Technique for model parameter adaptation of a dynamic model for lateral and longitudinal guidance of a motor vehicle

Abstract

Device (100) for estimating and adapting at least one model parameter of a dynamic model for lateral guidance and / or longitudinal guidance of a motor vehicle, in particular a commercial vehicle, comprising: - at least one sensor (102) or at least one sensor interface for detecting at least one actual value of a movement state of the transverse guide and / or longitudinal guide; - an estimation unit (106) designed to estimate the at least one model parameter of a dynamic model by comparing at least one desired value of the motion state calculated from the dynamic model of the lateral guidance and / or longitudinal guidance of the motor vehicle with the detected at least one actual value; and - a control unit (108) which is designed to adapt the dynamic model of the lateral guidance and / or longitudinal guidance depending on the detected at least one actual value of the movement state and the estimated at least one model parameter and to control control signals of at least one actuator of the lateral guidance and / or longitudinal guidance depending on the adapted dynamic model, wherein the dynamic model comprises a kinematic single-track model and the motion state comprises a steering angle of a wheel, a steering wheel angle, an amount of speed and / or an amount of acceleration of the motor vehicle, and wherein the estimation unit (106) and / or the control unit (108) is further designed to calculate in advance a time series of N future target values ​​of the motion state, in particular target values ​​of the steering angle, steering wheel angle, speed and / or acceleration, by means of the dynamic model, wherein N is a natural number and greater than or equal to 2, and wherein the estimation unit (106) varies and estimates the at least one model parameter in the dynamic model based on a minimal deviation of several of the target values ​​from the detected actual value of the motion state.

Description

[0001] The invention relates to a device and a method for estimating and adapting at least one model parameter of a dynamic model for lateral and / or longitudinal guidance of a motor vehicle, in particular a commercial vehicle.

[0002] First, reference should be made to documents DE 10 2006 054 425 A1, DE 10 2010 050 278 A1, DE 10 2011 121 454 A1, DE 10 2008 030 667 A1, DE 10 2008 040 240 A1 and DE 10 2018 217 845 A1.

[0003] Document DE 10 2006 054 425 A1 proposes a method for determining a value of a model parameter of a vehicle reference model, with which a reference value of a first driving state variable can be determined. The method is characterized in that an estimated value of the model parameter is determined using an artificial neural network as a function of at least one second driving state variable and / or a variable specified by a driver.

[0004] Document DE 10 2010 050 278 A1 relates to a method for estimating a sideslip angle. The sideslip angle that develops during a motor vehicle's travel is determined from a linear single-track model. The parameters included in the model are continuously updated based on other variables, such as the yaw rate, which can be calculated using the linear single-track model and measured simultaneously. A Kalman filter performs a comparison such that a parameter, such as the slip stiffness of the vehicle's wheels, is adjusted so that the calculated values ​​for the other variable match the measured values ​​as closely as possible.

[0005] Document DE 10 2011 121 454 A1 describes a control device for a motor vehicle. To fully exploit the potential of a motor vehicle's chassis control system for specific driving situations, a central control instance can be provided that controls all existing actuators.

[0006] Document DE 10 2008 030 667 A1 relates to a method for estimating parameters for characterizing vehicle characteristics in a motor vehicle. At least two different driving situations are specified, with each driving situation being assigned an estimation device for estimation. Furthermore, it is determined that a first specified driving situation exists, whereupon the estimation device assigned to the first driving situation is activated. Based on the activation, the estimation device determines an estimated value for the associated parameter using an estimation method.

[0007] Document DE 10 2008 040 240 A1 discloses a vehicle with an electronic control system and a method for actively correcting driving dynamics characteristics as a function of driving dynamics parameters, comprising means for detecting and / or determining driving dynamics parameters. It is provided that the means for detecting driving dynamics parameters comprises at least three position-determining devices arranged at a distance from one another, and the position changes at at least three position-determining devices arranged at a distance from one another are detected, in particular simultaneously.

[0008] The document DE 10 2018 217 845 A1 describes a method for controlling a vehicle, comprising: predicting a first parameter of a vehicle state at each of a plurality of points in time as a function of a first parameter of a current vehicle state and a first model associated with the vehicle, predicting a second parameter of the vehicle state at each of the plurality of points in time as a function of a second parameter of the current vehicle state, the plurality of predictions of the first parameter of the vehicle state and a second model associated with the vehicle, and determining one or more input variables for the vehicle at each of the points in time as a function of the predictions of the first and second parameters of the vehicle trajectory at each of the plurality of points in time and desired first and second parameters of the vehicle trajectory at each of the plurality of points in time.

[0009] Driver assistance systems (FAS, also known as Advanced Driver Assistance Systems or ADAS) in heavy commercial vehicles often have automated longitudinal and / or lateral guidance. Suitable sensors are used to determine the respective target variable, which can also be referred to as a reference variable, for example in the form of a desired speed or a desired distance from a vehicle in front, or a desired relative deviation from the center of the road. The control strategies used calculate the control components for the vehicle's actuators, e.g. brakes, engine and steering, required to adjust the target variable at a given time, depending on the deviation between the determined target variable and the measured actual variable. The control variable often also consists of another controlled variable, such as a target acceleration, for a subordinate control system to control the actuators.

[0010] Conventional systems, such as cruise control or adaptive cruise control, are convenience systems that support the driver in their driving task but do not relieve them of their responsibility. Therefore, with cruise control, for example, the main focus is on precisely maintaining a steady-state setpoint speed. The point in time at which the setpoint speed is reached is of secondary importance in this application. In predictive cruise control systems, instead of a statically specified speed profile, an optimal setpoint speed profile based on the road topography is calculated using the location coordinates of a section of road ahead to increase efficiency. A fuel consumption advantage can only be achieved if the setpoint speed is maintained at the specified location.However, the accuracy requirements are comparatively low due to the low rates of change in the topography.

[0011] With the development of supportive assistance functions toward highly automated driving, input variables for longitudinal and lateral guidance are traditionally determined by so-called planning algorithms. These algorithms calculate a trajectory optimized for suitable quality criteria, taking into account the destination and the environment model calculated from the vehicle's onboard sensors, as a reference for vehicle control. This can be understood as a set of spatially or temporally discretized points in a state space ahead, which can contain, for example, the target acceleration, target speed, and position coordinates in the longitudinal, transverse, and lateral directions associated with the respective location or time, as well as the lane curvature.

