Methods for training a neural network and methods for determining the state of a road surface using such a trained neural network

Abstract

A computer-implemented method for training a neural network is described. This network is designed to determine information about the road surface of a machine (2) traveling on it, in order to determine a control parameter of the machine (2) based on this information. The method includes reading (S5) a state variable of the machine (2) as input data and reading (S6) information about the speed of the machine (2) over the ground as input data. Furthermore, the method includes reading (S7) information about at least one road surface as output data and training (S8) the neural network with the input and output data to learn the relationship between the input and output data.Furthermore, a method for determining information about a road surface using a neural network trained in this way is described, as is a method for determining a control parameter as a function of such determined information about the road surface. A computer device (4; 6) for executing such methods and a machine (2) with such a computer device (4) are also described.

Description

Technical area

[0001] The present invention relates to a computer-implemented method for training a neural network configured to determine information about a driving surface of a machine traveling thereon in order to be able to determine a control parameter of the machine depending on the information about the driving surface. Furthermore, the present invention relates to a computer-implemented method for determining information about a driving surface of a machine traveling thereon in order to be able to determine a control parameter of the machine depending on the information about the driving surface. Furthermore, the present invention relates to a computer-implemented method for determining a control parameter of the machine depending on specific information about the driving surface.Furthermore, the present invention relates to a computer device for carrying out such methods, as well as to a machine with such a computer device. State of the art

[0002] It is known from the prior art to determine the condition of a road surface of a machine traveling on it by evaluating image data. For example, DE102020001154 A1 discloses determining the road surface condition using cameras and radar and / or lidar-based sensors as a vehicle's environmental sensor system. Using this determined road surface condition and other acquired information, the damage status of the vehicle can then be determined, for example. However, such image-based determination of the road surface condition requires considerable computing power. Description of the invention

[0003] In a first aspect, the present invention relates to a computer-implemented method for training a neural network. The neural network may, for example, be a connected neural network, such as a fully connected neural network. The training may be reinforcement training or supervised training. The computer-implemented method may be executed by a computer device to train the neural network.

[0004] The neural network is configured to determine information about a driving surface of a machine traveling on it. The machine can be an agricultural machine or a construction machine; alternatively, the machine can be a motor vehicle, such as a passenger car or a truck. Determining information about a driving surface can comprise classifying the driving surface, classifying a roadway, or determining information about a driving surface condition. A driving surface can be, for example, a road, a field, or a meadow. For example, it can be determined whether the machine is on-highway or off-highway, i.e. whether the driving surface is a road or not. Information about a driving surface on which the machine is currently traveling can be determined. Alternatively or additionally, information about a driving surface on which the machine has already traveled or will still travel can be determined.This allows information about the ground the machine is directly in contact with to be determined. Alternatively or additionally, information about the ground around an area the machine is directly in contact with can be determined.

[0005] The neural network is configured to determine information about the driving surface in order to be able to determine a control parameter of the machine depending on the information about the driving surface. The control parameter can be determined using a further, additional method; alternatively, the determination can be carried out using the method described under the first aspect of the invention. A control parameter of the machine can, for example, comprise a control parameter for an engine or a transmission of the machine. For example, a control parameter for controlling the engine can be determined so that the power provided by the engine can be controlled using the control parameter. Alternatively or additionally, a shifting strategy of the transmission of the machine can be controlled using the control parameter, for example an upshift or a downshift of the transmission.

[0006] The method comprises reading in at least one state variable of the machine as input data. Exactly one state variable can be read in, or alternatively, multiple state variables can be read in. The state variable can comprise information about a movement of the machine and, alternatively or additionally, information about the operation of the machine. The movement of the machine can be a kinematic movement of the machine, wherein the movement can be a translational and, alternatively or additionally, a rotational movement of the machine or parts of the machine in up to three spatial directions. Information about the operation of the machine can, for example, comprise information about the operation of the engine or the transmission of the machine. The state variable can, for example, describe a state of the movement of the machine and, alternatively or additionally, the operation of the machine, such as the operation of the engine and, alternatively or additionally, the operation of the transmission.

[0007] The method further includes reading in information about the machine's speed over the ground as input data. Information about the machine's speed can include information about the speed in up to three spatial directions. The ground, for example, the driving surface, can serve as a reference. Thus, information about the machine's speed can be a relative speed of the machine with respect to the driving surface. Information about the machine's speed may not, for example, include direct information about the driving surface, such as the condition of the driving surface, i.e., whether a road or a meadow is being driven on.

