Vehicle control system

Abstract

A vehicle control system is provided for estimating a total roll angle of a vehicle. The vehicle control system includes a sensor system for outputting current sensor data representing information about a physical parameter related to a current movement of the vehicle. Further, the vehicle control system includes a total roll angle estimation unit including a neural network for estimating a time-dependent total roll angle of the vehicle currently moving on a road surface based on at least some of the current sensor data.

Description

Technical Field

[0001] The present invention relates to a vehicle control system, including a sensor system for outputting current sensor data representing information about physical parameters related to the current movement of a vehicle, in particular an automobile. Background Art

[0002] An increasing number of vehicles, in particular automobiles, are equipped with vehicle control systems that assist the driver of the vehicle in demanding driving situations. Vehicle control systems are also necessary in the context of autonomous driving. Physical parameters related to vehicle movement are provided as input data based on which the vehicle control system can analyze the characteristics of vehicle movement and control the movement through feedback operations of brakes, steering, engine, etc.

[0003] In this field, many techniques have been developed for determining yaw rate and sideslip angle for vehicle control purposes. Less attention has been paid to vehicle roll angle.

[0004] For example, a vehicle is equipped with an electronic stability program (ESP) or a similar vehicle dynamic stability control system that provides yaw rate control and sideslip correction control based on the detected yaw rate and estimated sideslip angle during current driving of the vehicle. Figure 1 A vehicle control system 100 providing ESP function provided by the prior art is shown. The vehicle control system 100 includes a vehicle control unit 110 and a sensor system 120. The sensor system 120 includes a plurality of sensor devices, such as a 3D inertial measurement unit (IMU) sensor device, a controller area network (CAN) bus sensor device, and a wheel speed sensor device. A model 130 of desired vehicle dynamics is stored in the vehicle control unit 110, and the vehicle control unit 110 includes a vehicle control 140 for controlling the yaw rate and sideslip of the vehicle equipped with the vehicle control system 100 through feedback operations of brakes, motors, and actuators. The deviation between the detected yaw rate and the yaw rate target value provided by the model 130 of desired vehicle dynamics is input into the vehicle controller 140 for controlling the yaw rate of the actual movement of the vehicle. Based on the sensor data provided by the sensor system 120, the sideslip angle βest is estimated by a sideslip estimation unit 150 including a Kalman filter unit 154. The deviation between the estimated sideslip angle βest and the sideslip angle target value provided by the model 130 of desired vehicle dynamics is input into the vehicle control 140 for controlling the sideslip of the actual movement of the vehicle.

[0005] The sideslip dynamics of a vehicle can be approximated by the following equation:

[0006] where φ = φ r + φ s

[0007] where, represents the time derivative of the lateral component of the vehicle speed vector, ay, IMU represents the lateral acceleration of the vehicle obtained by the IMU sensor device based on the sensor system 120, rIMU represents the yaw rate provided by the yaw rate sensor of the sensor system 120, vx represents the longitudinal component of the vehicle speed vector, and g represents the gravitational constant. Φ represents the total roll angle, which is obtained from the sum of the road slope angle (the lateral inclination angle of the road with respect to the horizontal plane) Φr and the (relative) vehicle roll angle Φs with respect to the (inclined) road surface.

[0008] However, the total roll angle Φ is usually unknown. Therefore, the estimation of the sideslip angle, especially when the vehicle is moving on a sloped road surface (Φr ≠ 0) (such as in a sloped curve), is too inaccurate due to the lack of knowledge of the total roll angle Φ (usually directly ignored or only roughly estimated). Overestimating the sideslip angle can lead to inappropriate interactions in vehicle control, especially in sloped curves.

[0009] In the prior art, this problem is solved by stopping the ESP operation or at least stopping its sideslip control based on a heuristic algorithm that identifies the presence of a sloped curve on which the vehicle is currently moving. However, this method not only requires a time-consuming calibration process but also reduces safety due to partial or complete suppression of the ESP sideslip feedback control. Another method includes providing additional sensor devices for directly measuring the actual sideslip of the vehicle. However, such high-precision sideslip angle sensors are very costly. SUMMARY OF THE INVENTION

[0010] In view of the above, the object of the present application is to provide a vehicle control system that can appropriately consider the current total roll angle of a moving vehicle without requiring a costly sideslip angle sensor device.

