0038] In an alternative embodiment, the assignment of a travel corridor comprises the application of a local hysteresis depending on the target object detection by the detection device (181). If a target object (ZO) is detected, the corridor width b corr This increases the tolerance to keep a target object (ZO) in detection, even during dynamic movements of the ego vehicle (100) and the preceding target vehicle (101).

[0039] In an alternative embodiment, the assignment of a driving corridor comprises the application of a temporal hysteresis depending on the target plausibility. The target plausibility describes a value between zero and one, with the value one being assigned the highest plausibility. The closer a detected target object (ZO) is to the core of the driving corridor or to the center line of the course curvature κ of the ego vehicle (100), the higher the target plausibility, or in other words, the more likely the detected object is a target object (ZO) to which the ACC function regulates. If the target plausibility exceeds a value of 0.4, the detected object is recognized as a target object. If the target plausibility falls below a value of 0.2, the detected target object is no longer considered as such.The target plausibility thus contributes to determining the entry or exit of potential target objects from the area of ​​the driving corridor of the ego vehicle (100).

[0040] In an alternative embodiment, the assignment of a driving corridor comprises the application of a temporal hysteresis depending on the change in target plausibility. The evaluation of the change in target plausibility contributes to the detection and / or categorization of a potential target object loss (ZOV). A fundamental distinction can be made between trivial ZOV (t-ZOV) and non-trivial ZOV (nt-ZOV). A trivial ZOV (t-ZOV) is defined as a ZOV in which the target object (ZO) is no longer detected by the detection sensor system (103) for more than 10 seconds. A non-trivial ZOV (nt-ZOV) is defined as a ZOV in which the target object (ZO) is detected again within 10 seconds of a loss of detection. Different driving situations or driving maneuvers of the preceding target vehicles (101) can be detected by means of the change in target plausibility.For example, if the preceding target vehicle (101) simply changes lanes, the target plausibility will decrease rather slowly, so the change will be small. During a turning maneuver or a zero-clearance maneuver due to cornering, the target plausibility will decrease more quickly than during a lane change, and the change will be correspondingly greater.

[0041] In an alternative embodiment, the assignment of a driving corridor includes the application of a speed-dependent maximum distance for target objects (ZO). In this way, objects with a correspondingly greater distance from the ego vehicle (100) are less likely to be considered as target objects (ZO).

[0042] In an alternative embodiment, the assignment of a driving corridor comprises one or more of the following applications, such as a variable driving corridor, a local hysteresis, a target plausibility, a change in target plausibility, and a maximum distance. If multiple potential target objects are detected, the assignment of the driving corridor comprises one or more of the following evaluation criteria, such as the evaluation of the smallest longitudinal distance to the ego vehicle (100), the smallest distance to the course center or the course curvature κ of the ego vehicle (100), or the lowest required target acceleration of the ego vehicle (100).

[0043] In summary, the detection section (181) thus comprises technical components of a detection sensor system (103) and a vehicle sensor system (141), wherein the detection sensor system (103) comprises one or more components, such as a camera, a RADAR or a LIDAR sensor, and wherein the vehicle sensor system (141) comprises one or more of the components such as a wheel speed sensor, a steering wheel angle sensor, a yaw rate sensor and a lateral acceleration sensor.

[0044] Here, the detection section (181) is characterized according to the invention in that the detection of the movement of the ego vehicle (100) includes the determination of the current course curvature κ, the vehicle acceleration x¨ and the curve speed v c,p and that the identification of a target object (ZO) comprises the assignment of a driving corridor to the ego vehicle (100), depending on one or more calculation variables, such as a corridor width b corr, a local hysteresis of the corridor width b corr , a target plausibility, a change in the target plausibility, a speed-dependent distance, a distance to the course center of the ego vehicle (100) and a target acceleration of the ego vehicle (100).

[0045] In particular, the detection section (181) is characterized according to the invention in that the detection of the movement of the ego vehicle (100) includes the determination of the current course curvature κ, the vehicle acceleration x¨ and the curve speed v c,p wherein the current course curvature κ is determined as a function of one or more of the physical input variables determined by means of the detection sensor system (103), such as the steering angle δ H , the yaw rate ψ, the lateral acceleration a y and the wheel speed, whereby the vehicle acceleration x¨ is determined as a function of one or more control signals such as, a preset time gap τset , a speed difference v rel , between the ego vehicle (100) and the preceding target vehicle (101) and a target and actual distance d set or d between the ego vehicle (100) and the preceding target vehicle (101) is determined, wherein the curve speed v c,p depending on a preset time gap τ pre , the current course curvature κ and the determined azimuth angle Φ max is determined, and that the detection of the movement of a target vehicle (101) determines the distance of the ego vehicle (100) to the target vehicle (101), depending on the azimuth angle Φ determined by means of a radar sensor max and that the identification of a target object (ZO) includes the assignment of a driving corridor to the ego vehicle (100), depending on one or more calculation variables, such as a corridor width b corr , a local hysteresis of the corridor width b corr, a target plausibility, a change in the target plausibility, a speed-dependent distance, a distance to the course center of the ego vehicle (100) and a target acceleration of the ego vehicle (100).

[0046] According to a further aspect of the invention, the cruise control device (180) according to the invention comprises a first movement prediction section (182) which determines a virtual trajectory of the ego vehicle (100) for a point in time in the future from the detected movement of the ego vehicle (100) and as a function of the cruise control function (ACC).

[0047] In one aspect of the invention, the determination of the virtual trajectory of the ego vehicle (100) comprises determining its movement, in the form of a predicted course curvature κ pred, for a future point in time. In some embodiments, determining the predicted course curvature κ pred the assumption that the trajectory curve determined by the detection section (181) as a snapshot of the course curvature κ is maintained, the current course curvature κ being the predicted course curvature κ pred , for a future point in time. In an alternative embodiment, determining the predicted course curvature κ pred the evaluation of the movement of at least one target object (ZO) detected by means of the detection section (181) and the derivation of the predicted course curvature κ pred by detecting the course curvature κ ZO of the target object (ZO) and its temporal progression. Taking into account a definable temporal offset τ ZO, due to the movement of the target object (ZO) and the ego vehicle (100), within the recorded temporal course of the course curvature κ ZO of the target object (ZO), the predicted course curvature κ pred defined as the course curvature κ ZO (t+ / -τ ZO ) of the target object (ZO) at the time of the offset. In an alternative embodiment, the predicted course curvature κ pred derived from the current GPS data of the ego vehicle (100).

[0048] In an advantageous embodiment, determining the predicted course curvature κ pred the prediction of the route of the ego vehicle (100) for a point in time in the future. Fig.Figure 3 shows the structure of an encoder-decoder LSTM model according to the invention for predicting the route. In one aspect of the invention, the encoder-decoder LSTM model is configured to receive two-dimensional x- and y-coordinates of the previous route of the last 100 meters of travel, detected by means of detection sensors (103) or position data from global or local coordinate systems, such as GPS data, as input variables, and to use these to predict a 50-meter route of the ego vehicle (100). For this purpose, the input layer comprises pre-filtered x- and y-coordinates as position data of the previous 100 meters, with the size of each sample being (100, 2).

[0049] The encoder network receives the data from the input layer and creates a vector that describes the internal representation of the input sequence, with the length of the vector corresponding to the number of units in that layer. The encoder network comprises an LSTM (“Long Short Term Memory”) cell with 50 units. The decoder network also comprises an LSTM cell with 50 units and transforms the learned internal representation of the input sequence from the encoder network into an output sequence. The output sequence comprises the x- and y-coordinates of the 50-meter route to be predicted.Because the encoder network produces a [2x100] matrix as output, the length of which is determined by the number of memory cells of the LSTM layer, and the decoder network comprises an LSTM layer that expects a three-dimensional input comprising samples, time steps, and features to obtain a decoded sequence, the encoder-decoder-LSTM model additionally includes a repeat vector as an intermediate layer. The repeat vector is configured to repeat the vector obtained from the encoder network to increase the dimension of the outputs for the decoder network's inputs. The output layer comprises a dense layer as a TimeDistributed wrapper. Each individual layer is activated with the hyperbolic tangent function.