[0008] The method further comprises reading in information about at least one driving surface as output data. Information about exactly one driving surface can be read in, or alternatively, information about several different driving surfaces can be read in. This information about at least one driving surface can be read in as a label. For example, information about at least one driving surface can be read in classes, where the classes represent driving surface classes, such as road, meadow, or field. Information about the driving surface can have been previously specified or classified by a user.

[0009] The method further includes training the neural network with the input and output data to learn a relationship between the input and output data. The previously read input and output data can be used for this purpose. The training can be performed using an error function.

[0010] This provides a method for training a neural network to determine information about the driving surface of a machine driving on it. Using a neural network determined in this way, information about a driving surface, or alternatively about different driving surfaces, can be determined. The trained neural network can, for example, output that the driving surface is a field or a meadow. The neural network can be trained in such a way that a relationship between the driving surface, the state variable of the machine, and information about the speed of the machine has been trained by the neural network. The neural network can therefore be trained in such a way that specific properties of the vehicle behavior, for example the state variable or information about the speed of the machine, are trained by the neural network in connection with information about the driving surface.Direct measurement of the driving surface, for example, using imaging sensors, is therefore not necessary for determining information about the driving surface using the trained neural network. Furthermore, determining information about the driving surface can be done more accurately with a neural network trained in this way because information about the speed of the machine is used, rather than just a state variable of the machine, for example, when only movement or operating information of the machine is used to determine information about the driving surface. For example, if the state variable includes movement information of the machine, the movement information can include information about the movement of the machine.It is not always possible to determine from this motion information alone which surface the machine is traveling on and whether the motion information originates from a movement of the machine, for example, due to damping spring movement of a machine's chassis, or from unevenness of the surface, such as in a field. Furthermore, if a state variable and information on the machine's speed are separately input as input data to train the neural network, information on the machine's speed can be input independently of the state variable, without the need to determine information on the machine's speed depending on the state variable, for example, motion information of the machine.This can eliminate an additional step in training the neural network because, for example, acceleration data as movement information of the machine does not have to be integrated to determine information about the machine's speed, which can save time and computing power when training the neural network. Thus, a method is shown in which a complex relationship between state variables, information about the machine's speed, and information about the driving surface can be trained by a neural network. Such a complex relationship cannot, for example, be represented using a rule-based model, or can only be represented in a highly simplified manner. The method thus represents a simplified and less complex way of providing a model, here, for example, a machine learning model trained by the trained neural network, for determining information about the driving surface.

[0011] According to a further embodiment, the method can be characterized in that physical measurement data can be measured and the state variable of the machine can be determined depending on the measured physical measurement data. Physical measurement data can be measured using one or more sensors of the machine. The physical measurement data can be from a specific machine or from a cohort of multiple machines. Physical measurement data can include, for example, acceleration data of the machine, such as transverse acceleration and, alternatively or additionally, rotational acceleration in up to three spatial directions. Alternatively or additionally, the physical measurement data can include information on the operation of the machine. The physical measurement data can include at least one of a rotational speed, a torque, or a power output of a motor or a transmission of the machine.For example, physical measurement data can include the output or input speed of the motor or transmission, a differential speed of the transmission, or a speed of a driven wheel of the machine. Depending on the physical measurement data, motion information and, alternatively or additionally, operating information can be determined as state variables. The specific state variable can be read in when input data is read in. One or more specific state variables relating to motion or operating information of the machine can be read in.

[0012] Thus, the method is designed to train the neural network with measured operating and movement information of the machine and to train the neural network as a function of this and as a function of a relationship between the measured physical measurement data, as well as the information on the speed of the machine and information on the driving surface as output data.

[0013] According to a further embodiment, the method can be characterized in that the measured physical measurement data can be delayed. In this case, the measured physical measurement data of at least one quantity, for example, acceleration or a torque on the transmission, can be delayed. Alternatively, several or all of the measured physical measurement data can be delayed. The delay can be performed using one or more time delay functions. For example, a low-pass filter can be used to delay the measured physical measurement data. In this case, a low-pass filter with a fixed parameter and, alternatively, a low-pass filter with different parameters can be used. Thus, measured physical measurement data delayed either according to one parameter or according to different parameters can be obtained.If measured physical data are delayed with different parameters, for example, with differently parameterized low-pass filters, different delayed physical data can be obtained. Furthermore, the delayed physical data can be read in as input data for training. At least a portion of the delayed physical data, or alternatively, all delayed physical data, can be read in as input data for training. This can be done in addition to reading in the state variable and information on the machine's speed as input data.