[0012] Since, in comparison to comfort-oriented driving functions, responsibility for the driving task is transferred to the ADAS in highly automated driving, this creates significantly more restrictive requirements for the spatial and temporal control behavior of the control system and the control accuracy, particularly in heavy commercial vehicles and in spatially confined driving situations. Depending on the situation, it may be necessary to maintain the target position and target orientation of the vehicle along the target trajectory to within a few centimeters or significantly less than one degree (1 °), as is the case, for example, when maneuvering a truck-trailer within a container terminal. Reaching the trajectory waypoints at the most precise time possible is of crucial importance for safety reasons, particularly when automatically performing lane changes or evasive maneuvers in the presence of other road users.

[0013] Given the significantly stricter accuracy requirements, inaccuracies in, for example, vehicle-specific model parameters and / or simplifications of dynamic models that exclude difficult-to-determine—particularly vehicle-specific—model parameters are significantly more significant in ADAS compared to comfort systems. Model parameters, such as vehicle-specific model parameters influenced by the slip stiffness of the front and rear axles, are complex to model due to a multitude of technical influences and are also time- and condition-dependent.

[0014] If the model parameters (e.g., vehicle-specific ones) are not taken into account during planning and control, a discrepancy between the desired and actual trajectory progressions will always occur. This can lead to undesirable driving and planning behavior even in relatively undesirable situations, such as frequent resetting of the trajectory planning or instability within the lateral guidance.

[0015] In addition to model parameters such as vehicle mass and wheelbase, inertia and dead times within the actuator system are relevant variables in a dynamic model for lateral and / or longitudinal guidance. For example, document EP 3 373 095 A1 describes a pure-pursuit control system in a vehicle group (i.e., a "platoon") with variable forward motion and actuator inertia compensation using a "lag compensator" (also called a "lead-leg compensator"), which increases the system's phase margin.

[0016] The achievable precision of vehicle control depends, particularly in the area of ​​lateral dynamics, on the quality of the dynamic model embedded in the planning algorithm. The relevant variables, such as tire parameters or moments of inertia, often vary over time and, in addition, act with greater or lesser systematic accuracy. They are unknown or only very roughly known, and can only be determined with considerable effort.

[0017] The object is therefore to provide a device and a method for estimating and adapting a model parameter of a dynamic model for the lateral and / or longitudinal guidance of a motor vehicle, in particular a commercial vehicle, which improves the accuracy of the lateral and / or longitudinal guidance. Alternatively or additionally, the object is to estimate nonlinearities in the dynamic models of the lateral and / or longitudinal guidance and / or to consider the model parameters by estimating nonlinearities in the dynamic models of the lateral and / or longitudinal guidance and / or to optimize the planning algorithm for the lateral and / or longitudinal guidance.

[0018] This object is achieved by the device and method, as well as a corresponding motor vehicle, having the features of the independent claims. Advantageous embodiments and applications of the invention are the subject of the dependent claims and are explained in more detail in the following description, with partial reference to the figures.

[0019] According to one aspect of the invention, a device is provided for estimating and adapting at least one model parameter of a dynamic model for the lateral and / or longitudinal guidance of a motor vehicle, in particular a commercial vehicle. The device comprises at least one sensor or at least one sensor interface for detecting at least one actual value of a motion state of the lateral and / or longitudinal guidance. Furthermore, the device comprises an estimation unit configured to estimate the at least one model parameter of a dynamic model by comparing at least one target value of the motion state calculated from the dynamic model of the lateral and / or longitudinal guidance of the motor vehicle with the detected at least one actual value.Furthermore, the device comprises a control unit which is designed to adapt the dynamic model of the lateral guidance and / or the longitudinal guidance depending on the detected at least one actual value of the movement state and the estimated at least one model parameter and to regulate control signals of at least one actuator of the lateral guidance and / or the longitudinal guidance depending on the adapted dynamic model.

[0020] The at least one sensor or the at least one sensor interface for detecting the actual value of the motion state of the motor vehicle can comprise a steering angle sensor, a steering wheel angle sensor, a tachometer, a radar sensor (for example on a front of the motor vehicle), a camera (for example in the optical or infrared spectrum) and / or a lidar sensor (i.e. a sensor for direction-resolved distance measurement, preferably using light, or using "light detection and ranging", LIDAR) or can be connected to such a sensor for data exchange. Alternatively or additionally, motion states of a vehicle traveling ahead can be received from the vehicle traveling ahead via a radio connection. For example, the distance or the relative speed to the vehicle traveling ahead can be detected by the sensor system of the motor vehicle and the absolute speed of the vehicle traveling ahead can be received via the radio connection.Alternatively or additionally, the sensor interface can receive data or signals for position determination from a navigation satellite system, for example from a “Global Positioning System” (GPS).

[0021] The motion state can be a controlled variable of the control unit. In particular, the output control signals can cause a change in the motion state by means of the at least one actuator depending on the detected at least one actual value of the motion state.

[0022] The movement condition may include driving at high speed with few curves (e.g., driving on a motorway), driving with tight curves and at different speeds, and / or maneuvering.

[0023] The dynamic model can be a quantitative (particularly numerical) description of the temporal development of the motion state of the lateral and / or longitudinal guidance of the motor vehicle. The motion state can be a point in a state space of the dynamic model of the motor vehicle.

[0024] The control unit can be configured to control the at least one actuator using the dynamic model, wherein, for example, a virtual wheelbase, a vehicle mass, a rolling resistance, and / or a moment of inertia is at least one model parameter of the dynamic model and / or the motion state is a dynamic quantity (i.e., a variable) of the dynamic model. Further model parameters of the dynamic model can include inertia and dead times of the actuators.

[0025] The dynamic model can also be referred to as a dynamic model or motion model of the motor vehicle. The actuator can also be referred to as an actuator. The target value can also be referred to as a target specification or required value. The actual value can also be referred to as an output variable or measured value.

[0026] The target value and the actual value can refer to a corresponding time t1. The target value can be determined by evaluating the dynamic model at time t1, whereby the dynamic model is initialized with an actual value of the motion state recorded at time t0. Time t0 can be prior to time t1.

[0027] The dynamic model can describe a transfer behavior between the target value and the actual value using a transfer function, for example in the frequency domain and / or by combining a first-order or higher-order delay element and a dead-time element. The delay element, for example a PT1 element, of the transfer function can describe a proportional transfer behavior with a first-order delay and with an actuator gain factor K. The dead-time element can describe a time period between a signal change at the system input and the signal response at the system output. The transfer function of the dynamic model can comprise a product of the transfer functions of the delay element, for example an inverse linear function of the time constant of the delay element, and the dead-time element, in particular an exponential dependence on the dead time.The dynamic model can include different transfer functions for acceleration and braking processes.