[0050] In one aspect of the invention, the creation of the encoder-decoder LSTM model according to the invention comprises training with a number of 700 to 1500 epochs, a learning rate of 0.0005, a gradient descent algorithm ("Adam optimizer") as the optimizer, and the mean square deviation as the loss function. The training further comprises encoding the input data into two states by the encoder network, repeating the first state for the number of time steps to be predicted by the repeat vector, decoding the thus determined initial states by the decoder network, and applying a dense layer to it to determine the number of features required for each time step.

[0051] In one aspect of the invention, training the encoder-decoder LSTM model includes preparing the training data. In some embodiments, GPS positions are acquired as measured data, from which longitude and latitude information is subsequently extracted to obtain x- and y-coordinates as two-dimensional NED position data. The GPS position data extracted in this manner are checked to determine whether the latitude information corresponds to ∈ [-90° 90°] and the longitude information corresponds to ∈ [-180° 180°]. The extracted position data is converted into time series data at a sampling rate of 1 Hz, which is then converted as a function of the distance traveled. This results in the representation of the position data as GPS coordinates and ultimately NED coordinates depending on the distance traveled. The distance traveled corresponds to the summed distances between the individual position data.Optionally, the position and distance data are interpolated and resampled to ensure that all data points are uniformly spaced one meter apart (sampling homogenization). Furthermore, origin points are set at intervals of 3,000 meters along the distance to avoid major deviations between the two-dimensional description and the actual position. The first origin point is arbitrarily set to a reference point on the Earth's surface within the WGS 84 ellipsoid model. The conversion of geodetic coordinates to NED coordinates aims to achieve a projection of GPS-based position data onto a fixed tangential plane. The preparation of the training data for training the encoder-decoder LSTM model involves the application of a sliding window algorithm to divide the entire training data into 150-meter-long sequences.Within a 150-meter-long sequence, a window is then defined for a fixed number of time steps, with a number n = 100 (meters) of data points serving as input variables for training to predict the number m = 50 (meters) of data points, with the sum of the numbers n and m defining the length of the window. Additionally, an offset is defined, with the window shifting by the offset on the time series data for each iteration of training within the 3000-meter-long training sequence (“sliding”), so that new input and new output variables are continuously trained.

[0052] In an alternative embodiment of the route prediction, the encoder-decoder LSTM model is configured to receive two-dimensional x- and y-coordinates of the previous route of the last 100 meters of travel, detected by means of detection sensors (103) or position data of global or local coordinate systems, such as GPS or NED data, as input variables and to predict 50 meters of the route of the ego vehicle (100) therefrom, wherein the last 100 meters of route are formed from the last 50 meters of the actually driven route and the predicted 50 meters of the previous route prediction.

[0053] From the knowledge of the future route, the future direction of movement of the ego vehicle (100) is thus also known in an advantageous manner, so that from this and information from operating parameters of the drive unit (120) and the power transmission (130), such as the vehicle speed and acceleration, the predicted course curvature κ pred , can be predetermined based on geometric conditions and independently of external signals such as GPS data.

[0054] Thus, the first movement prediction section (182) according to the invention is characterized in one embodiment in that it has a predicted course curvature κ pred for a future point in time as the virtual trajectory (108) of the ego vehicle (100), where the predicted course curvature κ predis derived from the prediction of the route of the ego vehicle (100). The first movement prediction section (182) comprises an encoder-decoder-LSTM network, with an input layer for samples of dimension (100, 2), an encoder layer comprising an LSTM cell with 50 units, a RepeatVector comprising a length of 50 units, a decoder layer comprising 50 units, and an output layer comprising a dense layer with 50 units, wherein each layer is activated with the hyperbolic tangent function and which is configured to receive two-dimensional x and y coordinates of the previous route of the last 100 meters of travel, detected by means of detection sensors (103) or position data of global or local coordinate systems, as input variables and to predict 50 meters of the future route of the ego vehicle (100) therefrom.

[0055] According to a further aspect, the cruise control device (180) according to the invention comprises a loss prediction section (183) which predetermines a target object loss (ZOV) of the preceding target vehicle (101) for a future point in time as a function of its detected movement and the virtual trajectory of the ego vehicle (100), wherein the virtual trajectory characterizes the movement state of the ego vehicle (100), and categorizes that target object loss (ZOV) into a trivial target object loss (t-ZOV) or a non-trivial target object loss (nt-ZOV).

[0056] According to one aspect of the invention, the loss prediction section (183) comprises a recurrent neural network with LSTM cells (RNN-LSTM) that predicts a target loss (ZOV) and categorizes it into a trivial target loss (t-ZOV) or a non-trivial target loss (nt-ZOV). Fig.Figure 4 shows the exemplary structure of the RNN-LSTM model for detecting and categorizing target object losses (ZOV). The inventive RNN-LSTM model comprises a recurrent neural network with four layers, comprising an input layer, two hidden layers, and an output layer. The input layer is a dense layer comprising a number of units corresponding to a fixed number of input variables. The input variables correspond to signals that are generated and processed within the vehicle system of the ego vehicle (100). These can be obtained, for example, via bus systems of the vehicle system, such as the CAN bus or Flexray or other bus architectures. Table T1 shows a selection of possible (bus) signals that can be used as input variables of the inventive RNN-LSTM model. (T1) Nr. Designation Description S01 ACC target acceleration Target acceleration value of the ego vehicle requested by the ACC function (100) S02 ACC target object Target object detected by ACC function (ZO) S03 relative speed Relative speed between target object (ZO) and ego vehicle (100) determined by sensor data fusion from image processing and radar S04 Distance to target vehicle Longitudinal distance between target object (ZO) and ego vehicle (100) determined by sensor data fusion from image processing and radar S05 Speed ​​Ego ESP signal for the absolute speed of the ego vehicle (100) S06 Acceleration Ego Absolute longitudinal acceleration of the ego vehicle (100) S07 Acceleration Gearbox longitudinal acceleration of the ego vehicle converted by the drive train in the transmission (100) S08 Target drive torque Target value for the drive torque of the engine / drive unit (120) of the ego vehicle (100) S09 Actual wheel torque Torque of the ego vehicle (100) transmitted to the wheels by the power transmission (130) S10 Measurement time sensor technology Current timestamp of the sensor data fusion S11 Drive torque limitation dynamic limitation of the drive torque of the engine / drive unit (120) of the ego vehicle (100) S21 Number of detection cycles Number of Flexray cycles since initialization or tracking of the target object S12 Acceleration rate Ego Change in the acceleration of the ego vehicle S16 Actual drive torque Actual value for torque of the drive unit (120) of the ego vehicle (100) S19 Length distance of page objects Longitudinal distance of the ego vehicle (100) to objects in the vehicle environment S23 Orientation of ego vehicle Direction of movement, pitch and yaw angle of the ego vehicle (100) S20 Position page objects Position of objects in the vehicle environment S14 Position side objects, left side track Position of objects in the left adjacent lane of the ego vehicle (100) S13 Position of side objects, right side track Position of objects in the right adjacent lane of the ego vehicle (100) S18 Radial distance side objects Radial distance of the ego vehicle (100) to objects in the vehicle environment S15 Target drive torque filtered Filtered target value for the drive torque of the engine / drive unit (120) of the ego vehicle (100) S17 Target clutch torque Target value for torque of the power transmission (130) of the ego vehicle (100) that the drive unit (120) must apply S22 Angle side objects Radial angular distance of the ego vehicle (100) to objects in the vehicle environment S23 Reference speed Reference speed of the ego vehicle (100)

[0057] In some embodiments, all of the 23 signals listed in Table T1 (S01 - S23) are used as input variables, wherein the input layer and the stretched layers of the RNN-LSTM model comprise a number of 23 units.

[0058] In a preferred embodiment, the first 11 of the signals (S01 - S11) listed in Table T1 are used as input variables, comprising the signals of the ego vehicle (100) such as the ACC target acceleration (S01), the ACC target object detection (S02), the relative speed between the ego vehicle (100) and the target object (ZO) (S03), the distance to the target vehicle (S04), the speed of the ego vehicle (100) (S05), the acceleration of the ego vehicle (100) (S06), the acceleration of the ego vehicle (100) which is converted by the power transmission (130) (S07), the target drive torque of the drive unit (120) (S08), the actual wheel torque (S09), the time stamp for the detection sensor system (103) (S10) and the drive torque limitation (S11).

