Vehicle control system

Abstract

To improve the robustness of the learning process and achieve more optimal vehicle control. [Solution] The vehicle control device according to the embodiment includes: a forward data output unit that acquires data relating to road surface vibrations in front of the vehicle while it is in motion; a vehicle behavior data processing unit that acquires vehicle behavior data relating to the behavior of the vehicle while it is in motion; and a vehicle control unit that pre-learns a control parameter learning model that takes the road surface vibration data as input and outputs control parameters for an actuator that controls the behavior of the vehicle, and controls the actuator based on the control parameters output from the learned control parameter learning model and the vehicle behavior data.

Description

Technical Field

[0001] The present invention relates to a vehicle control device.

Background Art

[0002] Conventionally, there has been known a vehicle control device that performs damping force control or the like by feeding back vehicle behavior detected in a traveling vehicle. For example, there is a technique of receiving feedback data of vehicle behavior detected and measured during vehicle travel, applying arithmetic processing specified by execution of a machine learning algorithm to the feedback data, controlling damper characteristics, and updating control variables used internally (see, for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in such conventional techniques, the feedback data of vehicle behavior is data based on information on the road surface after the vehicle has already passed. That is, in the conventional technique, such past road surface data is input and learning processing is performed. In other words, only the displacement pattern of the road surface is learned. Therefore, in the conventional technique, when the vehicle travels on a road surface different from that during learning, it is difficult to output an optimal value, and the robustness of vehicle control is reduced.

[0005] The present invention has been made in view of the above, and a main object thereof is to provide a vehicle control device that can improve the robustness of learning and realize more optimal vehicle control.

Means for Solving the Problems

[0006] The vehicle control device according to the present invention is a vehicle control device mounted on a vehicle, comprising: a forward data output unit that acquires data relating to road surface vibrations in front of the vehicle while it is in motion; a vehicle behavior data processing unit that acquires vehicle behavior data relating to the behavior of the vehicle while it is in motion; and a vehicle control unit that pre-trains a control parameter learning model that takes the road surface vibration data as input and outputs control parameters for an actuator that controls the behavior of the vehicle, and controls the actuator based on the control parameters output from the trained control parameter learning model and the vehicle behavior data. [Effects of the Invention]

[0007] The vehicle control device according to the present invention improves the robustness of learning and enables more optimal vehicle control. [Brief explanation of the drawing]

[0008] [Figure 1] Figure 1 is an exemplary perspective view showing a transparent view of a portion of the passenger compartment of a vehicle according to the first embodiment. [Figure 2] Figure 2 is a plan view showing the vehicle body of the first embodiment as seen through. [Figure 3] Figure 3 is an exemplary block diagram of the configuration of a vehicle control system in a vehicle according to the first embodiment. [Figure 4] Figure 4 is a functional block diagram of a vehicle control device according to the first embodiment. [Figure 5] Figure 5 is a flowchart showing an example of the procedure for vehicle control processing according to the first embodiment. [Figure 6] Figure 6 is a diagram illustrating the learning-based vehicle control (during learning) in the first embodiment. [Figure 7] Figure 7 is a diagram illustrating the learning-based vehicle control in the first embodiment (when the road surface pattern is the same as during learning). [Figure 8]Figure 8 is a diagram illustrating the learning-based vehicle control in the first embodiment (in the case of a different road surface pattern than that used during learning). [Figure 9] Figure 9 is a block diagram showing an example of the functional configuration of a vehicle control device according to the second embodiment. [Figure 10] Figure 10 is a diagram illustrating the storage of vehicle behavior data in driving history data in the second embodiment. [Figure 11] Figure 11 is a flowchart showing an example of the processing procedure from acquiring position information to outputting to a control parameter learning model in the vehicle control processing according to the second embodiment. [Figure 12] Figure 12 is a flowchart showing an example of the learning process procedure in the vehicle control process according to the second embodiment. [Figure 13] Figure 13 is a diagram illustrating the conventional learning-based vehicle control (during the learning phase). [Figure 14] Figure 14 is a diagram illustrating conventional vehicle control based on learning (in the case of the same road surface pattern as during learning). [Figure 15] Figure 15 is a diagram illustrating conventional learning-based vehicle control (in the case of a different road surface pattern than the one used during learning). [Modes for carrying out the invention]

[0009] Illustrative embodiments of the present invention are disclosed below. The configurations of the embodiments shown below, as well as the actions, results, and effects brought about by such configurations, are examples only. The present invention can be realized by configurations other than those disclosed in the following embodiments, and it is possible to obtain at least one of the various effects based on the basic configuration or derived effects.

[0010] The vehicle 1 of the present embodiment may be, for example, an automobile having an internal combustion engine (not shown) as a drive source, that is, an internal combustion engine vehicle, or an automobile having an electric motor (not shown) as a drive source, that is, an electric vehicle, a fuel cell vehicle, etc., or a hybrid vehicle having both of them as drive sources, or an automobile having other drive sources. Further, the vehicle 1 can be equipped with various transmission devices, and can be equipped with various devices necessary for driving the internal combustion engine and the electric motor, such as systems and components. In addition, the type, number, layout, etc. of the devices related to the drive of the wheels 3 in the vehicle 1 can be set in various ways.

[0011] (Configuration of Vehicle, Vehicle Control System) FIG. 1 is an exemplary perspective view showing a state in which a part of the passenger compartment of the vehicle according to the embodiment is seen through. FIG. 2 is a plan view showing a state in which the vehicle body of the vehicle according to the embodiment is seen through. FIG. 3 is an exemplary block diagram of the configuration of a vehicle control system included in the vehicle according to the embodiment.

[0012] First, an example of the configuration of the vehicle 1 according to the present embodiment will be described with reference to FIGS. 1 to 3. As illustrated in FIG. 1, the vehicle body 2 constitutes a passenger compartment 2a in which a passenger (not shown) rides. Inside the passenger compartment 2a, a steering unit 4, an acceleration operation unit 5, a braking operation unit 6, a shift operation unit 7, etc. are provided in a state facing the driver's seat 2b as a passenger.