[0014] By temporally delaying the measured physical measurement data and reading the temporally delayed physical measurement data as input data for training, recurrence can be provided. For example, acceleration data as motion information and, alternatively or additionally, torque data as machine operating information can be delayed in time and read in as input data for training. This allows recurrence to be implemented via input data using feature engineering without the neural network itself having to have a recurrent structure. Thus, the neural network, as a non-recurrent neural network, can be executed faster and is less computationally intensive than a comparable recurrent neural network. Nevertheless, recurrence can be trained by the neural network using the recurrent input data.With such a trained recurrence, a more precise determination of the information about the driving surface can be carried out with the then trained neural network, even with a computer device with less computing power for determining information about the driving surface with the trained neural network.

[0015] According to a further embodiment, the method can be characterized in that the speed of the machine over ground can also be determined. For example, the speed of the machine over ground can be determined at specific times, for example periodically. The speed of the machine over ground can be determined by the machine or by an external device. The determined speed can be read in when information about the speed of the machine over ground is read in.

[0016] Conventional machines usually already have means for determining the machine's ground speed. Accordingly, the method can utilize these means for determining the machine's ground speed, making retrofitting the machine unnecessary in order to determine the machine's ground speed. Determining the machine's speed can thus be particularly simple. Retrofitting with hardware and, alternatively or additionally, with software for determining the machine's speed is not necessary, for example. This makes such a method easy to apply to a large number of machines.

[0017] According to a further embodiment, the method can be characterized in that determining the speed can comprise measuring the speed with a sensor. The sensor can be part of the machine or part of an external device. The sensor can comprise, for example, a radar sensor, a stereo camera, or generally a non-contact sensor. For example, the machine can comprise a radar sensor or a stereo camera for measuring the speed. The machine can comprise one or more radar sensors or stereo cameras. This allows the speed to be determined independently of wheel slip of a drive wheel of the machine.

[0018] Measured data from the speed measurement step using a radar sensor or stereo camera can initially be independent of information about the driving surface. For example, measured speed data cannot provide direct information about the driving surface, such as whether the machine is traveling over a meadow or a road.

[0019] This allows for simple and precise measurement and determination of the machine's speed. Existing radar sensors on the machine can be used to measure speed. This eliminates the need to retrofit the machine with additional sensors, such as a camera. Furthermore, cameras may be less suitable for determining machine speed than radar sensors.

[0020] According to a further embodiment, the method can be characterized in that determining the speed can include estimating the speed using position data of the machine. The estimation of the speed can be performed, for example, using GPS and alternatively or additionally using GNSS. The estimation can be performed using devices of the machine, for example, using a GPS receiver and a computing device. The estimation of the speed can be performed alternatively or additionally to measuring the speed.

[0021] Thus, the speed can be determined by estimating it using the machine's position data, without the need for additional radar sensors or stereo cameras. Depending on the application and the accuracy required for training the neural network, estimating the speed may be sufficient to provide information about the machine's speed accurately enough to train the neural network. Alternatively, both the speed can be measured and estimated. In this case, the speed can be estimated more often, and the estimated speed can be validated against a measured speed.

[0022] A second aspect of the present invention relates to a computer-implemented method for determining information about the driving surface of a machine traveling thereon. With regard to features that have already been defined with respect to the first aspect of the invention and are also used in the second and further aspects of the invention, reference is made to the previous definition.

[0023] Determining information about a driving surface can comprise classifying a driving surface and, alternatively or additionally, determining the condition of the driving surface. The method can be carried out by a computer device. The machine can be an agricultural or construction machine. Alternatively or additionally, the machine can be a motor vehicle, such as a passenger car or a truck. The method can be configured to determine a control parameter of the machine depending on the information about the driving surface. The control parameter can be determined using a further method, alternatively using methods according to the second aspect. The control parameter can be a control parameter as described in the first aspect of the present invention. The method can use a neural network trained according to an embodiment of the first aspect of the present invention.