[0011] The above and other objects are achieved by the subject matter claimed in the independent claims. Other implementations are apparent from the dependent claims, the description, and the drawings.

[0012] According to a first aspect, a vehicle control system is provided, including: a sensor system for outputting current sensor data, the current sensor data representing information about physical parameters related to the current movement of the vehicle; a total roll angle estimation unit including a neural network, the neural network being configured to estimate a time-dependent total roll angle of the vehicle currently moving on a road surface based on at least some of the current sensor data.

[0013] It should be noted that the term "road" herein should be understood in the broadest sense and represents any driving surface, including roads, streets, highways, dirt roads, and even off-road paths, etc. In particular, the road surface can be a sloped road surface, such as the road surface of a sloped curve. The total roll angle is obtained by the sum of the road slope angle (the lateral inclination angle of the road with respect to the horizontal plane) and the (relative) vehicle roll angle of the vehicle with respect to the (tilted) road surface. The vehicle can be an automobile, such as an electric vehicle. The sensor system can include an inertial measurement unit sensor device, a wheel speed sensor device, and a wheel steering angle sensor device. Physical parameters related to the current movement of the vehicle can include the yaw angular velocity, longitudinal acceleration, and lateral acceleration of the vehicle, wheel speed, and steering angle, etc. Needless to say, the term "neural network" herein is used to represent an artificial neural network.

[0014] The vehicle control system according to the first aspect provides an accurate real-time estimation of the total roll angle of the vehicle, especially a vehicle currently moving in a sloped curve, based on sensor data and artificial intelligence (AI) (such as deep learning) implemented by a neural network, so that any control of the vehicle movement that may or should depend on the total roll angle of the vehicle in a certain aspect can be improved. For example, the sideslip control (see the following description) can be improved based on the provided estimation of the total roll angle, or the vehicle rollover can be detected or suppressed based on the provided estimation of the total roll angle.

[0015] According to one implementation, the neural network used by the total roll angle estimation unit is or includes a recurrent neural network. A recurrent neural network is designed to process data in a time series / sequence, and thus, an accurate real-time estimation of the total roll angle can be achieved. In particular, the recurrent neural network can be or include a long short-term memory (LSTM) neural network or layer. The LSTM neural network is a powerful recurrent neural network, especially designed to identify patterns in data sequences. In this context, they can be appropriately used to process the numerical time series of sensor data provided by the sensor system of the vehicle control system.

[0016] According to one implementation, the vehicle control system further includes a sideslip estimation unit configured to estimate a sideslip angle of the vehicle currently moving on the road surface based on at least some of the current sensor data and the estimated total roll angle. Since sideslip dynamics depend on the total roll angle, the accuracy of sideslip angle estimation can be improved by considering the estimated total roll angle provided by the neural network. Accordingly, any control of the vehicle based on the estimated sideslip angle can be improved.

[0017] According to one implementation, the sideslip estimation unit includes a Kalman filter unit, particularly an extended / nonlinear Kalman filter unit, configured to estimate the sideslip angle of the vehicle currently moving on the road surface based on at least some of the current sensor data, the estimated total roll angle, and a nonlinear vehicle model. The predictor-corrector algorithm of the Kalman filter can accurately estimate the sideslip recursively during vehicle movement.

[0018] As previously mentioned, the estimated sideslip angle can be used for sideslip control. Accordingly, according to one implementation, the vehicle control system further includes a sideslip control unit configured to control the sideslip of the vehicle currently moving on the road surface based on the estimated sideslip angle. Since the sideslip angle is estimated in consideration of the estimated total roll angle provided by the total roll angle estimation unit of the vehicle control system, sideslip control can be performed more accurately compared to the prior art.