[0013] The steering unit 4 is, for example, a steering wheel protruding from the dashboard 24. The acceleration operation unit 5 is, for example, an accelerator pedal located under the driver's feet. The braking operation unit 6 is, for example, a brake pedal located under the driver's feet. The shift operation unit 7 is, for example, a shift lever protruding from the center console. Note that the steering unit 4, the acceleration operation unit 5, the braking operation unit 6, the shift operation unit 7, etc. are not limited to these.

[0014] In addition, in the passenger compartment 2a, a display device 8 as a display output unit and an audio output device 9 as an audio output unit are provided. The display device 8 is, for example, an LCD (Liquid Crystal Display), an OELD (Organic Electroluminescent Display), or the like. The audio output device 9 is, for example, a speaker. Further, the display device 8 is covered with a transparent operation input unit 10 such as a touch panel. The passenger can visually recognize the image displayed on the display screen of the display device 8 through the operation input unit 10. Also, the passenger can execute an operation input by touching, pressing, or moving the operation input unit 10 with a finger or the like at a position corresponding to the image displayed on the display screen of the display device 8.

[0015] These display device 8, audio output device 9, operation input unit 10, etc. are provided, for example, in a monitor device 11 located at the center in the vehicle width direction, that is, the left - right direction, of the dashboard 24. The monitor device 11 can have operation input units (not shown) such as switches, dials, joysticks, push buttons, etc. Also, an audio output device (not shown) can be provided at another position in the passenger compartment 2a different from the monitor device 11, and audio can be output from the audio output device 9 of the monitor device 11 and the other audio output device. Note that the monitor device 11 can be used in combination with, for example, a navigation system or an audio system. Also, in the passenger compartment 2a, a display device 12 different from the display device 8 is provided.

[0016] Also, as illustrated in FIG. 2, the vehicle 1 is, for example, a four - wheel vehicle and has two left - right front wheels 3F and two left - right rear wheels 3R. Any of these four wheels 3 can be configured to be steerable. As illustrated in FIG. 3, the vehicle 1 has a steering system 13 that steers at least two wheels 3.

[0017] As illustrated in Figure 3, the steering system 13 includes actuators 101 and 104 and a torque sensor 13b. The steering system 13 is electrically controlled by an ECU 14 (Electronic Control Unit) or the like to operate the actuators 101 and 104. Here, as shown in Figure 2, actuator 101 is connected to the front wheel 3F and is a front steering actuator for steering the front wheel 3F. Actuator 104 is connected to the rear wheel 3R and is a rear steering actuator for steering the rear wheel 3R.

[0018] The steering system 13 is, for example, an electric power steering system or an SBW (Steer By Wire) system. The steering system 13 supplements the steering force by adding torque, i.e., assist torque, to the steering unit 4 using actuators 101 and 104, or by steering the wheels 3 using actuators 101 and 104. In this case, actuators 101 and 104 may steer one wheel 3 or multiple wheels 3. The torque sensor 13b detects, for example, the torque that the driver applies to the steering unit 4.

[0019] Furthermore, as illustrated in Figure 3, the vehicle body 2 is equipped with multiple imaging units 15, for example, eight imaging units 15a to 15f. Each imaging unit 15 is a digital camera incorporating an image sensor such as a CCD (Charge Coupled Device) or CIS (CMOS Image Sensor). Each imaging unit 15 can output video data at a predetermined frame rate. Each imaging unit 15 has either a wide-angle lens or a fisheye lens and can capture a range of, for example, 140° to 190° in the horizontal direction. The optical axis of each imaging unit 15 is set to point diagonally downward. Therefore, the imaging unit 15 sequentially captures the external environment around the vehicle body 2, including the road surface on which the vehicle 1 can move and the area where the vehicle 1 can park, and outputs it as captured images (image data).

[0020] The imaging unit 15a is located, for example, at the rear end 2e of the vehicle body 2 and is provided on the lower wall of the rear trunk door 2h. The imaging unit 15ba is located, for example, at the right side of the vehicle body 2, i.e., at the right end in the vehicle width direction, and is provided on the front side of the door mirror 2g, which is a right-side projection. The imaging unit 15bb is located, for example, at the right side of the vehicle body 2, i.e., at the right end in the vehicle width direction, and is provided on the rear side of the door mirror 2g, which is a right-side projection.

[0021] The imaging unit 15c is located, for example, at the front of the vehicle body 2, that is, at the front end in the vehicle's longitudinal direction, and is provided on the front bumper or the like. The imaging unit 15da is located, for example, at the left side of the vehicle body 2, that is, at the left end in the vehicle's width direction, and is provided on the front side of the door mirror 2g, which is a protruding part on the left side. The imaging unit 15db is located, for example, at the left side of the vehicle body 2, that is, at the left end in the vehicle's width direction, and is provided on the rear side of the door mirror 2g, which is a protruding part on the left side.

[0022] The imaging unit 15e is located, for example, on the right side of the vehicle body 2, that is, at the right end in the vehicle width direction, and is provided near the right-side door. The imaging unit 15f is located, for example, on the left side of the vehicle body 2, that is, at the left end in the vehicle width direction, and is provided near the left-side door.

[0023] The ECU 14 performs calculations and image processing based on image data obtained from multiple imaging units 15, enabling it to generate images with a wider field of view or a virtual overhead view image of the vehicle 1 as seen from above. The overhead view image can also be referred to as a planar image. In this embodiment, each imaging unit 15 captures an image of the area around the vehicle 1.