[0024] The method can comprise entering input data into the trained neural network. The input data can comprise at least one state variable of the machine and information on the speed of the machine over ground. One or more state variables of the machine can be entered as input data. For example, certain state variables such as movement information or operating information of the machine can be read in as input data depending on measured physical measurement data. For example, movement information can be read in as input data depending on acceleration data of the machine. Alternatively or additionally, operating information can be read in as input data depending on a drive torque of the machine's motor and alternatively or additionally on a differential speed of a transmission of the machine.Information about the machine's speed can, for example, be measured or estimated. Furthermore, the method can include outputting output data from the neural network, wherein the output data can include information about a driving surface. The output can be based on a learned relationship between the input data and the output data.

[0025] Thus, information about the driving surface can be determined in a simple manner using the trained neural network. Such a method for determining information about the driving surface can thus be carried out on a computer device with low computing power. For example, it is not necessary to evaluate computationally intensive image data to determine information about the driving surface. Such a method can therefore also be carried out on a computer device of the machine, such as a control unit, for example a transmission control unit, with relatively little computing power. A GPU, for example, is not necessary to determine information about the driving surface. Furthermore, cameras and means for determining image data are not required to determine information about the driving surface, and thus it is not necessary, for example, to install or retrofit these on the machine.Thus, the method can be used to determine, for example, whether the machine is traveling on or off-road. Depending on information about the road surface, further driving functions can then be controlled using the control parameter.

[0026] A third aspect of the present invention relates to a computer-implemented method for determining a control parameter of the machine. The method can be executed using a computer device, for example a computer device of the machine. Information about a driving surface can be determined using a method according to an embodiment of the second aspect of the present invention. The control parameter of the machine can be determined depending on the determined information about the driving surface. For example, a control parameter for an engine or a transmission of the machine can be determined. For example, a transmission tuning can be controlled using the control parameter. Alternatively or additionally, the control parameter can be determined to control a working process of the machine.The control parameter can be used, for example, to determine and control the ploughing depth of a machine's plow if the machine is an agricultural machine, for example. Alternatively, a gearshift strategy can be called up and controlled using the control parameter. If, for example, the information about the driving surface is determined to be changing from on-highway to off-highway and the machine is driving from a road onto a field, the gearshift strategy can be adapted. For example, the gearshift strategy can be adapted so that the machine drives with the lowest possible fuel consumption both when driving on the road and when driving across a field, whereby the gearshift strategy can be different for road and field. This makes it possible to react to a change in the condition of the driving surface and to adapt the gearshift strategy.Alternatively or additionally, a safety shutdown of the machine can be controlled using the specific control parameter. For example, if it is determined that the driving surface is not changing, but the input data regarding the machine's operating and movement information changes significantly, this may indicate a change in the machine's vibration behavior. For example, the acceleration of the machine may change significantly without the operation of the machine, e.g. a plough on an agricultural machine, or the driving surface changing. This may indicate a sudden defect in the machine. The control parameter can then be determined in such a way that the machine is transferred to a safety state, whereby the machine is braked, for example, and can be brought to a standstill in order to carry out a safety shutdown of the machine or parts of the machine.Many machine functions can depend on information about the road surface, for example, the coefficient of friction between the road surface and the drive wheel. To optimally control these machine functions, it may be necessary to determine information about the road surface in order to determine one or more control parameters based on this information. This allows for optimal, economical, and safe operation of the machine even on changing and varying road surfaces.

[0027] A fourth aspect of the present invention relates to a computer device for carrying out a method according to an embodiment of the three previously described aspects of the invention. The computer device can, for example, be a computer device located externally of the machine, such as a computer or a server. Such a computer device can be configured to carry out at least some steps of a method according to an embodiment of the first aspect of the present invention. Furthermore, the computer device can be a control unit, such as a transmission control unit of the machine. Such a computer device can be configured to carry out steps of methods according to an embodiment of the first, second, and third aspects of the invention. For example, the computer device of the machine can be configured to determine information about a driving surface.Furthermore, this computer device can be configured to determine a control parameter of the machine depending thereon. The computer device of the machine can be configured to measure the physical measurement data and determine the speed. In one embodiment, the computer device of the machine can be configured to perform all steps of the method according to the first aspect for training the neural network.