[0019] In the context of a vehicle dynamic stability control system (e.g., electronic stability control (ESC) or dynamic stability control (DSC), dynamic stability and traction control (DSTC) or active stability control (ASC)), accurate sideslip control is important as it helps a driver of the vehicle to control the yaw rate and sideslip through feedback operations of brakes, electric motors, actuators, etc. in demanding driving situations (e.g., high-speed driving in a ramp curve). Accordingly, according to one implementation, the sensor system includes a yaw rate sensor device and further includes a yaw rate control unit in addition to the sideslip control unit, wherein the yaw rate sensor device is configured to acquire the yaw rate of the vehicle currently moving on the road surface, and the yaw rate control unit is configured to control the yaw rate of the vehicle currently moving on the road surface based on the acquired yaw rate.

[0020] According to another implementation, the neural network of the total roll angle estimation unit is trained based on previously acquired values of the total roll angle, which are measured during a previous movement of the vehicle and are related to previously acquired sensor data acquired by the sensor system during the previous movement. For the training process, according to this implementation, the training data is provided by the real movement of a training vehicle, which is equipped with an expensive high-precision roll angle sensor and is similar to a vehicle on which a vehicle control system according to the first aspect and any of its implementations can be installed. The output of the expensive high-precision roll angle sensor of the training vehicle represents the desired / target value that the neural network must be trained for. It should be noted that the neural network can also be used for further self-learning after the training is completed based on the previous movement, for example, based on the interaction (steering, braking) of the driver of a vehicle equipped with the neural network (and without a high-precision roll angle sensor).

[0021] The previous (training) movement of the vehicle can include at least one of the following movements: moving in a loop on a slope road and a flat road at relatively high and relatively low speeds respectively; moving in a slope curve with different radii at different speeds and different road slope angles; moving in a serpentine shape at relatively high and relatively low speeds respectively on a flat road and a slope road with different road slope angles; turning at different turning radii during braking until stopping; moving at relatively high and relatively low speeds respectively during oversteering and understeering maneuvers. The use of various different training movements can improve the efficiency of the process of training the neural network.

[0022] According to a second aspect, a method for estimating the time-dependent total roll angle of a vehicle moving on a road surface is provided, including: outputting current sensor data, where the current sensor data represents information about physical parameters related to the current movement of the vehicle on the road surface; receiving at least some of the output sensor data (or data based on the sensor data) through a neural network; processing the received sensor data through the neural network to estimate the total roll angle.

[0023] According to an implementation of this method, the neural network is a recurrent neural network. The recurrent neural network may be a long short-term memory (LSTM) neural network.

[0024] According to an implementation, the method further includes: training the neural network based on previously acquired values of the total roll angle, which are measured during a previous movement of the vehicle and are related to previously acquired sensor data acquired by the sensor system during the previous movement.

[0025] These previous movements of the vehicle may include at least one of the following movements: moving in a loop on a slope road and a flat road at relatively high and relatively low speeds respectively; moving in a slope curve with different radii at different speeds and different road slope angles; moving in a serpentine shape at relatively high and relatively low speeds respectively on a flat road and a slope road with different road slope angles; turning at different turning radii during braking until stopping; moving at relatively high and relatively low speeds respectively during oversteering and understeering maneuvers.

[0026] According to another implementation, the method further includes: estimating a sideslip angle of the vehicle currently moving on the road surface based on at least some of the current sensor data and the estimated total roll angle. The sideslip angle can be estimated by a Kalman filter unit based on at least some of the current sensor data, the estimated total roll angle, and a non-linear vehicle model.

[0027] According to another implementation, the method further includes: controlling the sideslip of the vehicle currently moving on the road surface based on the estimated sideslip angle.

[0028] According to another implementation, the sensor data includes the yaw rate of the vehicle currently moving on the road surface, and the method further includes: controlling the yaw rate of the vehicle currently moving on the road surface based on the yaw rate.