[0024] Furthermore, as illustrated in Figure 1, the vehicle body 2 is equipped with multiple distance measuring units 16 and 17, such as four distance measuring units 16a to 16d and eight distance measuring units 17a to 17h. The distance measuring units 16 and 17 are, for example, sonars that emit ultrasonic waves and capture the reflected waves. Sonars can also be called sonar sensors or ultrasonic detectors. The ECU 14 can measure the presence or absence of objects such as obstacles located around the vehicle 1 and the distance to such objects based on the detection results of the distance measuring units 16 and 17. In other words, the distance measuring units 16 and 17 are examples of detection units that detect objects. The distance measuring unit 17 can be used, for example, to detect objects at a relatively short distance, while the distance measuring unit 16 can be used, for example, to detect objects at a relatively long distance that is further away than the distance measuring unit 17. In addition, the distance measuring unit 17 can be used, for example, to detect objects in front of and behind the vehicle 1, while the distance measuring unit 16 can be used to detect objects to the sides of the vehicle 1.

[0025] Furthermore, as illustrated in Figure 3, the vehicle control system 100 includes the ECU 14, monitoring device 11, steering system 13, distance measuring units 16 and 17, as well as the brake system 18, suspension system 30, steering angle sensor 19, accelerator sensor 20, shift sensor 21, wheel speed sensor 22, acceleration sensor 25, vehicle height sensor 26, spring sensor 27, actuator 107, etc., all of which are electrically connected via the in-vehicle network 23, which serves as an telecommunications line. The in-vehicle network 23 is configured, for example, as a CAN (controller area network). The ECU 14 is an example of an estimation device and vehicle control device.

[0026] The ECU 14 can control the steering system 13, brake system 18, suspension system 30, actuator 107, etc., by sending control signals via the in-vehicle network 23. The ECU 14 can also receive detection results from the torque sensor 13b, brake sensor 18b, steering angle sensor 19, distance measuring unit 16, distance measuring unit 17, accelerator sensor 20, shift sensor 21, wheel speed sensor 22, acceleration sensor 25, vehicle height sensor 26, etc., as well as operation signals from the operation input unit 10, etc., via the in-vehicle network 23.

[0027] The ECU14 includes, for example, a CPU14a (Central Processing Unit), ROM (Read Only Memory)14b, RAM (Random Access Memory)14c, display control unit14d, audio control unit14e, SSD (Solid State Drive, flash memory)14f, and the like.

[0028] The CPU 14a can perform various calculations and controls, such as image processing related to images displayed on the display devices 8 and 12, determining the target position of the vehicle 1, calculating the movement path of the vehicle 1, determining whether or not there is interference with an object, automatic control of the vehicle 1, release of automatic control, damping control, spring constant switching control, steering control, stabilizer control, and driving force control. The CPU 14a can read programs installed and stored in a non-volatile storage device such as the ROM 14b and execute calculations according to those programs.

[0029] RAM 14c temporarily stores various data used in calculations performed by CPU 14a. The display control unit 14d primarily performs image processing using image data obtained by the imaging unit 15 and image data synthesis displayed on the display device 8, among the calculation processes performed by ECU 14. The audio control unit 14e primarily processes audio data output by the audio output device 9, among the calculation processes performed by ECU 14. SSD 14f is a rewritable non-volatile storage unit that can store data even when the power to ECU 14 is turned off. Note that CPU 14a, ROM 14b, RAM 14c, etc., can be integrated within the same package. Also, ECU 14 may use other logic processors such as DSP (Digital Signal Processor) or logic circuits instead of CPU 14a. Furthermore, HDD (Hard Disk Drive) may be provided instead of SSD 14f, and SSD 14f and HDD may be provided separately from ECU 14.

[0030] The location information sensor 31 is a sensor that acquires the current position of vehicle 1, and is, for example, a GPS receiver. The location information sensor 31 sends the acquired current position as location information to the CPU 14a.

[0031] The brake system 18 includes, for example, an ABS (Anti-lock Brake System) to suppress brake lock, an Electronic Stability Control (ESC) to suppress skidding of the vehicle 1 during cornering, an electric brake system to enhance braking force (perform brake assist), and a BBW (Brake By Wire). The brake system 18 applies braking force to the wheels 3 and thus to the vehicle 1 via an actuator 18a. The brake system 18 can also detect signs of brake lock, wheel spinning, skidding, etc., from the rotational difference between the left and right wheels 3, and perform various controls. The brake sensor 18b is, for example, a sensor that detects the position of the movable part of the braking operation unit 6. The brake sensor 18b can detect the position of the brake pedal as a movable part. The brake sensor 18b includes a displacement sensor.

[0032] Furthermore, the actuator 107 is a rear drive force control actuator as shown in Figure 2, and is electrically controlled by the ECU 14, etc., to control the drive force of the rear wheel 3R.

[0033] The steering angle sensor 19 is a sensor that detects the amount of steering of the steering unit 4, such as a steering wheel. The steering angle sensor 19 is constructed using, for example, a Hall element. The ECU 14 obtains the amount of steering of the steering unit 4 by the driver, the amount of steering of each wheel 3 during automatic steering, etc. from the steering angle sensor 19 and performs various controls. The steering angle sensor 19 detects the rotation angle of the rotating part included in the steering unit 4.

[0034] The accelerator sensor 20 is, for example, a sensor that detects the position of a movable part of the acceleration control unit 5. The accelerator sensor 20 can detect the position of the accelerator pedal as a movable part. The accelerator sensor 20 includes a displacement sensor.

[0035] The shift sensor 21 is, for example, a sensor that detects the position of a movable part of the gear shift operation unit 7. The shift sensor 21 can detect the position of a movable part such as a lever, arm, or button. The shift sensor 21 may include a displacement sensor or may be configured as a switch.