[0059] In a particularly preferred embodiment, the first 6 of the signals listed in Table T1 (S01 - S06) are used as input variables, comprising the signals of the ego vehicle (100) such as the ACC target acceleration (S01), the ACC target object detection (S02), the relative speed between the ego vehicle (100) and the target object (ZO) (S03), the distance to the target vehicle (S04), the speed of the ego vehicle (100) (S05) and the acceleration of the ego vehicle (100) (S06).

[0060] The selection of the 23 bus signals listed in Table T1 as input variables for the inventive RNN-LSTM model is made during the production of the loss prediction section (183) of the inventive cruise control device (180). In one aspect of the invention, the production of the cruise control device (180) comprises the preselection of six basic signals, which correspond to physical input variables of the RNN-LSTM model. These six basic signals result from information from the ACC function, which describes the current control intervention in the movement of the ego vehicle (100), sensor data from the detection sensor system (103), which provides information about the movement of the target object (ZO), and vehicle-internal information about the current movement of the ego vehicle (100) itself.In one embodiment, the ACC target acceleration (S01) and the ACC target object detection (S02) comprise preselected signals related to the ACC information; the relative speed (S03) and the distance to the target vehicle (S04) comprise preselected signals related to the movement of the target object (ZO); and the speed (S05) and acceleration (S06) of the ego vehicle (100) comprise preselected signals related to the movement of that ego vehicle (100). The ACC target acceleration (S01) indicates the target value currently stored in the ACC function for the control intervention on the acceleration of the ego vehicle (100), which is to be considered representative of the entire control intervention of the ACC function in the vehicle control system.The ACC target object detection (S02), on the other hand, provides a value that is representative of the state of the target object detection (ZOE) and thus provides information about the control behavior of the ACC function, related to free travel or following travel. The relative speed (S03) between the ego vehicle (100) and the target object (ZO) as well as the distance to the target vehicle (S04) provide basic information about the relative movement of the ego vehicle (100) and the target object (ZO). The speed (S05) and the acceleration (S06) of the ego vehicle (100), in turn, provide information about the movement of the ego vehicle (100) in the immediate future. All six basic signals are required to detect or predict the tracking of the target object (ZO) and the detection of its loss.

[0061] In a preferred embodiment, the 17 remaining signals from table T1 are selected during manufacture of the cruise control device (180) by analyzing the correlation of all signals present in the vehicle system to the 6 preselected signals. In some embodiments, all bus and Flexray signals available in the vehicle system are evaluated using regression methods. In one embodiment, the regression method comprises a regression model based on the random forest approach. In an alternative embodiment, the regression method comprises a model based on the feature importance selection approach, which represents a special sub-form of the random forest approach.In some alternative embodiments, the regression method comprises any other known method for selecting from a plurality of signal profiles those that correlate with further preselected signal profiles. In one embodiment, the signals listed in Table T1 result from a feature importance ranking (FIR) of operational design domain (ODD)-specific, trained random forests, where the ODDs can be: city driving, country road driving, or highway driving. The FIR is created from ∼ 10 5A listed ranking of signals is created for each ODD, ordered by decreasing correlation to the six defining, preselected signals. For each ODD, three lists of 100 signals each are created using FIR for sequences of 30 s, 60 s, and 90 s, resulting in a total of nine lists. To ensure robust, ODD- and sequence-independent signal selection, the intersection of all nine lists is then determined. The result is the 17 remaining signals listed in Table T1.

[0062] In one aspect of the invention, the production of the cruise control device (180) comprises producing training data for producing the RNN-LSTM model of the loss prediction section (183). In some embodiments, the production of training data for producing the RNN-LSTM model of the loss prediction section (183) comprises collecting measurement data, extracting relevant signals from the measurement data, preprocessing the training data, and implementing a sliding window algorithm on the preprocessed training data.

[0063] In one aspect of the invention, the collection of measurement data comprises recording continuous time series data, which originate in particular from the detection sensor system (103) and the vehicle sensor system (141). Within the time series data, the time stamps for the ego vehicle (100) are marked when a target object detection (ZOE) and a target object loss (ZOV) occur, for example, by evaluating the target vehicle detection signal of the ACC function (S02). The marked time series data generated in this way are then divided into smaller sequences, each containing one or more target object losses (ZOV). The divided sequences comprise different data types from different sources, such as sensor data fusion (SDF), radar, image processing, other sensors, but also signal data internal to the vehicle system, whereby in known vehicle systems, typically several tens of thousands of measurement signals can be generated.The sampling rate for most measurement signals is typically 40 ms (milliseconds). Measurement signals with a different sampling rate can be adjusted to 40 ms, for example, during data processing. Within the sequences, smaller segments are defined that encompass individual target loss events (TOV) and / or target detection events (TOE). For this purpose, segments with a length of 32 individual time steps of 40 ms each, with a total length of 1.24 seconds, are generated from the sequences. The individual segments are further categorized into trivial target loss events (t-TOV) and non-trivial target loss events (nt-TOV).To categorize as non-trivial target object loss (nt-ZOV), conditions are checked, such as the dead time (Δt ≤ 10s) until a new target object detection (ZOE), exclusively four-wheeled target objects (ZO) and both the ego vehicle (100) and the preceding target vehicle (101) as target object (ZO) must not blink.

[0064] In one aspect of the invention, the extraction of relevant signals comprises the previously described feature importance analysis using a regression model based on the random forest approach to determine the correlation of the totality of signals available within the recorded measurement data to basic, preselected signals.

[0065] In one aspect of the invention, the preprocessing of the training data comprises resampling, scaling, one-hot encoding, or sine transforming the measurement data, as well as dividing the entire data set and applying the sliding window algorithm. In one embodiment, the resampling of the measurement data comprises adjusting the sampling rate of all signals in the measurement data. In a preferred embodiment, the resampling of the measurement data comprises a sampling rate of 40 ms. In one embodiment, the scaling of the data comprises applying a MinMax scaler to adjust all value ranges of the signals to a uniform value range. A uniform value range is particularly necessary for training the RNN-LSTM model, since different value ranges can have varying influences on the result of the neural network's calculations.In one embodiment, one-hot encoding of the measurement data comprises translating labeled signals into numerical values. For example, individual signals that characterize one or more states of a system or subsystem can be provided with text characters instead of numerical values. However, this is not suitable for training neural networks, so numerical values ​​are assigned to the corresponding text-based states during one-hot encoding. In one embodiment, sine transformation of the measurement data comprises applying a sine transformation function to periodic signals to prevent signal jumps over the course of the time series data. In one embodiment, the entire set of data sets corresponding to the entire set of measurement data comprises dividing it into a training data set, a validation data set, and a test data set.In a preferred embodiment, 90% of the measurement data is used as training dataset, 5% as validation dataset and 5% as test dataset.

[0066] In some embodiments, applying the sliding window algorithm to the measurement data involves dividing the sequences into windows across the individual time steps of the sampling. In some embodiments, different windows are formed within the sequences, comprising an input window, a shift parameter, and a prediction window. During model training, the time steps of the input window are fed to the neural network to be trained as input variables. The shift window counts the time steps for which the prediction is to be calculated, and the prediction window then corresponds to the prediction to be made. In each iteration of training, the windows are then shifted by one time step in the time series of the sequence, and a new prediction is performed.In one embodiment, a sequence within which a target object loss (ZOV) occurs comprises 32 time steps of 40 ms each. In a preferred embodiment, the input window comprises 8 time steps, the offset parameter comprises 6 time steps, and the prediction window comprises one time step. Thus, in the first training iteration, time steps 1–8 are fed to the model as input, with the model now being asked to make the prediction for the 14th time step. In the second training iteration, time steps 2–9 are fed to the model as input, and it is asked to make the prediction for the 15th time step. Until the training now predicts the 32nd time step, the model now runs through 19 training iterations or windows.Within the 32 time steps, the target object loss occurs at an arbitrary location, which the model is expected to predict within the correct window during training. This procedure is then repeated for all training data.

[0067] In one aspect of the invention, the production of the cruise control device (180) comprises training the RNN-LSTM model of the loss prediction section (183) using the preprocessed data, comprising sequences with a length of 32 time steps. In some embodiments, the training of the RNN-LSTM model of the loss prediction section (183) comprises the optimization of hyperparameters of the model. A distinction is made between constant and variable hyperparameters. In one embodiment, the constant hyperparameters include an input length of 8 time steps, an offset parameter (prediction horizon) of 6 time steps, a prediction length of 1 time step, a sequence length of 32 time steps, a number of training epochs of 50, the activation function of the output layer as a softmax function, balanced class weights, and a loss function as categorical cross entropy.