[0029] The method according to the second aspect and the implementations of the method according to the second aspect provide the same advantages as the above-mentioned vehicle control system according to the first aspect and its implementations, and can be implemented in the above-mentioned vehicle control system according to the first aspect and its implementations. On the other hand, the vehicle control system according to the first aspect and its implementations can be used to execute the method according to the second aspect and its implementations.

[0030] According to a third aspect, there is provided a computer program product including computer-readable instructions for performing the steps of the method according to the second aspect or any of its implementations when run on a computer.

[0031] According to a fourth aspect, there is provided a vehicle, an automobile, an electric vehicle, an automated guided vehicle, or a mobile robot, which includes the vehicle control system according to the first aspect or any implementation of the first aspect.

[0032] One or more embodiments will be described in detail in the accompanying drawings and the following description. Other features, objectives, and advantages will be apparent from the specification, the drawings, and the claims. Description of the Drawings

[0033] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0034] Figure 1 Shows a vehicle control system provided by the prior art.

[0035] Figure 2 Shows a vehicle control system provided by an embodiment.

[0036] Figure 3 Shows the total roll angle estimated by a method for estimating the time - related total roll angle of a vehicle moving on a road surface by a vehicle control system or an estimator provided by an embodiment.

[0037] Figure 4 Shows a neural network architecture applicable to a vehicle control system provided by an embodiment.

[0038] Figure 5 Shows a vehicle control system provided by an embodiment.

[0039] Figure 6 Shows the operation of an extended Kalman filter of a total roll angle estimation unit of a vehicle control system provided by an embodiment.

[0040] Figure 7 Shows the operation of another extended Kalman filter of another total roll angle estimation unit of a vehicle control system provided by an embodiment.

[0041] Figure 8 Shows a method for estimating the time - related total roll angle of a vehicle moving on a road surface provided by an embodiment. Detailed implementation

[0042] This document provides a vehicle control system for a vehicle (such as an automobile). The provided vehicle control system includes a total angle estimation unit, and the total angle estimation unit is used to estimate the total roll angle of the vehicle through a neural network. For example, the estimation of the total roll angle can be used to improve the estimation of the sideslip angle of the vehicle and sideslip control, especially in the context of the operation of an ESP system. For example, the estimation of the total roll angle can be used for vehicle rollover detection and prevention.

[0043] In particular, the provided vehicle control system can accurately control the sideslip of a vehicle moving in a ramp curve based on the estimated total roll angle. The provided vehicle control system is reliably applicable to driving situations on a road surface with a zero lateral slope / tilt angle and a ramp road surface. For example, when the vehicle is moving in a ramp curve with a relatively large tilt angle greater than 5° or 10°, the vehicle control system may be particularly useful. However, the total angle estimation unit does not falsely detect a large tilt angle (false positive) on a flat road, which would again significantly deteriorate the sideslip angle estimation.

[0044] Figure 2 Figure 2 shows an embodiment of the vehicle control system 200 provided by the present invention. The vehicle control system 200 can be installed in a vehicle (e.g., an automobile), and the vehicle control system 200 includes a sensor system 210 for providing a plurality of physical parameters related to the movement of the vehicle. The sensor system 210 can include an inertial measurement unit sensor device, a wheel speed sensor device, and a wheel steering angle sensor device. The inertial measurement unit sensor device can include a yaw rate sensor device, a sensor device for detecting the longitudinal acceleration of the vehicle, and a sensor device for detecting the lateral acceleration of the vehicle.

[0045] In addition, the vehicle control system 200 includes a total roll angle estimation unit 220 for estimating the total roll angle of the vehicle during the current movement of the vehicle. The total roll angle of the vehicle is estimated based on the data provided by the sensor system 210. In particular, some physical parameters, such as the longitudinal speed of the vehicle, can be estimated based on the sensor data by a suitable algorithm and used to estimate the total roll angle. The total roll angle estimation unit 220 includes a neural network 225, which is trained to estimate the total roll angle based on the sensor output data provided by the sensor system 210. The vehicle control system 200 can also include vehicle control for controlling components of the vehicle, such as brakes, engines, motors, actuators, etc. (see also the following description), based on the estimated total roll angle.