[0036] The wheel speed sensor 22 is a sensor that detects the amount of rotation of the wheel 3 and the number of rotations per unit time. The wheel speed sensor 22 outputs the number of wheel speed pulses indicating the detected rotation speed as a sensor value. The wheel speed sensor 22 may be constructed using, for example, a Hall element. The ECU 14 calculates the amount of movement of the vehicle 1 based on the sensor value obtained from the wheel speed sensor 22 and executes various controls. In some cases, the wheel speed sensor 22 may be provided in the brake system 18. In that case, the ECU 14 obtains the detection result of the wheel speed sensor 22 via the brake system 18. The vehicle height sensor 26 is a sensor that detects the stroke displacement of each wheel.

[0037] The suspension system 30 is positioned between the vehicle body 2 and the wheels 3 of the vehicle 1. The suspension system 30 includes a spring that absorbs vibrations of the vehicle 1 caused by impacts from the road surface, and a variable damping damper that dampens the vibrations of the spring and can change the damping force of the spring's vibrations. In this embodiment, the suspension system 30 works in cooperation with the ECU 14 to control a damping force adjustment device such as a solenoid actuator to change the damping force of the variable damping damper. In this way, the suspension system 30 realizes a damping force control system that dampens vibrations of the vehicle body in the vertical, lateral, and longitudinal directions caused by impacts from the road surface.

[0038] Acceleration sensors 25 are provided near the vehicle body 2 side (also referred to as the sprung mass) relative to the suspension system 30 and near the vehicle 1 side (also referred to as the unsprung mass) relative to the suspension system 30. The acceleration sensors 25 include vertical acceleration sensors (i.e., sprung mass G sensor and unsprung mass G sensor) that detect and output the vertical acceleration of the vehicle body 2, a longitudinal acceleration sensor that detects and outputs the longitudinal acceleration of the vehicle body 2 (vehicle 1), and a lateral acceleration sensor that detects and outputs the lateral acceleration, which is the lateral (widthwise) acceleration of the vehicle body 2 (vehicle 1).

[0039] Here, the vertical acceleration sensors consist of a sprung-load acceleration sensor (also called a sprung-load G sensor) that detects sprung-load acceleration (also called sprung-load G), and an unsprung-load acceleration sensor (also called an unsprung-load G sensor) that detects unsprung-load acceleration (also called unsprung-load G).

[0040] The suspension system 30 is electrically controlled by the ECU 14, etc., to operate actuators 102, 103, 105, and 106 for damping force control and stabilizer control.

[0041] Here, actuator 102 is a front active stabilizer actuator for controlling the stabilizer on the front wheel 3F side, as shown in Figure 2, and is provided on the suspension on the rear wheel 3R side, as shown in Figure 2, and is a rear active stabilizer actuator for controlling the stabilizer on the rear wheel 3R side.

[0042] Furthermore, as shown in Figure 2, actuator 103 is provided on the suspension on the front wheel 3F side and is a front damping force control actuator for controlling the damping force on the front wheel 3F side. Actuator 106 is provided on the suspension on the rear wheel 3R side and is a rear damping force control actuator for controlling the damping force on the rear wheel 3R side.

[0043] The configurations, arrangements, and electrical connection types of the various sensors and actuators described above are merely examples and can be configured (changed) in various ways.

[0044] (Configuration of the vehicle control system) Next, we will describe the vehicle control system in which the ECU14 functions. Hereafter, the ECU14 may be referred to as the vehicle control system 14.

[0045] Figure 4 is a functional block diagram of a vehicle control device 14 according to an embodiment. As shown in Figure 4, the vehicle control device 14 mainly comprises, as an example, a vehicle behavior data processing unit 1401, a vehicle control unit 1402, a learning inference switching unit 1403, a reward calculation unit 1404, a learning unit 1405, and a forward data output unit 1406 as functional units.

[0046] These functional units are realized by the CPU 14a reading and executing a program stored in the memory unit 50d. Here, the memory unit 50d includes ROM 14b, RAM 14c, and SSD 14f. That is, the program may, for example, include modules corresponding to each block of the vehicle control device 14 shown in Figure 4. It is also possible to realize each functional unit module using independent hardware such as circuits including ASICs (Applocation Specific Integrated Circuits).

[0047] The vehicle behavior data processing unit 1401 acquires detection signals (i.e., sensor data) of the vehicle 1's behavior detected by the vehicle behavior sensor, and calculates vehicle behavior data using the sensor data itself or based on the sensor data. Furthermore, the vehicle behavior data processing unit 1401 obtains vehicle behavior data of the moving vehicle 1 by controlling actuators 102, 103, 105, and 106 based on the control gain output from the learning unit 1405, which will be described later.

[0048] Here, vehicle behavior data refers to data relating to the behavior of vehicle 1, such as sprung mass acceleration (sprung mass G), vehicle height, vehicle speed, sprung mass velocity, vehicle height change velocity, and unsprung mass velocity. The vehicle behavior sensors include the sprung mass G sensor within the acceleration sensor 25, the wheel speed sensor 22, the vehicle height sensor 26, etc.

[0049] For example, the vehicle behavior data processing unit 1401 receives the sprung mass acceleration (sprung mass G), which is the detection signal from the sprung mass G sensor in the acceleration sensor 25, and outputs the sprung mass acceleration as vehicle behavior data. The vehicle behavior data processing unit 1401 receives the wheel speed, which is the detection signal from the wheel speed sensor 22, calculates the vehicle speed, which is the speed of vehicle 1, from the wheel speed, and outputs the calculated vehicle speed as vehicle behavior data. The vehicle behavior data processing unit 1401 also receives the vehicle height, which is the detection signal from the vehicle height sensor 26, and outputs the vehicle height as vehicle behavior data. Furthermore, the vehicle behavior data processing unit 1401 integrates the sprung mass acceleration detected by the sprung mass G sensor to obtain the sprung mass velocity, and outputs the sprung mass velocity as vehicle behavior data. The vehicle behavior data processing unit 1401 differentiates the vehicle height sensor data detected by the vehicle height sensor to obtain the vehicle height change rate, and outputs the vehicle height change rate as vehicle behavior data. The vehicle behavior data processing unit 1401 calculates the unsprung speed from the difference between the sprung speed and the vehicle height change speed, and outputs the unsprung speed as vehicle behavior data.