[0028] A fifth aspect of the present invention relates to a machine having a computer device according to an embodiment of the fourth aspect of the present invention. The machine may be an agricultural machine or a construction machine. Alternatively or additionally, the machine may be a motor vehicle such as a passenger car or a truck. The machine may have a computer device, such as a control unit, for example for carrying out steps of a method according to an embodiment of the first, second and third aspects of the present invention. The machine may be in communicative connection with a further, external computer device, which may be configured to carry out steps of a method according to an embodiment of the first aspect of the present invention.

[0029] A sixth aspect of the present invention relates to a system comprising a machine according to an embodiment of the fifth aspect of the present invention and a computer device configured as a server according to an embodiment of the fourth aspect of the present invention, wherein this computer device is configured to carry out steps of a method according to an embodiment of the first aspect of the present invention. The computer device configured as a server can be configured to first train the neural network. For this purpose, state variables, information on the speed and information on the driving surface can have been previously determined and made available to the external computer device configured as a server. The steps of measuring physical measurement data and determining the speed can be carried out by a computer device of the machine.Information about this can be provided by the machine and, alternatively, by other machines and transmitted to the server, for example, via a communication device. Furthermore, this computer device can be configured to transmit information about the trained neural network to the computer device of the machine. This computer device of the machine can be configured to determine information about the driving surface based on the information received about the trained neural network and to further determine a control parameter based on this determined information about the driving surface. Short description of the characters Fig. 1 schematically shows steps of a method for training a neural network. Fig. 2 schematically shows steps of a method for determining information of a driving surface by means of a neural network and for determining a control parameter depending on the information of the driving surface. Fig. Figure 3 shows schematically a machine and computer equipment for carrying out the Fig. 1 and Fig. 2 schematically shown procedures. Detailed description of embodiments

[0030] Fig. Figure 1 shows a computer-implemented method for training a neural network. Fig. The computer device 6 shown in Figure 3 is configured to execute steps of such a method for training the neural network. The neural network is configured to receive information about a driving surface of a machine 2 traveling thereon, as shown schematically in Fig. 3. Depending on the information about the driving surface, a control parameter of machine 2 can be determined.

[0031] The method comprises measuring S1 physical measurement data. In this process, physical measurement data relating to movement information and operating information of machine 2 are measured. For example, acceleration data of machine 2 is measured. Furthermore, speeds and torques of a motor and a transmission of machine 2 are measured. Furthermore, a state variable of machine 2 is determined S2 depending on the measured physical measurement data. For example, the state variable determined is that the machine is in a certain mode of operation, i.e., that certain speeds and torques are present at the transmission and motor. In this process, operating information is determined as a state variable. Furthermore, the state variable of machine 2 is determined to be that machine 2 has a certain translational and rotational acceleration, i.e., acceleration information of machine 2 is determined as a state variable.

[0032] Furthermore, a temporal delay S3.0 of the measured physical data is applied. The temporal delay S3.0 is performed using differently parameterized low-pass filters.

[0033] Furthermore, a speed of machine 2 over ground is determined S4. The determination S4 is performed by measuring S4.1 the speed with a sensor 8 of machine 2. Sensor 8 is embodied as a radar sensor. In addition, the speed is estimated S4.2 using position data of machine 2. Estimation S4.2 is performed using GPS or GNSS. Thus, both estimation S4.2 and measurement S4.1 are performed, with the estimated speed values ​​being validated using the position data against measured speed values ​​using sensor 8.

[0034] The method further comprises reading in S5 of at least one state variable of the machine 2 as input data. The determined state variables are read in as input data during reading S.5. Furthermore, the time-delayed physical measurement data is read in S3.1 as input data. This enables feature engineering and trains the neural network to recurrence. Furthermore, information on the speed of the machine 2 over the ground is read in S6 as input data. The determined speed is read in during reading S6. Furthermore, information on at least one driving surface is read in S7 as output data. Information on different driving surfaces, classified into different driving surfaces, for example meadow, road or field, is read in as output data.

[0035] The method further comprises training S8 the neural network with the input data and the output data. Thus, the neural network learns a relationship between the input data and the output data.