[0068] In one embodiment, the hyperparameters to be optimized include the number of LSTM units of the first hidden layer, the number of LSTM units of the second hidden layer, the learning rate, the weight initialization distribution, the optimizer type, the regularizer for the first hidden layer, the batch size, and the type of activation functions.

[0069] In one aspect of the invention, the trained RNN-LSTM model of the loss prediction section (183) comprises the following hyperparameters as listed in Table T2. (T2) Nr. Designation Value 1. Number of LSTM units first hidden layer 256 2. Number of LSTM units 64 3. Batch size 128 4. initialization Glorot normal distribution 5. optimizer Adam 6. Activation of the hidden layers Selu 7. Learning rate 0,001 8. Sample weighting (t-ZOV and nt-ZOV) 0,50904 9. Class weighting Balanced 10. Sample weighting (n-ZOV) 0,2545 11. Regularizer (I2) 0,04 12. Dropouts between hidden layers 0,6

[0070] According to some embodiments, the categorization of target object losses (ZOV) includes trivial target object losses (t-ZOV) and non-trivial target object losses (nt-ZOV), as well as implicitly no target object losses (n-ZOV). In general, a target object loss (ZOV) is considered the state of no longer detecting a preceding target vehicle (101) or target object (ZO) by the ACC function and thus by the detection sensor system (103). The state of detection or no longer detecting the target object (ZO) can be implemented as a signal within one or more control units of the ego vehicle (100), preferably the drive control device (160) and the inventive cruise control device (180). Target object losses (ZOV) can be characterized by their cause and type.A target object loss (ZOV) can be caused, for example, by external circumstances of the vehicle's surroundings, such as the geometry of the route, obstacles, or detection limits of the detection sensors (103), or by the behavior of one or more target objects (ZO) themselves, such as leaving the driving corridor due to a lane change, a turning maneuver, or even by preceding target vehicles (101) entering or exiting gaps. In addition, the target object (ZO) can be detected again after a target object loss (ZOV) and at a time interval therefrom, leading to a renewed target object detection (ZOE). For the purposes of the invention, trivial target object losses (t-ZOV) are preferably those target object losses (ZOV) that are attributable to the behavior of the target objects (ZO), whereby the lost target objects (ZO) are not detected again at a characteristic time interval from the target object loss (ZOV).In contrast, non-trivial target object losses (nt-ZOV) within the meaning of the invention are those target object losses (ZOV) that are preferably attributable to the external circumstances of the vehicle's surroundings, and wherein the lost target objects (ZO) are detected again within a characteristic time interval. In some embodiments, the characteristic time interval is between time t. out the target object loss (ZOV) and the time t in of target object detection (ZOE) 10 seconds, so that Δt = t in - t out ≤ 10s for a non-trivial target loss (nt-ZOV) and correspondingly Δt > 10s for a trivial target loss (t-ZOV).

[0071] In an alternative embodiment, the target object losses (ZOV) are additionally categorized according to the vehicle class of the preceding target vehicle (101) as the target object (ZO), with cars, trucks, and at least four-wheeled vehicles representing a condition for a non-trivial target object loss (nt-ZOV), and all remaining vehicles being assigned to trivial target object losses (ZOV). In one embodiment, the categorization of the target object losses (ZOV) according to the vehicle class comprises the evaluation of a signal from the detection sensor system (103) for the vehicle class. This signal can be a camera-based signal that, based on image recordings, categorizes vehicles at least into cars or trucks, or four-wheeled vehicles and other vehicles.

[0072] In an alternative embodiment, the target object losses (ZOV) are additionally categorized according to the state of the turn signal of the ego vehicle (100) and the corresponding target object (ZO), wherein a detected active turn signal of the ego vehicle (100) and / or the target object (ZO) is evaluated as an indication of a lane change and thus categorized as a non-trivial target object loss (nt-ZOV). In some embodiments, the state of the turn signal of the ego vehicle (100) is evaluated via a bus signal already present in the vehicle system regarding the state of the turn signal. In some embodiments, the state of the turn signal of the target object (ZO) is evaluated via the evaluation of camera images from the detection sensor system (103), which can detect the state of the turn signal of one or more target objects (ZO) using image recognition methods.

[0073] The loss prediction section (183) according to the invention is thus advantageously configured to receive at least the 6 basic signals from the ego vehicle (100) mentioned above, comprising the ACC target acceleration (S01), the ACC target object detection (S02), the relative speed between the ego vehicle (100) and the target object (ZO) (S03), the distance to the target vehicle (S04), the speed of the ego vehicle (100) (S05) and the acceleration of the ego vehicle (100) (S06) as input variables by means of an RNN-LSTM model, and to predict therefrom a target object loss (ZOV) for 6 time steps of 40 ms in the future and to categorize the predetermined target object loss as trivial target object loss (t-ZOV) or non-trivial target object loss (nt-ZOV). Fig.Figure 5 shows the example velocity and acceleration curve of a virtual trajectory with a non-trivial target object loss (nt-ZOV). The exit time t out indicates the target object loss (ZOV) by the detection section (181) and the entry time t in indicates the target object (re-)detection (ZOE) by the detection section (181). The time interval between t out and t in , also known as Δt = t in - t out representable, corresponds to the dead time, whereby, if Δt ≤ 10s applies, a non-trivial target object loss (nt-ZOV) is present. For this purpose, the loss prediction section (183) is now configured to permanently evaluate the historical, recorded measurement data and signals for a past period using the RNN-LSTM model according to the invention and to predict the target object loss at a time t pa possible target object loss (ZOV) for an exit time t out In a preferred embodiment, the period for the evaluation of the historical, recorded measurement data and signals corresponds to the length of the input window of the sliding window algorithm, comprising the 8 time steps before the prediction time t p . Accordingly, in that preferred embodiment, the prediction horizon comprises the 6 time steps of the prediction window, which represent the distance between t p and t out define, where the exit time t out corresponds to the prediction window. In one aspect of the invention, the loss prediction section (183) determines the exit time t out predetermined.

[0074] In summary, the loss prediction section (183) according to the invention is characterized in that it receives the time series data of at least six of the measurement data and signals recorded during the journey by the detection sensors (103) and the vehicle sensors (141) of the ego vehicle (100), comprising the ACC target acceleration (S01), the ACC target object detection (S02), the relative speed between the ego vehicle (100) and the target object (ZO) (S03), the distance to the target vehicle (S04), the speed of the ego vehicle (100) (S05) and the acceleration of the ego vehicle (100) (S06) as an input variable and uses this to predict a target object loss (ZOV) for a future point in time.

[0075] In an alternative embodiment, the loss prediction section (183) according to the invention is particularly characterized in that it additionally receives the time series data of at least 5 further measurement data and signals recorded during the journey by the detection sensor system (103) and the vehicle sensor system (141) of the ego vehicle (100), comprising the acceleration of the ego vehicle (100), which is implemented by the power transmission (130) (S07), the target drive torque of the drive unit (120) (S08), the actual wheel torque (S09), the time stamp for the detection sensor system (103) (S10) and the drive torque limitation (S11) as an input variable and uses this to predict a target object loss (ZOV) for a future point in time.

[0076] The loss prediction section (183) according to the invention comprises an RNN-LSTM model, comprising a dense layer as input layer, which comprises a number of units corresponding to the number of input variables, a first stretched layer, comprising an LSTM cell with 256 units, a second hidden layer, comprising an LSTM cell with 64 units and a dense layer as output layer, with 3 units, wherein the hidden layers are activated with a selu function and the output layer with a softmax function, wherein the initialization is carried out by means of Glorot normal distribution.

[0077] In one aspect of the invention, the cruise control device (180) according to the invention comprises a second movement prediction section (184) that determines a virtual trajectory (109) of the preceding target vehicle (101) as a function of the detected movement of the preceding target vehicle (101), the calculated trajectory (108) of the ego vehicle (100), and the predetermined target object loss (ZOV). Referring to Fig. 5 should be used in the case that at the prediction time t pa target object loss (ZOV) was predicted by the loss prediction section (183) and this was categorized as a non-trivial target object loss (nt-ZOV), now at least the speed profile and the acceleration profile of the preceding target vehicle (101) as a target object (ZO), as a virtual trajectory (109), so that the ACC function can control the ego vehicle (100) by means of the speed control section (185) in a following drive on the virtual trajectory (109) instead of switching to the free drive control. In this way, it can be advantageously prevented according to the invention that an intolerable jerk is caused by the corresponding ACC function in the event of sudden target object detection (ZOE). Here, the areas in the speed and acceleration profile which are determined by the exit speed v out and the exit acceleration a outof the target object (ZO) at exit time t out the target object loss (ZOV) and the entry velocity v in and the entry acceleration a in of the target object (ZO) at entry time t in Target object detection (ZOE), a period in which the ACC system operates virtually blindly and has no information about the target object (ZO). However, to predict the virtual trajectory of the target object (ZO), the following information is required at the prediction time t p already the information about the speeds v out and v in as well as the accelerations a out and a in needed.