[0046] The total roll angle Φ estimated by any embodiment of the vehicle control system provided herein and any embodiment of the method for estimating the total roll angle of a moving vehicle provided herein, in particular, the total roll angle estimated by the total roll angle estimation unit 220 of the vehicle control system 200 is Figure 3 shown. Figure 3 Figure 3 shows an automobile 310 moving on a ramp road surface 320, which is laterally inclined at a road slope angle Φr relative to the horizontal plane. The total roll angle Φ is obtained by the sum of the road slope angle (the lateral inclination angle of the road with respect to the horizontal plane) Φr and the (relative) vehicle roll angle Φs of the vehicle with respect to the (inclined) road surface 320. It should be noted that, in Figure 3 the example shown, when measured on the left side of the plane oriented perpendicular to the road surface 320, the vehicle roll angle Φs has a negative sign. Any other sign convention can also be used when considered appropriate.

[0047] In particular, the vehicle control system 200 can estimate the total roll angle Φ for relatively large road slope angles Φr > Φs or Φr >> Φs. For relatively large road slope angles, accurate estimation of the total roll angle is particularly important for accurate sideslip control (see the following description).

[0048] The neural network 225 is trained based on previous training maneuvers of vehicles similar to the vehicle equipped with the vehicle control system 200. During these previous training maneuvers, the total roll angle Φ corresponding to the sensor output data acquired during the previous training maneuvers is provided as the target value / desired output of the neural network for training the neural network 225. The total roll angle Φ for training is acquired by appropriately designed expensive high-precision total roll angle sensors, which are not provided for mass-produced products and do not exist in vehicles equipped with a vehicle control system such as the vehicle control system 200.

[0049] The training maneuvers may include at least one of the following maneuvers of the training vehicle and / or any combination thereof deemed appropriate: moving in a loop on a slope road and a flat road at relatively high and relatively low speeds respectively (e.g., in the range of 250 km / h to 50 km / h or a sub-range thereof); moving in a slope curve with different radii (e.g., in the range of 50 m to 150 m or a sub-range thereof) at different speeds (e.g., in the range of 250 km / h to 50 km / h or a sub-range thereof) and different road slope angles (e.g., 5° to 25° or a sub-range thereof); moving in a serpentine shape on a flat and slope road with different road slope angles (e.g., 5° to 25° or a sub-range thereof) at relatively high and relatively low speeds respectively (e.g., in the range of 250 km / h to 50 km / h or a sub-range thereof); moving during braking, turning at different turning radii until stopping; moving at relatively high and relatively low speeds respectively (e.g., in the range of 250 km / h to 50 km / h or a sub-range thereof) during oversteering and understeering maneuvers respectively. Using a combination of all or some of these operations can speed up the training process and improve the accuracy of the acquired total roll angle estimation.

[0050] In principle, any suitable neural network can be used for the neural network 225. However, the neural network must be suitable for processing time series of data. For example, the neural network can be or include a (deep) recurrent neural network. In particular, the recurrent neural network can be or include long short-term memory (LSTM) or other neural networks mimicking long-term memory. The LSTM layer consists of a set of recurrently connected blocks called memory blocks. These blocks can be considered a differentiable version of the memory chips of a digital computer. Each block contains one or more recurrently connected memory units and three multiplicative units - an input gate, an output gate, and a forget gate - providing continuous write, read, and reset operations for the units.

[0051] Figure 4Shows a specific example of a suitable LSTM neural network architecture 400. The LSTM neural network architecture 400 includes an input layer 410 for receiving sensor data or input data based on sensor data. The data includes the yaw rate, longitudinal acceleration, and lateral acceleration of the vehicle, wheel speed, and steering angle, as well as parameters estimated based on sensor output data, such as longitudinal tire force and vehicle speed. According to a specific example, the data input to the input layer 410 of the LSTM neural network architecture 400 is a multivariate time series of 100 samples of 14 different features / parameters provided by a vehicle sensor system.