[0050] Specifically, the vehicle behavior data processing unit 1401 outputs vehicle behavior data such as sprung mass acceleration (sprung mass G), vehicle height, vehicle speed, sprung mass velocity, vehicle height change rate, and unsprung mass velocity to the vehicle control unit 1402 as feedback values ​​for vehicle behavior data (hereinafter referred to as "vehicle behavior FB data"). The vehicle behavior data processing unit 1401 also outputs sprung mass G to the reward calculation unit 1404. The vehicle behavior data processing unit 1401 outputs vehicle speed to the forward data output unit 1406.

[0051] The learning-inference switching unit 1403 outputs a switching instruction to the learning unit 1405 to switch between learning and inference processing (i.e., an instruction to learn or inference).

[0052] The reward calculation unit 1404 receives the sprung mass acceleration from the vehicle behavior data processing unit 1401, calculates a reward value using a predetermined reward function, and outputs the calculated reward value to the learning unit 1405. In other words, the reward calculation unit 1404 obtains the sprung mass acceleration of the vehicle 1, which is driven by controlling actuators 102, 103, 105, and 106 based on the control gain output from the learning unit 1405, from the vehicle behavior data processing unit 1401, and calculates a reward value.

[0053] The forward data output unit 1406 receives an image of the area in front of the vehicle 1 from the imaging unit 15, specifically the imaging unit 15c which images the area in front of the vehicle 1, and also receives the vehicle speed from the vehicle behavior data processing unit 1401. The forward data output unit 1406 then calculates the road surface displacement relative to distance from the image. The forward data output unit 1406 further outputs the vibration frequency characteristics of the road surface displacement (road surface displacement relative to the time axis, or road surface displacement spectrum relative to the frequency axis) that are expected to occur when the vehicle is assumed to travel along the road surface in front at the current speed, based on the displacement of the road surface in front and the current vehicle speed, as forward road surface displacement data to the control parameter learning model 1411. Alternatively, the road surface displacement data of the left and right wheel positions may be converted into heave, roll, and pitch input components for the vehicle and output to the control parameter learning model 1411 as road surface displacement data. Here, road surface displacement data is an example of data related to road surface vibration.

[0054] The learning unit 1405 has a control parameter learning model 1411. The control parameter learning model 1411 is a trained model that receives road surface displacement data information from the forward data output unit 1406 and outputs a control gain. The control parameter learning model 1411 is composed of, for example, a neural network.

[0055] The control gain is a parameter that forms the basis of the control current that the suspension system 30 commands to actuators 102, 103, 105, and 106 for damping force control. The control gain is an example of a control parameter.

[0056] The control parameter learning model 1411 is shipped in a state where it has been pre-trained by machine learning based on road surface displacement data and reward values ​​obtained by actually driving the vehicle 1, or based on road surface displacement data and reward values ​​obtained by machine learning based on vehicle simulations that mimic the vehicle 1, before the release of the vehicle control device 14. After the release of the vehicle control device 14, it is further sequentially trained by the learning unit 1405 using reinforcement learning methods.

[0057] The learning unit 1405 learns the control parameter learning model 1141 based on the road surface displacement data ahead. Specifically, it takes the control gain output from the control parameter learning model 1411 and the reward value output from the reward calculation unit 1404 as input and performs reinforcement learning on the control parameter learning model 1411 so that the reward value is maximized. In other words, the learning unit 1405 performs reinforcement learning on the control parameter learning model 1411 by updating parameters such as the weight values ​​of the neural network that constitutes the control parameter learning model 1411 so that the reward value is maximized.

[0058] The vehicle control unit 1402 receives the vehicle behavior FB value (vehicle behavior FB data) output by the vehicle behavior data processing unit 1401 and the control gain output by the learning unit 1405. The vehicle control unit 1402 then calculates a target control amount based on the input vehicle behavior FB value and control gain, and outputs the calculated target control amount to the suspension system 30. As a result, the suspension system 30 commands the actuators 102, 103, 105, and 106 to control current to reach the target control amount and performs damping force control, etc.

[0059] (Vehicle control processing) Next, the vehicle control process by the vehicle control device 14 according to this embodiment, which is configured as described above, will be explained. Figure 5 is a flowchart showing an example of the procedure for vehicle control processing according to the first embodiment. The forward data output unit 1406 receives vehicle speed from the vehicle behavior data processing unit 1401 and forward-facing images from the imaging unit 15c, acquires road surface displacement data, and outputs it to the control parameter learning model 1411 (S101). As a result, the control parameter learning model 1411 outputs the control gain.

[0060] Next, the learning / inference switching unit 1403 determines whether to perform inference or learning (S102). Whether to perform inference or learning can be instructed, for example, by the driver, while the vehicle 1 is in motion.

[0061] If inference is determined (S102: inference), the learning unit 1405 enters inference mode and outputs the control gain from the control parameter learning model 1411 to the vehicle control unit 1402 without the search component (S103). As a result, the vehicle control unit 1402 calculates the target control amount using the control gain and vehicle behavior FB data and outputs it to the suspension system 30, thereby performing damping force control, etc.

[0062] On the other hand, if learning is determined in S102 (S102: Learning), the learning unit 1405 enters learning mode and outputs the control gain output from the control parameter learning model 1411, with a search component, to the vehicle control unit 1402 (S104). As a result, the vehicle control unit 1402 calculates the target control amount using the control gain and vehicle behavior FB data and outputs it to the suspension system 30 to perform damping force control, etc., while the vehicle behavior data processing unit 1401 acquires vehicle behavior data under such control (S105).