[0078] According to one aspect of the invention, the second movement prediction section (184) calculates the acceleration of the preceding target vehicle (101) using the free road term of the so-called "Intelligent Driver Model", or also Intelligent Free Driver Model" (IFDM), according to equation 4a. Here, the acceleration a F of the preceding target vehicle (101) of an acceleration variable a, multiplied by the difference between 1 and the quotient of the detected speed v raised to the power of the free exponent δ F of the preceding target vehicle (101) and a reference speed v0, which corresponds, for example, to the exit speed v out corresponds. aF=a(1−(vFv0)δ)

[0079] Using the constants from equation 4b, the acceleration is reformulated by a differential equation from equation 4c. A=ain−aout(vinvout)δ−1B=ain−A∗(vvout)δ a(t)=v'(t)=A*(vvout)δ+B

[0080] The solution of the differential equation according to equation 4c calculates the speed v of the observed, preceding target vehicle (101) until the entry time t in the target object detection (ZOE), wherein the speed trajectory of the preceding target vehicle (101) is measured in a time interval from the exit time t out of the target object loss (ZOV) until the time of occurrence t in The time of occurrence t in However, the target object detection (ZOE) is not known in advance and is therefore determined according to equation 4d. tin(δ)=∫voutvin1A∗(Ivvout)δ+Bdv

[0081] In one aspect of the invention, the exit velocity v out and the exit acceleration a out of the target object (ZO) at exit time t outof the target object loss (ZOV) by extrapolation of recorded measurement data, by the detection section (181), from previous measurement points, before the prediction horizon of the prediction time t p Preferably, the last 10 measurement points before the prediction time t p as the basis for the extrapolation. In particular, the extrapolation is preferably carried out from the last 10 measurement points of the speed v and the acceleration a of the preceding target vehicle (101) before the prediction time t p using limited mathematical functions, such as those based on the hyperbolic tangent.

[0082] In one aspect of the invention, the inlet variables of the inlet velocity v in and the entry acceleration a in at the time of entry t inTarget object detection (ZOE) is determined using a transfer matrix. According to the invention, a transfer matrix is ​​a mathematical representation of the conditional and highly localized frequencies of reentry variables extracted from historical data, given their values ​​at target loss. In one aspect of the invention, the transfer matrix comprises frequency distributions for the kinematic variables of the exit velocity v out and the exit acceleration a out of the target object (ZO) at exit time t out of the target object loss (ZOV). For specified intervals from historical measurement and training data of the exit velocity v out and the exit acceleration a out of the target object (ZO) at exit time t out of the target object loss (ZOV) and assigned to these frequencies. For example, a relationship between recorded exit velocity v outand the corresponding frequency. The transfer matrix therefore includes a list of intervals for the exit velocity v out , preferably with an interval size of 3km / h and in a range from 0 to 180km / h, and for the exit acceleration a out , preferably with an interval size of 0.3 m / s 2 and in a range of -5 m / s 2 up to 5 m / s 2 and the corresponding frequencies, so that both the exit velocity v out and the exit acceleration a out Frequency distributions can be represented.

[0083] In one aspect of the invention, the production of the transfer matrix comprises the evaluation of historical measurement data, including the exit velocity v out , the exit acceleration a out , the entry velocity v in and the entry acceleration a inThe evaluation involves the following steps to create the transfer matrix: Each interval for the exit velocity v out , starting with the first interval [0; 3] km / h, where for each value v out an iteration loop is executed from the historical measurement data, comprising going through each interval for the exit acceleration a out , starting with the first interval [-5; -4.7]m / s 2 ; List of all entry values ​​for the entry speed v in and the entry acceleration a in from the historical data, within the specified intervals; interpolation of the entry values ​​for the entry velocity v in and the entry acceleration a in from adjacent intervals, in the absence of entry values ​​and repeating the steps until all intervals for the exit acceleration a outwere processed. Then repeat all steps for the further intervals of the exit velocity v out , until all intervals have been processed. The result of the transfer matrix now includes a tabular listing of the intervals of the exit velocity v out as an index in the first column, corresponding values ​​for the exit velocity v out , from the historical time series data in the second column, corresponding intervals for the exit acceleration a out in the third column, corresponding values ​​for the exit acceleration a out , from the historical time series data in the fourth column, corresponding intervals for the entry velocity v in in the fifth column, corresponding values ​​for the entry velocity v in, from the frequency distributions in the sixth column, corresponding intervals for the entry acceleration a in in the seventh column, corresponding values ​​for the entry acceleration a in , from the frequency distributions in the eighth column. In an alternative embodiment, the transfer matrix additionally includes a list of the corresponding distance to the target object (ZO) in a ninth column. In an alternative embodiment, the transfer matrix additionally includes a list of the corresponding entry time t in the recognition of the target object (ZOE) in a tenth column. This allows the predetermination of the entry velocity v in and the entry acceleration a in of the target object (ZO) at entry time t in the target object detection (ZOE) in one aspect of the invention, by means of a frequency distribution-based assignment to the exit velocity v outand exit acceleration a out of the preceding target vehicle (101), which are predetermined by extrapolating previous measurement data of the speed v and the acceleration a of the preceding target vehicle (101), which were recorded by the detection section, to the exit time t out which was predetermined by the loss prediction section (183).

[0084] If the exit velocity v out and the exit acceleration a out for the exit time t out at prediction time t p by interpolation of previous measurement data and the entry velocity v in and the entry acceleration a in for the entry time t in known from the transfer matrix, only the free exponent δ has to be estimated in order to solve the differential equation according to equation 4d in order to determine the entry time tin to be predetermined. In one aspect of the invention, the free exponent δ is estimated by calculating the integral according to equation 4d for intervals of δ in the range [-100; -1] and [1; 100].

[0085] If the entry time t in thus known, in one aspect of the invention the predetermination of the travelled distance s(t) is carried out as a virtual trajectory of the preceding target vehicle (101) according to equation 4e. s(t)=∫touttv(t')dt'

[0086] In one aspect of the invention, the production of the travel control device (180) according to the invention comprises the production of the second movement prediction section (184) by means of extraction of relevant signals, according to table T3, from the recorded measurement data and signals. (T3) Nr. Designation S24 Speed ​​of the ego vehicle at ZOV in km / h S25 Relative speed at ZOV in m / s S26 Speed ​​of the target object at ZOV in km / h S27 Acceleration of the ego vehicle at ZOV in m / s2 S28 Relative acceleration at ZOV in m / s2 S29 Acceleration of the target object at ZOV in m / s2 S30 Distance between ego vehicle and target vehicle at ZOV in m S31 Timestamp at ZOV in s S32 Speed ​​of the ego vehicle upon re-entry (ZOE) in km / h S33 Relative re-entry velocity (ZOE) in m / s S34 Speed ​​of the target object upon re-entry (ZOE) in km / h S35 Acceleration of the ego vehicle upon re-entry (ZOE) in m / s2 S36 relative acceleration on re-entry (ZOE) in m / s2 S37 Acceleration of the target object upon re-entry (ZOE) in m / s2 S38 Distance between ego vehicle and target vehicle in case of target loss in m S39 Timestamp on re-entry (ZOE) in s S40 Time difference entry tent - dead time in s

[0087] In some embodiments, the creation of the second motion prediction section (184) comprises one or more of the previously described processes, such as collecting measurement data, extracting relevant signals from the measurement data, and preprocessing the training data. In some embodiments, the measurement data and signals comprise the same ones as for creating the first motion prediction section (182) and the loss prediction section (183).