[0028] At least in one exemplary embodiment, the at least one model parameter in the dynamic model and thus in the control can be adapted and / or taken into account by means of the at least one sensor and the estimation unit, for example without installing additional sensors for the dedicated measurement of non-linear effects of the dynamic model. In the same or a further exemplary embodiment, the motor vehicle itself (for example by means of an internal vehicle control unit) can estimate and / or adapt the at least one model parameter, for example without external measuring technology during an inspection and / or during use (in particular while driving or when it is parked). In the same or a further exemplary embodiment, the output of the control signals to at least one actuator of the longitudinal guidance and / or lateral guidance can depend on the at least one adapted model parameter.

[0029] The device may further comprise a storage unit configured to store the detected at least one actual value. The at least one sensor or the at least one sensor interface may output the detected at least one actual value to the storage unit. The estimation unit may be configured to read the detected at least one actual value from the storage unit or to receive it from the storage unit. The estimation unit may comprise an in-vehicle computing unit. Alternatively or additionally, the estimation unit may comprise an external computing unit, for example, a central computing unit of a vehicle fleet operator.

[0030] According to a second aspect, a method is provided for estimating and adapting at least one model parameter of a dynamic model for the lateral and / or longitudinal guidance of a motor vehicle, in particular a commercial vehicle. The method comprises the step of detecting at least one actual value of a motion state of the lateral guidance and / or the longitudinal guidance using at least one sensor or at least one sensor interface. Furthermore, the method comprises the step of estimating at least one model parameter of a dynamic model by comparing at least one target value of the motion state calculated from the dynamic model with the detected at least one actual value.The method further comprises the step of controlling at least one actuator of the lateral guidance and / or the longitudinal guidance depending on the detected actual state, wherein a control unit of the lateral guidance and / or the longitudinal guidance outputs control signals to the actuator depending on the dynamic model adapted by the estimated at least one model parameter.

[0031] Optionally, the method comprises a step of storing the recorded at least one actual value.

[0032] The method may further comprise each individual feature described in the context of the device and / or a corresponding method step, and vice versa. The following features relate to both the device of the first aspect and the method of the second aspect. Although device features of the first aspect are described below, these also apply as corresponding method steps of the second aspect.

[0033] The at least one sensor or the at least one sensor interface can comprise a receiving module for receiving signals from a global navigation satellite system, preferably including differential correction signals. The global navigation satellite system can comprise the so-called Global Positioning System (GPS). The differential correction signals can correspond to Differential GPS (DGPS).

[0034] The dynamic model can include slip stiffnesses on the front and / or rear axle, at least one center of gravity distance between the front and rear axles, a distance between the front and rear axles, a mass of the motor vehicle, tire parameters, a rolling resistance, and / or a moment of inertia (for example, about a vertical axis) of the motor vehicle as model parameters. Alternatively or additionally, the dynamic model can include inertias of actuators and / or dead times of actuators as model parameters. The at least one center of gravity distance between the front and rear axles can include a distance between the centers of gravity of the respective axles. The distance between the centers of gravity of the front and rear axles can also be referred to as the actual wheelbase. Alternatively or additionally, the center of gravity distances of the front and rear axles can each be measured relative to the center of gravity of the motor vehicle.

[0035] The dynamic model includes a kinematic single-track model. The motion state includes a wheel steering angle, a steering wheel angle, a speed, and / or acceleration of the vehicle. For example, in the case of rear-axle steering of the vehicle, the steering angle can be the steering angle of the rear wheel.

[0036] The kinematic single-track model can comprise coupled differential equations (e.g., first-order linear differential equations) for a position coordinate of the longitudinal guidance of the motor vehicle (also referred to as the longitudinal position coordinate), a position coordinate of the lateral guidance of the motor vehicle (also referred to as the transverse position coordinate), and / or a yaw angle of the motor vehicle. The yaw rate, i.e., the change in the yaw angle, can be a function of the steering angle and, optionally, the speed (e.g., the magnitude of the speed) of the motor vehicle. A wheelbase of the motor vehicle can be a (e.g., free) parameter of the dynamic model. Furthermore, the yaw angle can depend on a parameter of the dynamic model, for example, the wheelbase of the motor vehicle.

[0037] The dynamic model with slip stiffnesses can include a coupled (e.g. linear) first-order differential equation system for the change of the position coordinate in the transverse direction of the vehicle and the yaw rate.

[0038] The estimation unit and / or the control unit are further configured to predict a temporal series of N future target values ​​of the motion state, in particular target values ​​of the steering angle, steering wheel angle, speed, and / or acceleration, using the dynamic model, where N is a natural number. N is equal to or greater than 2, optionally equal to or greater than 3.

[0039] The estimation unit varies the at least one model parameter in the dynamic model (e.g., for estimation). Additionally, the estimation unit is configured to estimate the at least one model parameter based on a minimal deviation of several of the target values ​​from the detected actual value of the motion state.

[0040] Predictive calculation using the dynamic model can also be referred to as model-predictive calculation or model-predictive calculation. The at least one model parameter of the dynamic model of the lateral guidance can be estimated by calculating the difference between a model-predictively calculated target steering wheel angle (or target steering angle)—for example, calculated using varied values ​​of the model parameter—and the associated, in particular, recorded actual steering wheel angle (or actual steering angle), and determining the minimum of all differences—for example, for varied values ​​of the model parameter. Alternatively or additionally, a series of model-predictively calculated target steering wheel angles (or target steering angles) can be compared with a single recorded actual steering wheel angle (or actual steering angle).In a further alternative, a series of recorded actual steering wheel angles (or actual steering angles) can be compared with a single target steering wheel angle (or target steering angle). The at least one model parameter of the dynamic model of the longitudinal guidance can be determined by calculating the difference between a model-predictively calculated target acceleration—for example, calculated using varied values ​​of the model parameter—and the associated recorded actual acceleration, in particular assigned to the same point in time, and determining the minimum of all differences—for example, for varied values ​​of the model parameter. Alternatively or additionally, a series of model-predictively calculated target accelerations can be compared with a single recorded actual acceleration. In a further alternative, a series of recorded actual accelerations can be compared with a single target acceleration.

[0041] Each (for example, discrete) index i=1..N of the series can advantageously be assigned a time t i be assigned or correspond to. The series of future target values ​​indexed with i=1..N at times t i can be calculated based on an actual value measured at time t0 as the initial value. The time points can be ordered in ascending order, ie t0 <t1<..<t N . The comparison of the target value(s) with the values ​​at times t i measured actual values ​​can be directly after the time t i Alternatively or additionally, the target value(s) and actual value(s) can be calculated after an end time, for example t N , can be compared. From the index i of the minimum of the difference between the target value (e.g., target value of the steering wheel angle) and the actual value (e.g., actual value of the steering wheel angle), at least one model parameter of the dynamic model of the longitudinal guidance and / or the lateral guidance can be estimated.