[0036] Fig. Figure 2 shows a computer-implemented method for determining information about the driving surface of a machine 2 traveling thereon, as well as a computer-implemented method for determining a control parameter of the machine 2. The control parameter is determined as a function of the information about the driving surface. Fig.The method schematically shown in Figure 2 is carried out by a computer device 4 of the machine 2. The computer device 4 is a transmission control unit of the machine 2. Information about the neural network previously trained by the computer device 6 is sent wirelessly and via a communication connection to the computer device 4 of the machine 2. Using this trained neural network, the computer device 4 determines information about the driving surface. Input data is input S9 into the trained neural network. The input data comprise at least one state variable of the machine and information about the speed of the machine 2 above ground. The state variable is determined as a function of measured physical data. This state variable comprises information about the operation and movement of the machine 2.Thus, input data is input into the neural network depending on information about the torque and speed of the transmission and engine of machine 2, as well as depending on the acceleration of machine 2. The information about the speed of machine 2 includes both the measured and estimated speed of machine 2 over the ground. This is done using sensor 8. Furthermore, output data from the neural network is output S10, wherein the output data includes information about a driving surface. The neural network is configured to determine output data depending on the input data.

[0037] Furthermore, depending on the determined information of the driving surface, a determination S11 of the control parameter of the machine 2 takes place. For example, if it is output in step S10 that the driving surface changes from road to field, it is determined in step S11 that a control parameter relating to a shifting strategy of the transmission of the machine 2 is determined as the control parameter in order to adapt the shifting strategy so that the machine 2 can previously drive optimally on the road and now optimally in the field.

[0038] The computer device 4 of the machine 2 is further configured to perform the steps of measuring S1 the physical measurement data and determining S4 the speed. Determining S4 the speed comprises measuring S4.1 and estimating S4.2 the speed. Furthermore, the computer device 4 is configured to send information about this to the computer device 6 so that this computer device 6 can perform the remaining steps for training the neural network. Reference symbol 2 machines 4, 6 Computer setup 8 Sensor S1 Measuring physical data S2 Determining a state variable of the machine S3.0 Delaying the measured physical data S3.1 Reading in the time-delayed physical measurement data S4 Determine the speed of the machine S4.1 Measuring the speed S4.2 Estimating speed S5 Reading the state variable of the machine as input data S6 Reading information about the speed of the machine as input data S7 Reading in information about the driving surface as output data S8 Training the neural network with the input data and the output data S9 Reading input data into the trained neural network S10 Outputting output data of the neural network S11 Determining a control parameter QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] DE 102020001154 A1

[0002]

Claims

[1] Computer-implemented method for training a neural network which is set up to determine information about a driving surface of a machine (2) traveling thereon in order to be able to determine a control parameter of the machine (2) as a function of the information about the driving surface, comprising the steps of: reading in (S5) at least one state variable of the machine (2) as input data; reading in (S6) information about the speed of the machine (2) over the ground as input data; reading in (S7) information about at least one driving surface as output data; training (S8) the neural network with the input data and output data in order to learn a relationship between the input data and the output data. [2] Method according to claim 1, characterized bythat furthermore a measurement (S1) of physical measurement data and a determination (S2) of the state variable of the machine (2) takes place as a function of the measured physical measurement data, wherein the determined state variable is read in as input data during the reading in (S5). [3] Method according to claim 2, characterized by that a temporal delay (S3.0) of the measured physical measurement data takes place and a reading (S3.1) of the temporally delayed physical measurement data takes place as input data for training (S8). [4] Method according to one of the preceding claims, characterized by that further a determination (S4) of the speed of the machine (2) over ground takes place, and the determined speed is read in when reading in (S6) information on the speed of the machine (2) over ground. [5] Method according to claim 4, characterized bythat determining (S4) the speed comprises measuring (S4.1) the speed with a sensor (8). [6] Method according to one of claims 4 or 5, characterized by that determining (S4) the speed comprises estimating (S4.2) the speed using position data of the machine (2). [7] Computer-implemented method for determining information about a driving surface of a machine (2) traveling thereon in order to be able to determine a control parameter of the machine (2) depending on the information about the driving surface, the method using a neural network trained according to the preceding claims, the method comprising the steps of: Inputting (S9) input data into the trained neural network, wherein the input data comprise at least one state variable of the machine (2) and information on the speed of the machine (2) over the ground, and outputting (S10) output data of the neural network, wherein the output data comprise information on a driving surface. [8] Computer-implemented method for determining a control parameter of the machine (2), wherein information of a driving surface is determined using a method according to the preceding claim, and a determination (S11) of the control parameter of the machine (2) takes place as a function of the determined information of the driving surface. [9] Computer device (4; 6) for carrying out a method according to one of the preceding claims. [10] Machine (2) with a computer device (4) according to the preceding claim.

Images