[0052] In Figure 4 the example shown, the data received by the input layer 410 is processed by two bilLSTM layers 420, one forward LSTM layer and one backward LSTM layer. The outputs of the two bilLSTM layers 420 are received by a linear layer 430, which implements a linear activation function for outputting an estimated total roll angle.

[0053] Figure 5 Shows a vehicle control system 500 provided by an embodiment that represents or includes an electronic stability program (ESP). It can be similar to Figure 2 the vehicle control system 200 shown. The vehicle control system 500 includes a vehicle control unit 510 and a sensor system 520. The sensor system 520 includes multiple sensor devices, such as 3D inertial measurement unit (IMU) sensor devices, CAN bus sensor devices, and wheel speed and steering angle sensor devices. For example, a model 530 of desired vehicle dynamics is stored in the vehicle control unit 510, and the vehicle control unit 510 includes a vehicle control 540, which includes a feedback (FB) yaw rate controller 542 and an FB sideslip controller 544, and the FB yaw rate controller and the FB sideslip controller are respectively used to control the yaw rate and sideslip of a vehicle equipped with the vehicle control system 500 through feedback operations of brakes, motors, and actuators.

[0054] The deviation of the detected yaw rate from the yaw rate target value provided by the model 530 of the desired vehicle dynamics is input into the vehicle control 540 / FB yaw rate controller 542 for controlling the yaw rate of the actual movement of the vehicle. The sideslip estimation unit (observer) 550 estimates the sideslip angle βest (= arctan (lateral vehicle speed component / longitudinal vehicle speed component)) based on the sensor data provided by the sensor system 520, and inputs the deviation of the estimated sideslip angle βest from the sideslip angle target value provided by the model 530 of the desired vehicle dynamics into the vehicle control 540 / FB sideslip controller 544 for controlling the sideslip of the actual movement of the vehicle.

[0055] Unlike Figure 1 the configuration of the prior art shown, the sideslip estimation unit 550 of the vehicle control system 500 provided by one embodiment includes a total roll angle estimation unit (observer) 552, and the total roll angle estimation unit (observer) implements a deep learning roll angle model in a recurrent neural network (such as similar to Figure 4 the LSTM neural network shown). The total roll angle estimation unit 552 outputs the estimated total roll angle based on the sensor data output by the sensor system 520, particularly the sensor data output by the wheel speed sensors of the sensor system 520. The estimated total roll angle is input into a Kalman filtering unit 554 (for example, an extended Kalman filtering unit), which is also included by the total roll angle estimation unit 552. The Kalman filtering unit 554 is used to recursively estimate the sideslip angle of the vehicle. The sideslip dynamics of the vehicle can be approximated by the following equation:

[0056] where φ = φ r + φ s

[0057] where, represents the time derivative of the lateral component of the vehicle speed vector, ay,IMU represents the lateral acceleration of the vehicle obtained based on the IMU sensor device of the sensor system 120, rIMU represents the yaw rate provided by the yaw rate sensor of the sensor system 120, vx represents the longitudinal component of the vehicle speed vector, g represents the gravitational constant. Φ represents the total roll angle, which is obtained by the sum of the road slope angle Φr and the vehicle roll angle Φs relative to the road surface on which the vehicle travels (see also Figure 3 ). It should be noted that the longitudinal speed vx can be obtained through or based on the sensor data output by the sensor device (such as a CAN sensor device) of the sensor system 120.

[0058] Differing from the prior art, since the total roll angle estimation unit 552 is provided, the total roll angle Φ can be accurately considered in the sideslip dynamics. Based on the sensor data and the estimated total roll angle, the Kalman filter unit 554 accurately estimates the sideslip angle βest in a time-dependent manner based on the predictor-corrector algorithm. Therefore, based on the accurate estimation of the sideslip angle βest provided by the Kalman filter unit 554 / sideslip estimation unit 550, the vehicle control 540 can appropriately control the sideslip by controlling the brakes, motors, and actuators.