[0063] Next, the reward calculation unit 1404 calculates a reward value from the sprung mass acceleration, which is one of the vehicle behavior data acquired by the vehicle behavior data processing unit 1401 (S106). Then, the learning unit 1405 updates the parameters of the control parameter learning model 1411 so that the reward value is maximized (S107). As a result, the control gain output from the control parameter learning model 1411 with updated parameters is output to the vehicle control unit 1402, and the vehicle control unit 1402 calculates a target control amount using the control gain and the vehicle behavior FB data and outputs it to the suspension system 30, thereby performing damping force control, etc.

[0064] (Overview) Figures 13-15 illustrate conventional learning-based vehicle control. Figure 13 shows the case where vehicle 1 travels on road surfaces A, B, and C in that order during learning. Figure 14 shows the case where vehicle 1 travels on the same road surface pattern (road surfaces A, B, and C in that order) after learning, i.e., during inference. Figure 15 shows the case where vehicle 1 travels on a different road surface pattern (road surfaces B, A, and C in that order) during inference. In the examples in Figures 13-15, the road surface vibration is shown as the input to the control parameter learning model, and the control gain and reward value are shown as the output of the control parameter learning model. The reward value is indicated as High or Low.

[0065] Conventional technology involved receiving feedback data of vehicle behavior detected and measured while the vehicle was in motion, applying computational processing identified by a machine learning algorithm to the feedback data, controlling the damper characteristics, and updating the internally used control variables.

[0066] However, in this conventional technology, as shown in Figure 13, the control parameter learning model takes feedback data regarding road surface vibrations on the road surface the vehicle has passed over as input, outputs control parameters, and obtains reward values ​​for the output control parameters on subsequent road surfaces. Therefore, the road surface on which the input data is based and the road surface on which the reward values ​​are based are different, and the only causal relationship between these input / output and reward value datasets is the road surface pattern (occurrence order). In other words, the model learns the relationship of road surface patterns (occurrence order), for example, that it is optimal to output control gain b after road surface vibration A. Therefore, as shown in Figure 14, when a vehicle travels on a road surface with the same road surface pattern (in the order of road surfaces A, B, C) as the road surface used during learning shown in Figure 13, the control parameter learning model can output control gains a, b, and c, which are appropriate control parameters for road surfaces A, B, and C respectively, and perform appropriate vehicle control. Here, control gain a is assumed to be the optimal control gain for road surface A, control gain b is the optimal control gain for road surface B, and control gain c is the optimal control gain for road surface C.

[0067] However, for example, when a vehicle travels on a road surface with a different pattern than the road surface pattern shown in Figure 13, such as the one shown in Figure 15 (i.e., road surfaces B, A, C in that order), the conventional control parameter learning model has difficulty outputting optimal control parameters such as control gains, resulting in a reduction in the robustness of vehicle control.

[0068] In contrast, the vehicle control device 14 according to this embodiment includes a forward data output unit 1406 that acquires road surface displacement data in front of the moving vehicle 1, a vehicle behavior data processing unit 1401 that acquires vehicle behavior data relating to the behavior of the moving vehicle 1, and a vehicle control unit 1402 that pre-learns a control parameter learning model 1411 that takes forward road surface displacement data as input and outputs control gains for actuators 102, 103, 105, and 106, and controls actuators 102, 103, 105, and 106 based on the control gains output from the learned control parameter learning model 1411 and vehicle behavior FB data.

[0069] Therefore, in this embodiment, the control parameter learning model 1411 outputs a control gain using the displacement of the road surface ahead of where the vehicle 1 will travel, rather than the road surface the vehicle has just passed, and obtains a reward value after actually traveling on that road surface ahead with the output control gain. Thus, the road surface on which the input data is based and the road surface on which the reward value is based coincide, and since the control parameter learning model 1411 is trained with this input / output and reward value dataset, it is possible to learn the causal relationship between the characteristics of the road surface and the control gain that is optimal for that road surface. Accordingly, according to this embodiment, by using the road surface displacement data ahead, it is not only possible to eliminate the phase lag, but it is also possible to output the optimal control gain even when the vehicle 1 travels on a different road surface than the one used during training. Thus, according to this embodiment, highly robust, or in other words, highly generalizable, learning can be performed, and optimal vehicle control can be achieved.

[0070] Figures 6-8 illustrate the learning-based vehicle control in the first embodiment. Figure 6 shows the case where vehicle 1 travels on road surfaces A, B, and C in that order during learning. Figure 7 shows the case where vehicle 1 travels on the same road surface pattern (road surfaces A, B, and C in that order) as during learning, i.e., during inference. Figure 8 shows the case where vehicle 1 travels on a different road surface pattern (road surfaces B, A, and C in that order) during inference than during learning. In addition, the examples in Figures 6-8 show the road surface displacement (an example of road surface vibration) as the input to the control parameter learning model 1411, and the control gain and reward value as the output of the control parameter learning model 1411.

[0071] In the vehicle control device 14 according to this embodiment, as shown in Figure 6, the control parameter learning model 1411 takes the road surface displacement that the vehicle 1 will travel on (for example, road surface displacement B on road surface B) as input, outputs a control gain (for example, control gain b), obtains a reward value after actually traveling on the road surface ahead (for example, road surface B) with the outputted control gain, and performs learning with this set of learning data (for example, input: road surface displacement B, output: control gain b, reward value: reward value obtained by traveling on road surface B with control gain b). Therefore, as shown in Figure 7, when traveling on road surfaces with the same road surface pattern as during learning, such as road surfaces A, B, and C, the control parameter learning model 1411 can naturally output the optimal control gain for each road surface. Furthermore, in this embodiment, as shown in Figure 8, even when traveling on road surfaces with different road surface patterns than during learning, such as road surfaces B, A, and C, the control parameter learning model 1411 can output the optimal control gain for each road surface. Furthermore, even when driving on road surface A', which has a different road surface profile but the same characteristics as road surface A (for example, the same vibration frequency components and heave, roll, and pitch input components to the vehicle), the optimal control gain a can be output in the same way.