[0042] The estimation unit can be configured to estimate a value of a virtual wheelbase of the kinematic single-track model as a model parameter, wherein the virtual wheelbase takes into account or represents the at least one center of gravity distance of the front axle and / or rear axle and / or the distance between the front axle and rear axle and / or (for example, non-linear) effects, in particular slip stiffnesses. The center of gravity distance between the front axle and rear axle of the motor vehicle can also be referred to as the actual wheelbase or correspond thereto. The virtual wheelbase can be different from the actual wheelbase, preferably in order to represent (for example, non-linear) effects, in particular slip stiffnesses.The virtual wheelbase can be determined by minimizing a lateral position error depending on the (e.g. current) lateral acceleration by varying the virtual wheelbase in the dynamic model of the lateral guidance.

[0043] The fit of at least one model parameter can be estimated using an artificial neural network. The artificial neural network can comprise a so-called "General Regression Neural Network" (GRNN).

[0044] The input space of the artificial neural network may include a lateral acceleration, a steering angle, a steering wheel angle, a yaw rate and / or a curvature of the trajectory of the motor vehicle.

[0045] A transfer function of the dynamic model of the lateral guidance and / or the longitudinal guidance can comprise a PT1 element (for example, as a factor, in particular in the frequency domain). The estimation unit can further be configured to determine the PT1 element of the transfer function by determining at least one parameter of the PT1 element and / or from a list of PT1 elements by comparing the at least one target value calculated from the dynamic model with the at least one detected actual value. A list of the dynamic models can comprise the list of PT1 elements. The transfer function in the frequency domain can comprise a product of an exponential term describing the dead time of the actuator and polynomial terms in the numerator and denominator of a delay element. A PT1 element can correspond to a linear polynomial in the denominator of a delay element.

[0046] The PT1 element can be a linear (i.e., "P" for "proportional") first-order (i.e., "1" for first order) transfer element in time (i.e., "time" for "T"). The PT1 element can model a relaxation time of the lateral and / or longitudinal guidance. The list of PT1 elements can correspond to different relaxation times.

[0047] The transfer function can model the lateral and / or longitudinal guidance provided by the actuator. The transfer function can represent a causal relationship between the lateral and / or longitudinal guidance provided by the actuator. In a frequency domain representation, the transfer function can be a quotient of the motion state (as an output signal of the lateral and / or longitudinal guidance provided by the actuator) and the control signal output to the actuator (as an input signal of the lateral and / or longitudinal guidance provided by the actuator).

[0048] A transfer function of the dynamic model of the lateral guidance and / or the longitudinal guidance can comprise a PD1 element (for example, as a factor, in particular in the frequency domain). The estimation unit and / or the control unit can further be configured to determine and / or adjust the inertia of the actuator. The inertia of the actuator can be adjusted in the frequency domain by a polynomial in the numerator of the delay element of the transfer function. The PD1 element can comprise a linear (i.e., "P" for "proportional") controller with a first-order differentiating (i.e., "D" for "differential") component.

[0049] The lateral and / or longitudinal guidance can comprise a follow-up control for a vehicle in a vehicle convoy. A vehicle convoy can also be referred to as a platoon. The follow-up control can comprise so-called pure-pursuit control with variable look-ahead and actuator inertia compensation. The variable look-ahead can comprise a (for example, linear) dependence of a target steering angle of a following vehicle on an actual steering angle of the vehicle ahead in the vehicle convoy (preferably offset by a time constant). The actuator inertia compensation can also be referred to as lag compensation or lead-lag compensation. The lag compensation can comprise dominant pole compensation, which is incorporated into the actuator's transfer function as an additional transfer element. The time constant of the lag compensation element can be different from the time constant of the other transfer function elements.In particular, the time constant of the lag compensation element can be smaller than the time constant of a PT1 element or a higher-order polynomial element.

[0050] The estimation unit and / or the control unit can be configured to estimate and / or adapt the at least one model parameter of the dynamic model during driving operation. Alternatively or additionally, the estimation unit and / or the control unit can comprise at least one interface for changing the at least one model parameter of the dynamic model during driving operation. Here, the time specification "during driving operation" can also be implemented as "at runtime."

[0051] An estimate of at least one model parameter of the dynamic model can be adjusted depending on a detected driving state. The driving state can include a driving maneuver. The driving state can include driving at a constant speed (e.g., driving on a highway at high speed with hardly any curves), driving around tight curves at different speeds, maneuvering, acceleration, braking, and / or a lane change.

[0052] An initialization (e.g., an estimated value) of at least one model parameter of the dynamic model can be performed with measured or empirically determined initial values ​​and / or initial values ​​determined outside of driving operation or by offline fitting and / or be limited to a physically meaningful value range. For example, the estimate of the virtual wheelbase can be initialized using the value of the actual wheelbase. "Offline" refers to an estimate or calculation outside of ongoing driving operation. In particular, an estimate or calculation can be performed using an artificial neural network, such as a "General Regression Neural Network" (GRNN). An estimate or calculation during ongoing driving operation is also referred to as an "online" estimation or "online" calculation.Alternatively or additionally, an online estimation or calculation can involve a wireless connection between the motor vehicle and an external estimation or calculation unit, such as a fleet operator's computer system. The online estimation or calculation can be performed internally or externally within the vehicle.

[0053] According to a further aspect, a motor vehicle, in particular a commercial vehicle, is provided, which comprises a device according to the above aspect for estimating and adapting at least one model parameter of a dynamic model for lateral and / or longitudinal guidance. Alternatively or additionally, a motor vehicle is provided which is configured to carry out the method according to the above second aspect for estimating and adapting at least one model parameter.

[0054] In every aspect, the commercial vehicle can be a truck, a tractor, a bus or a crane truck.

[0055] The features described above can be implemented in any combination. Further features and advantages of the invention are described below with reference to the accompanying drawings. They show: Fig. 1 an embodiment of a device for estimating and adapting at least one model parameter of a dynamic model; Fig. 2 an embodiment of a method which is carried out by means of the device according to Fig. 1 is executable; Fig. 3 a schematic artificial neural network; and Fig. 4 an example of an estimation and adjustment of at least one model parameter in model predictive control.