[0059] The Kalman filter unit 554 included in the sideslip estimation unit 550 can be an extended / nonlinear Kalman filter unit that uses a nonlinear vehicle model and sensor data, as Figure 6 and Figure 7 shown. The nonlinear vehicle model can be a model based on classical physics, considering the rigid body movement of the vehicle based on the principles of kinematics and Newtonian dynamics. As an alternative or supplement, if considered appropriate, a vehicle model based on a neural network can be used.

[0060] According to Figure 6 the example shown, the extended Kalman filter unit 630 for estimating the sideslip angle of the vehicle receives the data output by the total roll angle estimation unit 610 and the sensor system 620. Based on the nonlinear vehicle model 632, the input sensor data is processed to obtain a prior state estimate of the sideslip angle. Based on the sensor model 734 provided with both the sensor data provided by the sensor system 620 and the total roll angle estimate provided by the total roll angle estimation unit 610, the prior state estimate (prediction) is recursively updated / corrected in order to obtain an estimate βest of the sideslip angle of the vehicle.

[0061] According to Figure 7 the example shown, the extended Kalman filter unit 730 for estimating the sideslip angle of the vehicle receives the data output by the total roll angle estimation unit 710 and the sensor system 720. Based on the sensor model 732, the input data (both the estimate of the total roll angle and the sensor data) is processed to obtain a prior state estimate of the sideslip angle. Based on the nonlinear vehicle model 734 provided with the sensor data, the prior state estimate (prediction) is recursively updated / corrected to obtain an estimate βest of the sideslip angle of the vehicle.

[0062] All embodiments of the above vehicle control system can be used to execute the method 800 for estimating the time-dependent total roll angle of a vehicle moving on a road surface, as Figure 8As shown. Method 800 may be implemented in any of the above embodiments of the vehicle control system. The method 800 for estimating the time-related total roll angle of a vehicle moving on a road surface includes the following steps: Output (S810) current sensor data obtained for the current movement of the vehicle. The sensor data represents information about physical parameters (yaw rate, acceleration, wheel speed, wheel steering angle, etc.) related to the current movement of the vehicle. The neural network of the total roll angle estimation unit receives at least some of the sensor data (or data based on at least some of the sensor data) (S820). The neural network was previously trained based on previously obtained sensor data and actual measurements of the total roll angle related to the previously obtained sensor data, and it processes (S830) the received data to obtain an estimate of the total roll angle of the vehicle. The total roll angle estimated thereby can be used to control the sideslip of the vehicle currently traveling on the road surface.

[0063] All of the previously discussed embodiments are not intended as limitations, but rather as examples for illustrating the features and advantages of the present invention. It should be understood that some or all of the above features may also be combined in different ways.