[0088] In summary, the second movement prediction section (184) according to the invention is characterized in that the determination of the virtual trajectory (109) of the preceding target vehicle (101), the predetermination of the entry speed v in , the entry acceleration a in and the time of entry t in the recognition (ZOE) of the preceding target vehicle (101) at the prediction time t pwhich is calculated by means of a frequency distribution-based transfer matrix, depending on the exit velocity v out and the exit acceleration a out of the preceding target vehicle (101) are estimated, wherein the exit speed v out and the exit acceleration a out by extrapolating previous measurement data for the speed v f and acceleration a F of the preceding target vehicle (101) recorded by the detection section (181) to the exit time t out , which is determined by the loss prediction section (183) at the prediction time t p was predetermined.

[0089] In one aspect of the invention, the travel control device (180) according to the invention comprises a speed control section (185) that controls the speed of the ego vehicle (100) depending on the predetermined target object loss (ZOV), its categorization, and the determined virtual trajectory (109) of the preceding vehicle (101). The speed control section (185) controls the travel of the ego vehicle (100), in particular the speed, taking into account information received from the loss prediction section (183) and the second movement prediction section (184).

[0090] In one aspect of the invention, the cruise control device (180) according to the invention carries out a method for controlling the speed of the ego vehicle (100). Fig. Figure 6 shows the process flow of such a procedure. Referring to Fig.6, the method for controlling the speed of the ego vehicle (100) is preferably carried out during the execution of an ACC function of the ego vehicle (100). Thus, the method according to the invention begins in a zeroth step (S100) with the activation of an ACC function. In a first step (S110), the movement of the ego vehicle (100) and one or more preceding vehicles (101) is detected by the detection sensor system (103) of the detection section (181).

[0091] For example, if no preceding vehicle (101) is detected, the cruise control section (185) receives this information from the detection section (181) and controls the host vehicle (100) to activate acceleration and accelerate to and maintain a preset target speed, in what is known as free-run control. In one embodiment, free-run control by the cruise control section (185) includes transmitting information to the drive control device (160) and the power transmission control device (170) to control the drive unit (120) and the power transmission (130).

[0092] If, for example, several vehicles (101) traveling ahead are detected, the detection section (181) identifies one of the vehicles (101) traveling ahead as the target object (ZO), taking into account the specified physical variables, such as the determined course curvature κ of the ego vehicle (100), the assignment of a driving corridor to the ego vehicle (100), or a predetermined target acceleration of the ego vehicle (100), and others. If, for example, exactly one vehicle (101) traveling ahead is detected, the detection section (181) identifies it as the target object (ZO), taking into account the specified physical variables, such as the determined course curvature κ of the ego vehicle (100), the assignment of a driving corridor to the ego vehicle (100), or a predetermined target acceleration of the ego vehicle (100), and others.If a target object is identified, the process flow of the method according to the invention jumps to a second step (S120), the determination of the trajectory of the ego vehicle (100).

[0093] The determination of the trajectory of the ego vehicle (100) after the second step (S120) is carried out according to the invention by the first movement prediction section (182) and comprises the determination of a predicted course curvature κ pred for a point in time in the future, where the predicted price curvature κ predis derived from the prediction of the route of the ego vehicle (100). For this purpose, the first movement prediction section (182) comprises an encoder-decoder-LSTM network, with an input layer for samples of dimension (100, 2), an encoder layer comprising an LSTM cell with 50 units, a RepeatVector comprising a length of 50 units, a decoder layer comprising 50 units, and an output layer comprising a dense layer with 50 units, wherein each layer is activated with the hyperbolic tangent function. In one embodiment, the determination of the trajectory of the ego vehicle (100) by means of the encoder-decoder-LSTM network comprises receiving two-dimensional x and y coordinates of the previous route of the last 100 meters of travel as input variables and predicting 50 meters of the future route of the ego vehicle (100).If the trajectory of the ego vehicle (100) is predetermined for a point in time in the future, the process flow of the method according to the invention jumps to a third step (130), the prediction of a target object loss (ZOV).

[0094] The prediction of the target object loss (ZOV) is carried out according to the invention by the loss prediction section (183) and comprises a query as to whether a target object loss (ZOV) has been predicted at all in a first, subordinate step (S131), the categorization of a potential target object loss (ZOV) in a second, subordinate step (S132) and the query as to whether the predetermined target object loss (ZOV) has been categorized as a non-trivial target object loss (nt-ZOV) in a third, subordinate step (S133).Wherein the prediction of a target object loss (ZOV) comprises receiving the time series data of at least six of the measurement data and signals recorded during the journey by the detection sensor system (103) and the vehicle sensor system (141) of the ego vehicle (100), comprising the ACC target acceleration (S01), the ACC target object detection (S02), the relative speed between the ego vehicle (100) and the target object (ZO) (S03), the distance to the target vehicle (S04), the speed of the ego vehicle (100) (S05) and the acceleration of the ego vehicle (100) (S06) as an input variable and predicting a target object loss (ZOV) for a future point in time therefrom.

[0095] For example, if a target object (ZO) has been identified by the detection section (181) in the first step (S110) and no target object loss (ZOV) is predicted by the loss prediction section (183) in the third step, the speed control section (181) receives this information and, in one embodiment, controls the ego vehicle (100) to deactivate acceleration and / or influence the movement of the ego vehicle (100) to bring about a preset target distance from the identified target object (ZO).

[0096] For example, when a target object loss (ZOV) is predicted by the loss prediction section (183) in the third step (S130), the process flow jumps to the second sub-step (S132) to categorize the predicted target object loss (ZOV).The categorization of the predetermined target object loss (ZOV) by the loss prediction section (183) comprises the application of a pre-trained RNN-LSTM model according to the invention, comprising a dense layer as input layer comprising a number of units corresponding to the number of input variables, a first stretched layer comprising an LSTM cell with 256 units, a second hidden layer comprising an LSTM cell with 64 units and a dense layer as output layer with 3 units, wherein the hidden layers are activated with a selu function and the output layer with a softmax function, wherein the initialization is carried out by means of a Glorot normal distribution.

[0097] For example, if the loss prediction section (183) categorized a predetermined target object loss (ZOV) in the second sub-step (S132) as a trivial target object loss (t-ZOV), i.e., not a non-trivial target object loss (nt-ZOV), the process flow jumps to the negative path in the third sub-step (S133), wherein the speed control section (185) receives this information and controls the ego vehicle (100) to a target speed.

[0098] For example, if a predetermined target object loss (ZOV) was categorized as a non-trivial target object loss (nt-ZOV) by the loss prediction section (183) in the second subordinate step (S132), the process flow jumps in the third subordinate step (S133) to the fourth step (S140), the prediction of the virtual trajectory (109) of the identified target object (ZO). The prediction of the virtual trajectory of the preceding vehicle (101) identified as the target object (ZO) is carried out in the fourth step (140) by the second movement prediction section (184), comprising the prediction of the entry speed v in , the entry acceleration a in and the time of entry t in the recognition (ZOE) of the preceding target vehicle (101) at the prediction time t p, which is calculated using a frequency distribution-based transfer matrix, depending on the exit velocity v out and the exit acceleration a out of the preceding target vehicle (101) are estimated, wherein the exit speed v out and the exit acceleration a out by extrapolating previous measurement data for the speed v and acceleration a of the preceding target vehicle (101), which were recorded by the detection section (181), to the exit time t out , which is determined by the loss prediction section (183) at the prediction time t pwas predetermined. If the virtual trajectory (109) of the preceding vehicle (101) identified as the target object (ZO) is predetermined, the process jumps to a fifth step (S150), the control of the ego vehicle (100) to a desired distance from the virtual trajectory (109) of the target object (ZO). If the virtual trajectory is not within the detection range of the ego vehicle's sensors, it is deformed by the second movement prediction section (184) using certain mathematical operations (e.g., virtual shortening of the ego-virtual ZO distance, virtual reduction of the road curvature, etc.) until it is within the detection range of the ego vehicle.

[0099] Thus, according to the invention, the speed control section is advantageously configured to transmit information to the drive control device (160) and / or to the power transmission device (170) in order to control the ego vehicle (100), to prevent acceleration and to adjust the speed of the ego vehicle (100) in such a way as to bring about a preset target distance to a preceding vehicle (101) identified as a target object (ZO), whenever no target object loss (ZOV) has been predetermined, or to bring about a preset target distance to a virtual trajectory (109) of the preceding vehicle (101) identified as a target object (ZO), whenever a target object loss (ZOV) has been predetermined and this has been categorized as a non-trivial target object loss (nt-ZOV), orto activate an acceleration and to regulate the speed of the ego vehicle (100) to a preset target speed whenever a target object loss (ZOV) has been predetermined and this has been categorized as a trivial target object loss (t-ZOV) or when no preceding target vehicle (101) has been identified as a target object (ZO).