[0056] Fig. 1 shows an embodiment of a device, generally designated by reference numeral 100, for estimating and adapting at least one model parameter of a dynamic model for the lateral and / or longitudinal guidance of a motor vehicle, using the example of a commercial vehicle. The device 100 comprises a sensor 102 or a sensor interface designed to detect actual value(s) of the motion state. Optionally, the device 100 comprises a storage unit 104 designed to store actual value(s). Furthermore, the device 100 comprises an estimation unit 106 designed to estimate at least one model parameter of a dynamic model for the lateral and / or longitudinal guidance based on a comparison of determined target values ​​and measured actual values.The control unit 108 of the device 100 is designed to control an actuator of the lateral guidance and / or longitudinal guidance using control signals depending on the dynamic model adapted by the estimated at least one model parameter. All units 104, 106, 108 and sensors 102 or sensor interfaces can be internal to the vehicle. Alternatively or additionally, the storage unit 104 and / or the estimation unit 106 can be external to the vehicle. For example, the one or more recorded actual values ​​of the motion state can be sent via a radio connection during driving operation to the operator of the vehicle fleet to which the commercial vehicle belongs. A central, in particular external to the vehicle, estimation unit 106 can determine the at least one model parameter based on the received actual values ​​and send them to the commercial vehicle via the radio connection or another radio connection during driving operation.Alternatively or additionally, the one or more actual values ​​can be stored on an in-vehicle storage unit 104 during driving operation. The data from the storage unit 104 can be read out by an external estimation unit 106 outside of driving operation, for example, during maintenance or when parked in the haulage company's fleet. The external estimation unit 106 can determine the at least one model parameter based on the read data, in particular past actual values ​​of the motion state, and transmit it to the in-vehicle control unit 108 of the commercial vehicle.

[0057] Fig. 2 shows an exemplary embodiment of a method, generally designated by reference numeral 200, for estimating and adapting at least one model parameter of a dynamic model of the lateral guidance and / or the longitudinal guidance of a motor vehicle, in particular a commercial vehicle. In a first step 202 of the method 200, one or more actual values ​​of the motion state of the lateral guidance and / or the longitudinal guidance of the vehicle are recorded by means of a sensor or a sensor interface. Step 202 can be performed by the at least one sensor 102 or the at least one sensor interface of the device 100. In an optional step 204, the actual value or values ​​can be stored. Step 204 can be executed by the memory unit 104 of the device 100.In a further step 206, the at least one model parameter is determined by comparing the actual value(s) with the target value(s) calculated from a dynamic model. Step 206 can be performed by the estimation unit 106 of the device 100. In step 208, an actuator is controlled by the lateral guide and / or the longitudinal guide sending control signals to the actuator depending on the dynamic model adapted by the estimated at least one model parameter. Step 208 can be performed by the control unit 108 of the device 100.

[0058] In conventional methods for at least partially automated vehicle operation, a motion planning algorithm based on an environment model generated using, among other things, environmental sensors and high-precision digital map data is used to determine a future target driving state suitable for the given driving task. This target driving state is then passed on to a vehicle longitudinal and lateral guidance system (hereinafter also referred to as follow-up control) as a target value. The target driving state x determined by the motion planning soll,t can determine the target values ​​belonging to a fixed time t or location, in particular in the longitudinal and transverse direction of the vehicle, for location coordinates (x, y), speeds (v x , v y ), acceleration (a x , a y ) along different coordinate axes and / or yaw angle ψ and yaw rate (time change of yaw angle ψ): xset,t=[t, x, y, vx, vy, ax, ay, ψ˙, ...]T.

[0059] A set of N spatially or temporally discretized target driving states, mostly within a certain temporal and / or spatial forecast horizon, is called the target trajectory X soll designated: Xsoll={xsoll,t0, xsoll,t1, …, xsoll,tN}.

[0060] Here, the target driving conditions are ordered in time from an initial time t0 (t0 <t1<..<t N). Due to limited computing power, conventional motion planning methods use highly simplified motion models to calculate various sets of vehicle states within a given look-ahead horizon that lead to the selected destination. In an optimization step, the optimal target trajectory is identified with regard to suitable quality criteria, taking into account vehicle-specific boundary conditions such as the wheelbase and / or the mass of the vehicle and / or a load state and / or an acceleration capability, and is then passed on to the subsequent control.

[0061] Here, actuators, in particular vehicle actuators, can refer to a majority of the actuators of the motor vehicle.

[0062] The task of the follow-up control is then to calculate suitable specifications for the vehicle actuators (engine, transmission, brakes, steering) in order to determine the actual state x ist,tof the vehicle to the target state along the trajectory X soll Some conventional control strategies take a reference state x from the (especially target) trajectory soll,tk , where the time look ahead can be selected depending on driving speed and control dynamics k is a fixed index of a point on a target trajectory X soll . Other conventional control strategies use several (especially target) states or the entire (especially target) trajectory in model predictive procedures in order to implement a driving profile that is optimal in terms of control effort and following behavior.

[0063] Both methods therefore use implicit or explicit knowledge about the dynamics of the automated vehicle.

[0064] If the dynamic models conventionally used for motion planning inadequately represent the vehicle, the probability that the vehicle guided by the follow-up control system will not be able to follow the (target) trajectory with the required accuracy increases. The probability that the vehicle guided by the follow-up control system will no longer be able to follow the (target) trajectory with the required accuracy also increases if the vehicle dynamics modeled explicitly or implicitly in the control system deviate significantly from the actual vehicle behavior. The control accuracy with which the automated vehicle follows the (target) trajectory specified by the planning system depends on how precisely the vehicle dynamics models used in the planning and control system describe the actual vehicle behavior.

[0065] Due to the limited available computing power, especially within the vehicle itself, increasing the accuracy of dynamic models by significantly increasing model depth and / or model order is only of limited use. Many parameters that influence dynamics, such as tire slip stiffness or straight-line stability, are difficult to measure and / or vary over time. Therefore, the approach of measuring as many physical parameters as possible and incorporating them (e.g., explicitly) into the model is also limited in its results, especially with continuous adjustment during driving.

[0066] Embodiments of the device enable the determination of dominant dynamic components (e.g., nonlinear) that are not represented by simple motion models during runtime within suitable driving situations using suitable calculation methods. The determined dynamic components are converted into a few, not necessarily physically motivated, model parameters and transmitted to the vehicle's planning and control algorithm and / or the follow-up control algorithm. By adapting the motion planning and follow-up control depending on the estimated dynamic components, the interaction of both components (motion planning and follow-up control) increases the accuracy with which the vehicle automatically follows a given trajectory.

[0067] Preferably, the device and the method for estimating and adapting at least one model parameter of a dynamic model for lateral guidance and / or longitudinal guidance comprises the storage of past values ​​of desired driving conditions (along the desired trajectories) and of actual driving conditions, which are detected by suitable sensors in and / or on the vehicle.

[0068] An advantageous embodiment of method 200 includes a sensor system, for example, a high-precision DGPS-supported inertial measurement sensor system, for precisely determining the position and motion state of the vehicle. The vehicle can also be referred to as an ego vehicle. Another embodiment of method 200 includes a self-localization function using GPS, self-motion data from the vehicle's own sensors, and the environment model.