[0072] Furthermore, in the vehicle control device 14 according to this embodiment, the vehicle behavior data processing unit 1401 further acquires vehicle behavior data of the moving vehicle 1 by controlling actuators 102, 103, 105, and 106 based on control gains. The vehicle control device 14 also includes a reward calculation unit 1404 that calculates a reward value from the acquired vehicle behavior data using a predetermined reward function, and a learning unit 1405 performs reinforcement learning on the control parameter learning model 1411 so as to maximize the reward value. Therefore, according to this embodiment, since the control parameter learning model 1411 is trained by reinforcement learning to learn the displacement of the road surface ahead of where the vehicle 1 will travel, more robust learning can be performed, and more optimal vehicle control can be achieved.

[0073] Furthermore, in the vehicle control device 14 according to this embodiment, the forward data output unit 1406 determines the condition of the road surface in front of the vehicle 1 based on the image captured by the imaging unit 15c, which is capable of capturing images in front of the vehicle 1, and determines road surface displacement data based on the road surface condition and the vehicle speed, which is the speed of the vehicle included in the vehicle behavior data. Therefore, according to this embodiment, the displacement of the road surface that the vehicle will actually travel on is determined and used to train the control parameter learning model 1411, so that more robust learning can be performed and more optimal vehicle control can be achieved.

[0074] Furthermore, in the vehicle control device 14 according to this embodiment, the control parameter learning model 1411 is pre-learned before the vehicle is released, using road surface displacement data, the control parameters output using that data as input, and the reward values ​​obtained when driving on that road surface. It is further learned in the user's operating environment. In the conventional technology, it is not possible to output the optimal control parameters on unlearned road surfaces that differ from the road surface pattern used during learning, so the results learned in advance before the vehicle is released cannot be effectively utilized on the road surface that the user actually drives on. However, in this embodiment, it is possible to output the optimal control parameters even on unlearned road surfaces that differ from the road surface pattern used during learning, so the results learned in advance can be utilized on the road surface that the user actually drives on. Therefore, according to this embodiment, when a user drives the vehicle 1, they can drive with the optimal control gain (control parameters) from the beginning, and learning can be performed in accordance with the user's environment.

[0075] (Second embodiment) In the first embodiment, road surface displacement data, which is data relating to road surface vibrations ahead, was acquired based on images captured by the imaging unit 15. In this second embodiment, however, data relating to road surface vibrations ahead is acquired based on the current position of the vehicle 1 and its past driving history.

[0076] (Configuration of the vehicle control device 514) The configuration of the vehicle 1 and the vehicle control system according to the second embodiment are the same as those in the first embodiment. Figure 9 is a block diagram showing an example of the functional configuration of the vehicle control device 514 according to the second embodiment. As shown in Figure 9, the vehicle control device 514 according to this embodiment mainly comprises, as an example, a vehicle behavior data processing unit 1401, a vehicle control unit 1402, a learning inference switching unit 1403, a reward calculation unit 1404, a learning unit 1405, a location information acquisition unit 1501, a driving history determination unit 1502, a memory determination unit 1503, a memory processing unit 1504, a forward data output unit 1505, and a memory unit 1510 as functional units.

[0077] In this embodiment as well, these functional units are realized by the CPU 14a reading and executing a program stored in the memory unit 50d. The program may, for example, include modules corresponding to each block in the vehicle control device 514 shown in Figure 9, excluding the memory unit 1510. Furthermore, each functional unit module can also be realized by independent hardware such as circuits including ASICs.

[0078] Here, the vehicle behavior data processing unit 1401, the vehicle control unit 1402, the learning inference switching unit 1403, and the reward calculation unit 1404 are the same as in the first embodiment.

[0079] The memory unit 1510 is a storage medium such as an HDD or SSD. The memory unit 1510 stores the driving history data 1511. In this embodiment, the driving history data 1511 associates location information, vehicle speed, and road surface vibration data. When vehicle 1 travels on a road surface at a location where there is no previous driving history, the driving history data 1511 stores the location information of that road surface and the road surface vibration data in association with each other. The location information registered is the location coordinates.

[0080] Here, the road surface vibration data refers to vibration data generated in the vehicle 1 when it is traveling on the road surface. In this embodiment, the road surface vibration data includes vertical displacement, velocity, and acceleration generated in the vehicle 1 when it is traveling on the road surface.

[0081] The location information acquisition unit 1501 acquires the position coordinates of the vehicle 1's current position, as detected by the location information sensor 31, as location information. The location information sensor 31 and the location information acquisition unit 1501 are, for example, a navigation system.

[0082] The driving history determination unit 1502 determines whether or not the driving history data 1511 exists in the storage unit 1510. The memory determination unit 1503 determines whether the current sprung mass acceleration exceeds a second threshold when the driving history determination unit 1502 determines that there is no driving history. Here, the second threshold is an example of a predetermined threshold.

[0083] When vehicle 1 travels on a road surface for which no travel history data 1511 has been stored, the memory processing unit 1504 stores the travel history data 1511 in the memory unit 1510 by associating the position information, vehicle speed, and road surface vibration data at the point where the vehicle behavior data, which is the vertical acceleration of the sprung mass, exceeds a second threshold. Here, as data related to road surface vibration, the memory processing unit 1504 stores in the travel history data 1511 data obtained by frequency analysis of the displacement, velocity, and acceleration of the vertical vibrations that occurred in vehicle 1 when traveling on the road surface, which can be calculated from the vehicle behavior data. In particular, by storing the unsprung mass velocity (or unsprung mass acceleration), it becomes possible to accurately reflect vibration characteristics close to the characteristics of the road surface in the learning process without being affected by control parameters. As mentioned above, the unsprung mass velocity can be calculated from the difference between the sprung mass velocity, which is obtained by integrating the sprung mass acceleration (sprung mass G), and the vehicle height change velocity, which is obtained by differentiating the vehicle height sensor.