[0100] Referring to the Fig. 7A and Fig. 7B, the method according to the invention will be explained in more detail using the device according to the invention using a preferred embodiment. Fig.Figure 7A shows a driving scenario comprising an ego vehicle (100) and a preceding target vehicle (101), which are traveling on a common roadway (102) within a common lane. The driving scenario depicted represents the case of a non-trivial target object loss (nt-ZO) of the preceding target vehicle (101), wherein the preceding target vehicle (101) is at an exit time t out is initially no longer detected by the detection sensors (103) of the ego vehicle (104), by a target object loss (ZOV) and at a later time t inis recognized by a target object recognition (ZOE). The ego vehicle (100) comprises one or more elements of a detection sensor system (103), which covers a conical, dashed detection area (104) of the vehicle surroundings of the ego vehicle (100). The course of the roadway (102) forms a right-hand bend, at the entrance to which the preceding target vehicle (101) is positioned. In the right-hand bend, there is also an obstacle (105), for example a building, which further restricts the view of the exit of the bend for the ego vehicle (100). The ego vehicle (100) and the preceding vehicle (101) are shown in solid lines at a first point in time and in dotted lines at a second point in time. The first point in time corresponds to the prediction time t p The second time point corresponds to the exit time t out . In Fig.7B shows a third time point. The third time point marks the entry time t in , i.e. the recognition of the target object (ZOE).

[0101] If, for example, the ACC function of the ego vehicle (100) is activated and the cruise control device (180) according to the invention executes the method according to the invention, the detection section (181) detects the preceding target vehicle (101) at the prediction time t with the means of the detection sensor system (103), for example by means of a radar sensor. p, according to the first step (S110) of the process flow of the method according to the invention. Furthermore, the detection section (181) identifies the detected preceding target vehicle (101) as a target object (ZO) by detecting the movement of the ego vehicle (100), by determining the course curvature κ, by assigning a driving corridor, by detecting the target acceleration of the ego vehicle (100), and by assigning the preceding target vehicle (101) to the driving corridor. The information about the detection of the preceding target vehicle (101) and its identification as a target object (ZO) is then transmitted to the first movement prediction device (182).

[0102] The first motion prediction device (182) then determines the current trajectory of the ego vehicle (100) using an encoder-decoder LSTM network according to the invention, according to a second step (S120) of the method according to the invention. Here, the first motion prediction device (182) determines a predicted course curvature κ pred as a virtual trajectory (106) of the ego vehicle (100) for a future point in time, taking into account a predicted route. The information about the virtual trajectory (108) of the ego vehicle (100) is transmitted to the loss prediction section (183).

[0103] The loss prediction section (183) receives the information about the virtual trajectory (108) of the ego vehicle (100) from the first movement prediction section (182) and about the movement of the preceding target vehicle (101) from the detection section (181) and uses this information to determine a target object loss (ZOV) in advance. In the illustrated embodiment, the target object loss (ZOV) occurs at the second time, the exit time t out, wherein the ego vehicle (100) is located at the entrance to the curve and the preceding target vehicle (101) is located at the exit of the curve, wherein the preceding target vehicle (101) is no longer detected by the detection range (104) of the detection sensor system (103) of the ego vehicle (100) due to the course of the roadway (102) and the restriction of visibility by the obstacle (105).The loss prediction section (183) receives the time series data of at least six of the measurement data and signals recorded during the journey by the detection sensors (103) and the vehicle sensors (141) of the ego vehicle (100), comprising the ACC target acceleration (S01), the ACC target object detection (S02), the relative speed between the ego vehicle (100) and the target object (ZO) (S03), the distance to the target vehicle (S04), the speed of the ego vehicle (100) (S05) and the acceleration of the ego vehicle (100) (S06) as an input variable and, in a third step (S130) of the method according to the invention, determines therefrom a target object loss (ZOV) for the exit time t using an RNN-LSTM model according to the invention. outbefore. The predetermined target object loss (ZOV) is further categorized into a non-trivial target object loss (nt-ZOV) according to a second, subordinate step (S132). The information about the predetermined, non-trivial target object loss (nt-ZOV) is subsequently transmitted to the second motion prediction section (184).

[0104] The second motion prediction section (184) receives the information about the non-trivial target object loss (nt-ZOV) from the loss prediction section (183) and about the movement of the ego vehicle (100) and the preceding target vehicle (101) from the detection section (181) and determines therefrom a virtual trajectory (109) of the preceding target vehicle (101) in advance, according to the fourth step (S140) of the method according to the invention. Here, the second motion prediction section (184) determines the virtual trajectory (109) of the preceding target vehicle (101) by predetermining the entry speed v in , the entry acceleration a in and the time of entry t in the recognition (ZOE) of the preceding target vehicle (101) at the prediction time t p , which is calculated using a frequency distribution-based transfer matrix, depending on the exit velocity v outand the exit acceleration a out of the preceding target vehicle (101) are estimated, wherein the exit speed v out and the exit acceleration a out by extrapolating previous measurement data for the speed v f and acceleration a F of the preceding target vehicle (101) recorded by the detection section (181) to the exit time t out , which is determined by the loss prediction section (183) at the prediction time t p was predetermined. The information about the deformed, virtual trajectory (109) of the preceding target vehicle (101) is transmitted to the cruise control section (185).

[0105] Thus, at prediction time t p the speed control section (181) is aware that at the exit time t out, wherein the preceding target vehicle (101) reaches the curve exit and the detection sensors (103) of the ego vehicle (100) will no longer detect it, a non-trivial target object loss (nt-ZOV) will occur, resulting from the information of the loss prediction section (183), and that at the time of occurrence t in , wherein the ego vehicle (100) will have passed the curve to such an extent that the detection sensor system (103) detects the preceding target vehicle (101) again and how the preceding target vehicle (101) moves in the period from the exit time t out until the time of occurrence t in, is estimated to move, based on the information from the second movement prediction section (184). From this information, the cruise control section (185) now implements instructions to the drive control device (160) and / or to the power transmission control device (170) of the ego vehicle (100) in order to prevent the acceleration of the ego vehicle (100) and to influence its speed in such a way that a target distance, preset by the ACC function, from the deformed, virtual trajectory (109) of the preceding target vehicle (101) is established, according to the fifth step (S150) of the method according to the invention.

[0106] In this way, according to the invention, a jolt is advantageously prevented which would occur if the ego vehicle (100), without the inventive cruise control device (180), were to switch to a free-run control instead of a distance-based speed control on a virtual trajectory (109) of the preceding target vehicle (101) and were to suddenly and unexpectedly re-detect the preceding target vehicle (101) at the third time, whereby an abrupt adjustment of the speed would be brought about by increased acceleration. List of reference symbols 100 Ego Vehicle 101 Target vehicle ahead 102 roadway 103 Detection sensors 104 Detection range 105 Obstacle 106 traffic signs 107 Recognition 108 Virtual Trajectory Ego-Vehicle 109 Virtual trajectory target vehicle 120 drive unit 130 Power transmission unit 140 chassis 141 Vehicle sensors 150 vehicle control device 160 Drive control device 170 Power transmission control device 180 cruise control device 181 Recording section 182 First motion prediction section 183 Second motion prediction section 184 Loss prediction section 185 Cruise control section S100 ACC function activated S110 Detect vehicle ahead S120 Determine ego vehicle trajectory S130 Predetermine ZOV S131 ZOV predetermined? S132 categorize ZOV S133 nt-ZOV categorized? S140 Predetermine virtual trajectory ZO S150 Adjust to target distance S160 Regulate to target speed