[0069] One component of method 200 preferably includes the estimation and adaptation of model parameters of the dynamic model of the longitudinal dynamic and / or transverse dynamic actuator within the planning and / or control process. The transfer behavior between the target value and the output variable of the actuator can be described in a simplified manner using a transfer function in the frequency domain, which is common in systems theory and control engineering, according to equation (1) by combining, in particular a product, a first-order (or higher-order) delay element, for example, a PT1 element, and a dead-time element: GA(s)=YA(s)UA(s)=KATAs+1⋅e−TTs.

[0070] This includes K A the stationary gain and T A the time constant of the delay element, T Tis the dead time, and s is the independent complex variable in the frequency domain. In an advantageous embodiment of method 200, different transfer functions are used for acceleration and deceleration processes.

[0071] Another component of method 200 is the modeling of the vehicle's lateral dynamics within the planning and control process. According to one embodiment of method 200, the kinematic single-track model according to equation (2) is used as the dynamic model. Here, δ wheel the steering angle of the wheel of the modeled single-track vehicle, v the speed, and l the wheelbase, which, as a free parameter, represents the vehicle's dynamics. x, y, ψ are the longitudinal coordinates, the lateral coordinates, and the yaw angle, respectively: ψ˙=vl⋅tan(δwheel), x˙=v⋅cos(ψ), y˙=v⋅sin(ψ).

[0072] According to a further embodiment, which can be combined with the above embodiment, the dynamic model according to equation (3) is used. In this, c v and c h (the cornering stiffnesses on the front and rear axles), l v and l h (the centre of gravity distances between the front and rear axles, for example relative to the centre of gravity of the vehicle), m (the mass of the vehicle) and J z (the moment of inertia around the vertical axis) are the free parameters describing the dynamics of the vehicle, and δ LRW is the steering wheel angle. Compared to the embodiment according to equation (2), this embodiment can also describe the lateral movement of the vehicle in the dynamic range with non-zero slip angles: [y¨ψ¨]=[−cv+chmv−cvlv−chlhmvcvlv−chlhJz(v)−cvlv2−chlh2Jzv][y˙ψ˙]+[cvmcvlvJz]δLRW.

[0073] According to an advantageous embodiment, the method 200 comprises a model-predictive lateral control which calculates a series N of future steering wheel angles δ LRW,soll to stabilize the vehicle along a trajectory: δLRW,set=[δ1, δ2, …δN]T.

[0074] According to a further embodiment, method 200 implements the longitudinal control and / or transverse control using a pure-pursuit control with variable look-ahead and actuator inertia compensation. Its delay element, optionally with dominant pole compensation by a so-called lag compensator (also called a "lead-lag compensator"), is already known from document EP 3 373 095 A1, but without taking a dead time into account in the transfer function.

[0075] For example, method 200 includes various functions for estimating longitudinal dynamics parameters and / or lateral dynamics parameters for the dynamics model used, using the current and stored historical values ​​of the desired trajectories and the actual trajectories at runtime. If the estimation in method 200 is tied to specific driving situations or driving maneuvers—e.g., due to the observability criteria for states or parameters known from systems theory—the method includes functions and / or properties that recognize the required driving maneuvers and, for example, control the execution of the estimation methods in the form of state machines.

[0076] Certain driving situations or maneuvers can also be referred to as driving modes. These driving modes can include, for example, high-speed highway driving with few curves, (tight) cornering at varying speeds, and / or maneuvering.

[0077] According to a further advantageous embodiment, which can be combined with any other embodiment, the dynamic model-dependent components used in planning and control have suitable interfaces for changing model parameters during driving operation (i.e., at runtime). The change in the model parameters is initiated automatically.

[0078] According to a further advantageous embodiment, method 200 includes initializing the estimated values ​​of the respective model parameters with initial values ​​determined empirically and / or by offline fitting (adaptation outside of driving operation). Furthermore, the method includes limiting the estimated values ​​to physically reasonable value ranges (dead times in the range of seconds, vehicle mass that does not deviate too far from production figures, etc.). During runtime, driving maneuvers are detected and model parameters are adjusted, for example, continuously.

[0079] To determine dynamic model parameters, the estimation unit 106 preferably uses the stored historical values ​​of target specifications and implemented actual variables. In an advantageous embodiment of the method 200, the results are calculated directly at runtime. Another embodiment of the method 200 performs the calculation offline using recordings of suitable test runs, for example, during maintenance or while the vehicle is stationary in the fleet. The results can be determined, for example, in the form of characteristic maps or by offline-trained artificial neural networks (networks) at runtime to determine the respective estimated variable.

[0080] In an advantageous embodiment of method 200, artificial neural networks are used to estimate at least one model parameter based on the deviation between the target specification and the implementation of the request, e.g., a steering angle, an acceleration, an engine torque, and / or a brake pressure. In one variant of the method, individual historical values ​​of one or more control or state variables, or sequences of several historical values, can be used.

[0081] Fig. 3 shows an exemplary artificial neural network 300, which is designed as a so-called "General Regression Neural Network" (GRNN), which belongs to the class of radial basis neural networks, and comprises a layer of p neurons. The GRNN 300 further comprises input values ​​u 302 arranged in an N-dimensional vector and an output variable ŷ(u) 304. The GRNN 300 advantageously uses Gaussian curves, which belong to the class of radial basis functions (RBFs), as activation functions A i 312 with a focus on the basis functions and / or activation parameters δl,k˜ 306, where i=1..p indicates the neurons. According to the number of neurons, the input space u ∈ R N evenly distributed. The squared Euclidean distance C i 308 between input variable u k 302 with k=1,..,N and the center of gravity of the radial basis function δl,k˜ 306 is calculated from Ci=∑k=1N(uk−ςl, k˜)2. C i The input variables 302 can also be referred to as "training sample" or "input." The output variable 304 can also be referred to as "output."

[0082] The standard deviation δ 310 defines the width of the activation functions A i 312. In a conventional radial basis network, the activation functions 312 are defined as Ai=exp(−Ci2δ2). A GRNN 300 is characterized by the fact that the activation functions are normalized, Ai=exp(−Ci2δ2)∑j=1pexp(−Cj2δ2) Accordingly, a neuron with activation function A i 312 is activated more strongly, the closer the input value u k 302 is located at the respective support value and / or activation parameter 306 of the RBF. The output of the network is determined by a Θ^_i 314 weighted sum of all activations determined.

[0083] The GRNN 300 can first calculate in one step the weighted sum 316 of the non-normalized activation functions, Ai=exp(−Ci2δ2), 312 and divide by the common normalization factor in a subsequent step 318 to obtain the sum 320 of the normalized activation functions. Alternatively, the activation functions 312 can first be normalized in step 318, Ai=exp(−Ci2δ2)∑j=1pexp(−Cj2δ2), and in the subsequent step 320 the sum of the normalized activation functions is formed.