Claims

1. A vehicle control system (200, 500), characterized in that, Comprising: A sensor system (210, 520, 620, 720) for outputting current sensor data, the current sensor data representing information about physical parameters related to the current movement of the vehicle; A total roll angle estimation unit (220, 552, 610, 710) including a neural network (225, 400), the neural network being configured to estimate a time-dependent total roll angle of the vehicle currently moving on the road surface based on at least some of the current sensor data; 2. The vehicle control system (200, 500) according to claim 1, characterized in that, The neural network (225, 400) is a recurrent neural network; 3. The vehicle control system (200, 500) according to claim 2, characterized in that, The recurrent neural network (225, 400) includes a long short-term memory (LSTM) neural network (225, 400); 4. The vehicle control system (200, 500) according to any one of the above claims, characterized in that, Further comprising a sideslip estimation unit (550), the sideslip estimation unit (550) being configured to estimate a sideslip angle of the vehicle currently moving on the road surface based on at least some of the current sensor data and the estimated total roll angle; 5. The vehicle control system (200, 500) according to claim 4, characterized in that, The sideslip estimation unit (550) includes a Kalman filtering unit (554, 630, 730), the Kalman filtering unit being configured to estimate the sideslip angle of the vehicle currently moving on the road surface based on the at least some of the current sensor data, the estimated total roll angle, and a non-linear vehicle model; 6. The vehicle control system (200, 500) according to claim 4 or 5, characterized in that, Further comprising a sideslip control unit for controlling the sideslip of the vehicle currently moving on the road surface based on the estimated sideslip angle; 7. The vehicle control system (200, 500) according to claim 6, characterized in that, The sensor system (210, 520, 620, 720) includes a yaw rate sensor device and a yaw rate control unit, wherein the yaw rate sensor device is configured to acquire a yaw rate of the vehicle currently moving on the road surface, and the yaw rate control unit is configured to control the yaw rate of the vehicle currently moving on the road surface based on the acquired yaw rate; 8. The vehicle control system (200, 500) according to any one of the above claims, characterized in that, The neural network (225, 400) is trained based on previously acquired values of the total roll angle, the previously acquired values of the total roll angle being measured during a previous movement of the vehicle and being related to previously acquired sensor data acquired by the sensor system (210, 520, 620, 720) during the previous movement; 9. The vehicle control system according to claim 8, wherein The previous movement of the vehicle includes at least one of the following movements: Circular movement on a slope road and a flat road at relatively high and relatively low speeds respectively; Movement in a slope curve with different radii at different speeds and different road slope angles; Serpentine movement at relatively high and relatively low speeds respectively on a flat road and a slope road with different road slope angles; During braking, turning at different turning radii until stopping; During oversteering and understeering maneuvers, moving at relatively high and relatively low speeds respectively; 10. The vehicle control system (200, 500) according to any one of the above claims, characterized in that, The sensor system includes an inertial measurement unit sensor device, a wheel speed sensor device, and a wheel steering angle sensor device; 11. A method (800) for estimating a time-dependent total roll angle of a vehicle moving on a road surface, characterized in that, Comprising: Output (S810) current sensor data, where the current sensor data represents information about physical parameters related to the current movement of the vehicle on the road surface; Receive (S820) at least some of the output sensor data of the output sensor data through a neural network (225, 400); Process (S830) the received sensor data through the neural network (225, 400) to estimate the total roll angle.

12. The method (800) according to claim 11, wherein The neural network (225, 400) is a recurrent neural network.

13. The method (800) according to claim 12, wherein, The recurrent neural network (225, 400) is a long short-term memory (LSTM) neural network (225, 400).

14. The method (800) according to any one of claims 11 to 13, characterized in that, Further comprising: Train the neural network (225, 400) based on previously obtained values of the total roll angle, where the previously obtained values of the total roll angle are measured during a previous movement of the vehicle and are related to previously obtained sensor data acquired by a sensor system (210, 520, 620, 720) during the previous movement.

15. The method (800) according to claim 14, wherein The previous movement of the vehicle includes at least one of the following movements: Circular movement on a ramp road and a flat road at relatively high and relatively low speeds, respectively; Movement in a ramp curve with different radii at different speeds and different road slope angles; Serpentine movement on a flat road and a ramp road with different road slope angles at relatively high and relatively low speeds, respectively; During braking, turn at different turning radii until stopping; During oversteering and understeering maneuvers, move at relatively high and relatively low speeds, respectively.

16. The method (800) according to any one of claims 11 to 14, characterized in that, Further comprising: Estimate the slip angle of the vehicle currently moving on the road surface based on at least some of the current sensor data and the estimated total roll angle.

17. The method (800) according to claim 16, characterized in that, The slip angle is estimated by a Kalman filter unit (554, 630, 730) based on at least some of the current sensor data, the estimated total roll angle, and a non-linear vehicle model.

18. The method (800) according to claim 16 or 17, characterized in that, Further comprising: Control the slip of the vehicle currently moving on the road surface based on the estimated slip angle.

19. The method (800) according to any one of claims 11 to 18, characterized in that, The sensor data includes the yaw angular velocity of the vehicle currently moving on the road surface, and the method further includes controlling the yaw angular velocity of the vehicle currently moving on the road surface based on the yaw angular velocity.

20. A computer program product, characterized in that, Comprising computer-readable instructions for performing the steps of the method (800) according to any one of claims 11 to 19 when run on a computer.

Images