Claims

[1] A cruise control device (180) for an ego vehicle (100) which executes a distance-based cruise control function (ACC) to control the ego vehicle (100) in a following journey, depending on the travel of a preceding target vehicle (101), comprising a detection section (181) which detects the movement of the ego vehicle (100) and one or more preceding target vehicles (101) and identifies a target object (ZO) therefrom; a first movement prediction section (182) which, from the detected movement of the ego vehicle (100), at a prediction time t p and, depending on the cruise control function (ACC), a virtual trajectory (108) of the ego vehicle (100) is determined for a point in time in the future; a loss prediction section (183) which, at the prediction time t pa target object loss (ZOV) of the preceding target vehicle (101), depending on its detected movement and the virtual trajectory (180) of the ego vehicle (100), is predetermined for a future point in time and this predetermined target object loss (ZOV) is categorized into a trivial target object loss (t-ZOV) or a non-trivial target object loss (Nt-ZOV); a second movement prediction section (184), which at the prediction time t pa virtual trajectory (109) of the preceding vehicle (101) identified as the target object (ZO) is determined for a future point in time as a function of the detected movement of the preceding target vehicle (101) and the predetermined, categorized target object loss (ZOV); and a speed control section (185) that controls the speed of the ego vehicle (100) as a function of the predetermined, categorized target object loss (ZOV) and the determined deformed virtual trajectory (109) of the preceding vehicle (101) identified as the target object (ZO). [2] Device according to claim 1, characterized byin that the detection section (181) comprises technical components of a detection sensor system (103) and a vehicle sensor system (141), wherein the detection sensor system (103) comprises one or more components, such as a camera, a RADAR or a LIDAR sensor, and wherein the vehicle sensor system (141) comprises one or more of the components such as a wheel speed sensor, a steering wheel angle sensor, a yaw rate sensor and a lateral acceleration sensor. [3] Device according to claim 2, characterized by that the detection of the movement of the ego vehicle (100) allows the determination of the current course curvature κ, the vehicle acceleration x¨ and the curve speed v c,p and that the identification of a target object (ZO) comprises the assignment of a driving corridor to the ego vehicle (100), depending on one or more calculation variables, such as a corridor width b corr , a local hysteresis of the corridor width b corr, a target plausibility, a change in the target plausibility, a speed-dependent distance, a distance to the course center of the ego vehicle (100) and a target acceleration of the ego vehicle (100). [4] Device according to one of the preceding claims, characterized by that the first movement prediction section (182) has a predicted course curvature κ pred for a future point in time as the virtual trajectory (108) of the ego vehicle (100), where the predicted course curvature κ pred is derived from the prediction of the route of the ego vehicle (100). [5] Device according to claim 4, characterized byin that the first movement prediction section (182) comprises an encoder-decoder-LSTM network, with an input layer for samples of dimension (100, 2), an encoder layer comprising an LSTM cell with 50 units, a RepeatVector comprising a length of 50 units, a decoder layer comprising 50 units and an output layer comprising a dense layer with 50 units, each layer being activated with the hyperbolic tangent function and which is configured to receive two-dimensional x and y coordinates of the previous route of the last 100 meters of travel, detected by means of detection sensors (103) or position data of global or local coordinate systems, as input variables and to predict 50 meters of the future route of the ego vehicle (100) therefrom. [6] Device according to one of the preceding claims, characterized byin that the loss prediction section (183) receives the time series data of at least six of the measurement data and signals recorded during the journey by the detection sensors (103) and the vehicle sensors (141) of the ego vehicle (100), comprising the ACC target acceleration (S01), the ACC target object detection (S02), the relative speed between the ego vehicle (100) and the target object (ZO) (S03), the distance to the target vehicle (S04), the speed of the ego vehicle (100) (S05) and the acceleration of the ego vehicle (100) (S06) as input variables and uses these to predict a target object loss (ZOV) for a point in time in the future. [7] Device according to claim 6, characterized bythat the loss prediction section (183) additionally receives the time series data of at least 5 further measurement data and signals recorded during the journey by the detection sensors (103) and the vehicle sensors (141) of the ego vehicle (100), comprising the acceleration of the ego vehicle (100), which is implemented by the power transmission (130) (S07), the target drive torque of the drive unit (120) (S08), the actual wheel torque (S09), the time stamp for the detection sensors (103) (S10) and the drive torque limitation (S11) as input variables and uses this to predict a target object loss (ZOV) for a point in time in the future. [8] Device according to one of claims 6 or 7, characterized byin that the loss prediction section (183) comprises an RNN-LSTM model, comprising a dense layer as input layer, which comprises a number of units corresponding to the number of input variables, a first stretched layer, comprising an LSTM cell with 256 units, a second hidden layer, comprising an LSTM cell with 64 units and a dense layer as output layer, with 3 units, wherein the hidden layers are activated with a selu function and the output layer with a softmax function and wherein the initialization is carried out by means of Glorot normal distribution. [9] Device according to one of the preceding claims, characterized by that the determination of the virtual trajectory (109) of the preceding target vehicle (101) by the second movement prediction section (184), the predetermination of the entry speed v in , the entry acceleration a in and the time of entry t inthe recognition (ZOE) of the preceding target vehicle (101) at the prediction time t p which is calculated by means of a frequency distribution-based transfer matrix, depending on the exit velocity v out and the exit acceleration a out of the preceding target vehicle (101) are estimated, wherein the exit speed v out and the exit acceleration a out by extrapolating previous measurement data for the speed v f and acceleration a f of the preceding target vehicle (101) recorded by the detection section (181) to the exit time t out , which is determined by the loss prediction section (183) at the prediction time t p was predetermined. [10] Device according to one of the preceding claims, characterized bythat the speed control section (185) transmits information to the drive control device (160) and / or to the power transmission device (170) in order to control the ego vehicle (100) in order to prevent acceleration and to adjust the speed of the ego vehicle (100) in order to bring about a preset target distance to a preceding vehicle (101) identified as a target object (ZO), whenever no target object loss (ZOV) has been predetermined, or in order to bring about a preset target distance to a deformed, virtual trajectory (109) of the preceding vehicle (101) identified as a target object (ZO), whenever a target object loss (ZOV) has been predetermined and this has been categorized as a non-trivial target object loss (nt-ZOV), or in order to activate acceleration and adjust the speed of the ego vehicle (100) to a preset target speed to regulate, wheneverif a target object loss (ZOV) has been predetermined and categorized as a trivial target object loss (t-ZOV) or if no preceding target vehicle (101) has been identified as a target object (ZO). [11] Method for controlling the speed of an ego vehicle (100), by executing a speed control function (ACC) in order to control the ego vehicle (100) in a following journey, depending on the journey of a preceding target vehicle (101) (S100), wherein the movement of the ego vehicle (100) and one or more preceding target vehicles (101) is detected and a target object (ZO) is identified therefrom (S110), wherein from the detected movement of the ego vehicle (100), at a prediction time t p , depending on the cruise control function (ACC), a virtual trajectory (108) of the ego vehicle (100) is determined for a point in time in the future (S120), wherein at the prediction time tp a target object loss (ZOV) of the preceding target vehicle (101), depending on its detected movement and the virtual trajectory (108) of the ego vehicle (100), is predetermined (S130) for a future point in time, and that predetermined target object loss (ZOV) is categorized (S132) into a trivial target object loss (t-ZOV) or a non-trivial target object loss (Nt-ZOV), wherein at the prediction time t pa virtual trajectory (109) of the preceding target vehicle (101) identified as the target object (ZO) is determined for a future point in time as a function of the detected movement of the preceding target vehicle (101) and the predetermined, categorized target object loss (ZOV), and the speed of the ego vehicle (100) is controlled as a function of the predetermined, categorized target object loss (ZOV) and the determined virtual trajectory (109) of the preceding target vehicle (101) identified as the target object (ZO). [12] Method according to claim 11, characterized bythat the speed of the ego vehicle (100) is regulated to prevent acceleration and to adjust the speed of the ego vehicle (100) in such a way as to bring about a preset target distance to a preceding vehicle (101) identified as a target object (ZO), whenever no target object loss (ZOV) has been predetermined, or to bring about a preset target distance to a deformed, virtual trajectory (109) of the preceding vehicle (101) identified as a target object (ZO), whenever a target object loss (ZOV) has been predetermined and this has been categorized as a non-trivial target object loss (nt-ZOV), or to activate an acceleration and to regulate the speed of the ego vehicle (100) to a preset target speed, whenever a target object loss (ZOV) has been predetermined and this has been categorized as a trivial target object loss (t-ZOV) categorized or,if no preceding target vehicle (101) has been identified as a target object (ZO). [13] A method according to any one of claims 11 and 12, which is carried out by an apparatus according to any one of claims 1 to 10. [14] A vehicle comprising a cruise control device (180) according to any one of claims 1 to 10, which is arranged to carry out a method according to any one of claims 11 to 13.

Images