[0084] The calculation of output variable 304 by neural network 300 can be performed offline or online. Online here primarily means that the process is performed while driving. Additionally, the vehicle can be in constant radio contact with an external computing unit, such as a supercomputer, while driving, and the calculation can be performed outside the vehicle.

[0085] If the activation functions 312 of all neural networks become a vector A and the weights 314 become a vector Θ^_ In summary, the following rule results for determining the error e between predicted and measured estimate: e(u_)=y^(u_)−y(u_)=Θ^_TA_(u_)−Θ_TA_(u), where Θ_ is the vector of ideal weights, Θ^_ the vector of adjusted weights 314 based on training data, ŷ(u) the output 304 calculated by the neural network, for example the GRNN 300, and y(u) the actual output of the system.

[0086] The training of the network takes place at runtime, for example during driving, using an adaptation procedure in which the values ​​θ ̂i314 can be changed at runtime using stored setpoint values ​​and stored actual values ​​for states and / or manipulated variables so that the error e(u) is minimized.

[0087] This involves a conventional iterative gradient descent method with a so-called “momentum term” ΔΘ^_[l], ie the difference for optimization in the I-th iteration step, is used: Θ^_[l+1]=Θ^_[l]+ΔΘ^_[l].

[0088] The “momentum term” can be determined, for example, from the product of error and vector of activation functions with an iteration step width factor η and a parameter 0 ≤ α < 1 and the “momentum term” ΔΘ^_[l−1] of the previous, (I-1)-th, iteration step: ΔΘ^_[l]=−η e(u_)[l]A_(u_)[l]+αΔΘ^_[l−1].

[0089] According to one embodiment, the control system has a pure-pursuit controller that moves along a given reference trajectory X soll = [x soll , y soll , ...] T Depending on the speed, a defined target forecast Δx (difference between current position on the actual trajectory and point of the forecast or reference trajectory) is determined, which in turn is converted into a target curvature and target yaw rate, by which the vehicle is positioned at a (particularly specified) point x v (for example the longitudinal coordinate) the desired (target) value y v (for example, the transverse coordinate). With x soll , y soll In particular, the vehicle's own longitudinal coordinate 402 and the vehicle's own transverse coordinate 404 are Fig. 4.

[0090] Fig.4 shows an example of a model predictive control 400. A vehicle 410 with the vehicle's own longitudinal coordinate 402 and the vehicle's own transverse coordinate 404 ("Vehicle Coordinate System Fixed" or "VCF") is shown at two points 406 ("Point of Interest" or "POI"). For example, the spatial position of "POI 1" is reached at an earlier time than the spatial position of "POI 2." The origin of the vehicle's own coordinate system is centered on the rear axle of the vehicle. "POI 1" is assigned the index 9 on a reference trajectory or target trajectory 408. "POI 2" is assigned the index 12 on the reference trajectory or target trajectory 408. The model predictive control 400 can, for example, calculate the target value assigned to “POI 2” by forward integration from the time step at which “POI 1” is reached.

[0091] According to a further embodiment, which can be combined with all other embodiments, the "virtual" wheelbase of the kinematic single-track model can be estimated using an artificial neural network, in particular the GRNN 300. The wheelbase is referred to as "virtual" because, in addition to the effects of the variation of the actual wheelbase, other, otherwise unconsidered effects, such as slip stiffness, are also taken into account in its estimation. In the method 200 for estimating and adapting the "virtual" wheelbase, past values ​​of relevant variables are analyzed. The relevant variables for estimating the "virtual" wheelbase include vectors of the lateral acceleration a y and the actual steering wheel angle δ LRW,istIn a subsequent processing step, an estimate of the “virtual” wheelbase l is determined using the differential equations (2) of the kinematic single-track model, which is fed into the GRNN 300.

[0092] According to one embodiment, the lateral acceleration and the steering wheel angle represent the two dimensions of the input space, for example, of the GRNN 300, and are discretized, preferably within reasonable value ranges, into p (for example, p = 21) support points. At runtime, the weightings of the resulting p 2 (for example p 2 =441) neurons were adjusted to minimize the error between measurement and prediction.

[0093] According to a further embodiment, a network structure with a one-dimensional input space is selected, in particular the magnitude of the lateral acceleration. This is preferably within a reasonable range of values, for example, an acceleration of 0 m / s. 2 up to 3m / s 2 , divided into p (e.g., p = 21) support points. The weights are adjusted at runtime to minimize the error between the predicted and measured lateral position error.

[0094] A sub-function of the method ensures that only data collected in driving situations where the wheelbase can be observed with respect to lateral acceleration according to suitable system-theoretical criteria is used to train the networks. Furthermore, the estimated wheelbase is limited to suitable minimum and maximum values.

[0095] For example, the enabling function for estimating the virtual wheelbase includes highway driving at high and / or nearly constant speeds and few curves. The enabling function for estimating the virtual wheelbase may also include driving in curves at varying speeds. Alternatively or additionally, the enabling function for estimating the virtual wheelbase may exclude maneuvering above a predetermined limit for the articulation angle, yaw angle, or yaw rate.

[0096] Further embodiments of the method 200 take into account, in addition to the lateral acceleration and the steering angle, further state variables, such as the yaw rate and / or curvature, for example of the trajectory of the vehicle.

[0097] An advantageous embodiment of the method 200 for estimating and adapting the virtual wheelbase comprises the use of longer sequences of recorded historical values. The overall area under investigation is divided into discrete sampling points. Each discrete sampling point is assigned a sampling vector or input vector, for example, the input values ​​302 of the GRNN 300. The network assigned to the input vector 302, for example, the GRNN 300, can be trained online by using differently parameterized reference models, characterized, for example, by the differential equations (2) of the kinematic single-track model. Alternatively or additionally, the method for determining the virtual wheelbase can be trained offline using more complex reference models than the kinematic single-track model and by performing predetermined reference maneuvers, for example, a step specification of target steering angles, and stored in the dynamic model.

[0098] According to an embodiment of the method in which a model-predictive lateral control method is used, the estimated value of the virtual wheelbase is fed directly into its dynamic model.

[0099] Embodiments comprise a device and a method executed by the device for estimating and adapting at least one model parameter of a dynamic model for lateral guidance and / or longitudinal guidance, in which, based on information stored at runtime and with the aid of suitable methods, unconsidered dynamic effects are determined in the form of suitable parameters and made available to the planning algorithm and / or the control system. By determining suitable parameters in the dynamic model of the lateral guidance and / or longitudinal guidance, a significant improvement in the control accuracy of the trajectory tracking is achieved. Alternatively or additionally, the present invention provides a device and a method for estimating nonlinearities in the dynamics of a vehicle for adapting motion planning algorithms and tracking control algorithms for an at least partially automated vehicle.The present invention optimizes the control behavior of the follow-up control.

[0100] Although the invention has been described with reference to exemplary embodiments, it will be apparent to one skilled in the art that various changes may be made and equivalents may be substituted. Furthermore, many modifications may be made to adapt a particular driving situation or application to the teachings of the invention. Consequently, the invention is not limited to the disclosed embodiments, but encompasses all embodiments falling within the scope of the appended claims. List of reference symbols 100 device 102 Sensor / Sensor Interface 104 storage unit 106 Estimation Unit 108 Control unit 200 procedures 202 Recording an actual value 204 Saving the actual value 206 Estimation of a model parameter 208 Control of an actuator depending on the dynamic model adapted by the estimated model parameter 300 Artificial Neural Network 302 Input variable u 304 Output value ŷ 306 Activation parameter ς̃ 308 Squared Euclidean distance between input variable and center of gravity of the radial basis function C i 310 Width of the activation function δ 312 Activation function A i 314 Weighting factor Θ^_ 316 Weighted summation Σ 318 Normalization ÷ 320 Weighted and normalized summation Σ 400 Model predictive control 402 Longitudinal coordinate 404 transverse coordinate 406 point on the actual trajectory 408 point on the target trajectory 410 vehicles

Claims

[1] Device (100) for estimating and adapting at least one model parameter of a dynamic model for lateral guidance and / or longitudinal guidance of a motor vehicle, in particular a commercial vehicle, comprising: - at least one sensor (102) or at least one sensor interface for detecting at least one actual value of a movement state of the transverse guide and / or longitudinal guide; - an estimation unit (106) designed to estimate the at least one model parameter of a dynamic model by comparing at least one desired value of the motion state calculated from the dynamic model of the lateral guidance and / or longitudinal guidance of the motor vehicle with the detected at least one actual value; and - a control unit (108) which is designed to adapt the dynamic model of the lateral guidance and / or longitudinal guidance depending on the detected at least one actual value of the movement state and the estimated at least one model parameter and to control control signals of at least one actuator of the lateral guidance and / or longitudinal guidance depending on the adapted dynamic model, wherein the dynamic model comprises a kinematic single-track model and the motion state comprises a steering angle of a wheel, a steering wheel angle, an amount of speed and / or an amount of acceleration of the motor vehicle, and wherein the estimation unit (106) and / or the control unit (108) is further designed to calculate in advance a time series of N future target values ​​of the motion state, in particular target values ​​of the steering angle, steering wheel angle, speed and / or acceleration, by means of the dynamic model, wherein N is a natural number and greater than or equal to 2, and wherein the estimation unit (106) varies and estimates the at least one model parameter in the dynamic model based on a minimal deviation of several of the target values ​​from the detected actual value of the motion state. [2] The device of claim 1, further comprising: - a storage unit (104) which is designed to store the detected at least one actual value and to output it to the estimation unit (106). [3] Apparatus according to claim 1 or 2, wherein the at least one sensor (102) or the at least one sensor interface comprises a receiving module for receiving signals from a global navigation satellite system, preferably including differential correction signals. [4] Device according to one of claims 1 to 3, wherein the dynamic model comprises slip stiffnesses on the front and rear axles, at least one center of gravity distance of the front and rear axles, a mass of the motor vehicle, tire parameters, a rolling resistance and / or a moment of inertia about a vertical axis of the motor vehicle as model parameters. [5] Device according to one of the preceding claims, wherein the estimation unit (106) is designed to estimate a value of a virtual wheelbase of the kinematic single-track model as a model parameter, wherein the virtual wheelbase takes into account or represents at least one center of gravity distance between the front and rear axles, a distance between the front and rear axles and / or at least one slip stiffness. [6] Device according to one of the preceding claims, wherein the adaptation of the at least one model parameter is calculated by means of an artificial neural network (300). [7] Device according to claim 6, wherein the input space (302) of the artificial neural network (300) comprises a lateral acceleration, a steering angle, a steering wheel angle, a yaw rate and / or a curvature of the trajectory of the motor vehicle. [8] Device according to one of the preceding claims, wherein the transverse guidance and / or the longitudinal guidance comprises a follow-up control for a vehicle in a vehicle convoy. [9] Device according to one of the preceding claims, wherein the estimation unit (106) and / or the control unit (108) is designed to estimate and / or adapt the at least one model parameter of the dynamic model during driving operation and / or comprises at least one interface for changing the at least one model parameter of the dynamic model during driving operation. [10] Device according to one of the preceding claims, wherein the estimation of the at least one model parameter of the dynamic model is adapted depending on a detected driving operating state of the motor vehicle. [11] Device according to one of the preceding claims, wherein the at least one model parameter of the dynamic model is initialized with an initial value measured and / or determined outside of driving operation. [12] Method (200) for estimating and adapting at least one model parameter of a dynamic model for lateral guidance and / or longitudinal guidance of a motor vehicle, in particular a commercial vehicle, comprising the steps: - detecting (202) at least one actual value of a movement state of the transverse guide and / or longitudinal guide by means of at least one sensor or at least one sensor interface; - estimating (206) at least one model parameter of a dynamic model by comparing at least one desired value of the motion state calculated from the dynamic model with the detected at least one actual value; and - controlling (208) at least one actuator of the lateral guidance and / or longitudinal guidance depending on the detected actual state, wherein a control unit (108) of the lateral guidance and / or longitudinal guidance outputs control signals to the actuator depending on the dynamic model adapted by the estimated at least one model parameter, wherein the dynamic model comprises a kinematic single-track model and the motion state comprises a steering angle of a wheel, a steering wheel angle, an amount of speed and / or an amount of acceleration of the motor vehicle, and wherein the estimation unit (106) and / or the control unit (108) is / are further configured to predict a time series of N future target values ​​of the motion state, in particular target values ​​of the steering angle, steering wheel angle, speed and / or acceleration, by means of the dynamic model, wherein N is a natural number and is greater than or equal to 2, and wherein the estimation unit (106) varies and estimates the at least one model parameter in the dynamic model based on a minimal deviation of several of the target values ​​from the detected actual value of the motion state. [13] Motor vehicle (400), in particular commercial vehicle, comprising at least one actuator for lateral guidance and / or longitudinal guidance; and a device (100) for estimating and adapting at least one model parameter of a dynamic model for lateral guidance and / or longitudinal guidance according to one of claims 1 to 11.

Images