[0084] Figure 10 is a diagram illustrating the storage of vehicle behavior data in the driving history data 1511 in the second embodiment. As shown in Figure 11, when the vertical acceleration of the vehicle 1 exceeds the second threshold, the memory flag is turned on, and that section becomes the data storage section. Then, the memory processing unit 1504 stores data related to road surface vibration, such as unsprung speed, in the driving history data 1511 within the data storage section. At this time, the memory processing unit 1504 stores the unsprung speed as power spectral density (PSD) by frequency analysis of the time-series waveform in the data storage section. This makes it possible to reduce the storage capacity of the memory unit 1510 by storing the data as frequency analysis data, which is independent of the time length of the storage section.

[0085] Returning to Figure 9, when vehicle 1 is traveling at the location indicated by the location information, if the memory processing unit 1504 determines that the difference between the current vehicle speed acquired by the vehicle behavior data processing unit 1401 and the vehicle speed associated with the location information of the current travel location in the travel history data 1511 stored in the memory unit 1510 is less than the first threshold, it outputs the road surface vibration data associated with the location information of the current travel location in the travel history data 1511 to the forward data output unit 1505.

[0086] When vehicle 1 is traveling at the location indicated by the location information, if the memory processing unit 1504 determines that the difference between the current vehicle speed acquired by the vehicle behavior data processing unit 1401 and the vehicle speed associated with the location information of the current travel location in the travel history data 1511 stored in the memory unit 1510 exceeds a first threshold, the memory processing unit 1504 corrects the road surface vibration data associated with the location information of the current travel location in the travel history data 1511 based on the current vehicle speed, and outputs the corrected road surface vibration data to the forward data output unit 1505.

[0087] The forward data output unit 1505 receives location information from the location information acquisition unit 1501, vehicle speed (vehicle behavior data) from the vehicle behavior data processing unit 1401, and the judgment result from the driving history judgment unit 1502. The forward data output unit 1505 also receives road surface vibration data corresponding to the location information, which is output from the storage processing unit 1504 according to the judgment result from the storage judgment unit 1503. The forward data output unit 1505 then outputs the road surface vibration data input from the storage processing unit 1504 to the control parameter learning model 1411.

[0088] The learning unit 1405 uses the control parameter learning model 1411 to determine the control gain from the road surface vibration data corresponding to the position information output from the forward data output unit 1505. In this embodiment, the control parameter learning model 1411 takes road surface vibration data as input and outputs control parameters.

[0089] Furthermore, the learning unit 1405 learns the control parameter learning model 1411 in the memory determination unit 1503 when the vehicle 1 travels again for the second time or later at a location where the vehicle behavior data, which is the vertical acceleration of the vehicle's sprung mass, exceeds a second threshold.

[0090] (Vehicle control processing) Next, the vehicle control process by the vehicle control device 514 configured according to this embodiment will be described. First, the process from acquiring position information to outputting to the control parameter learning model 1411 in the vehicle control process will be described. Figure 11 is a flowchart showing an example of the processing procedure from acquiring position information to outputting to the control parameter learning model 1411 in the vehicle control processing according to the second embodiment.

[0091] First, the location information acquisition unit 1501 acquires location information of the vehicle 1's current location from the location information sensor 31 (S201). The location information is a position coordinate indicated by latitude and longitude, and in addition, the road ID of the road being traveled can also be acquired.

[0092] Next, the driving history determination unit 1502 determines whether or not the driving history data 1511 is stored in the storage unit 1510 (S202). If the driving history data 1511 is not stored in the storage unit 1510 (S202: No), the storage processing unit 1504 obtains the sprung mass vertical acceleration of vehicle 1 from the vehicle behavior data processing unit 1401 and determines whether or not the sprung mass vertical acceleration of vehicle 1 is greater than or equal to a second threshold (S204). If the sprung mass vertical acceleration of vehicle 1 is less than the second threshold (S204: No), the process ends.

[0093] On the other hand, if the vertical acceleration of the vehicle 1 is greater than or equal to the second threshold (S204: Yes), the vehicle behavior data is stored in the driving history data 1511, linked to the location information (S205).

[0094] In S202, if the driving history data 1511 is stored in the memory unit 1510 (S202: Yes), the memory determination unit 1503 obtains the current vehicle speed of vehicle 1 from the vehicle behavior data processing unit 1401 and obtains the previously stored driving history data 1511 in the memory unit 1510 via the memory processing unit 1504 (S206).

[0095] Next, the memory determination unit 1503 determines whether the current vehicle speed differs from the previously stored vehicle speed in the driving history data by a first threshold or more (S207). If the difference between the current vehicle speed and the previously stored vehicle speed in the driving history data is less than the first threshold (S207: No), the forward data output unit 1505 outputs the road surface vibration data stored in the driving history data 1511 to the control parameter learning model 1411 (S208). After that, the process moves on to S209.

[0096] In S207, if the difference between the current vehicle speed and the previously stored vehicle speed in the driving history data is greater than or equal to the first threshold (S207: Yes), the memory processing unit 1504 corrects the road surface vibration data stored in the driving history data 1511 with the current vehicle speed (S211). Then, the forward data output unit 1505 outputs the corrected road surface vibration data to the control parameter learning model 1411 (S212).

[0097] Next, the memory processing unit 1504 acquires vehicle behavior data from the vehicle behavior data processing unit 1401 (S209). Then, the memory processing unit 1504 overwrites and updates the driving history data 1511 by linking the road surface vibration data calculated from the acquired vehicle behavior data to the location information (S210). And then the process is finished.

[0098] Next, the learning process in the vehicle control process according to this embodiment will be described. Figure 12 is a flowchart showing an example of the learning process procedure in the vehicle control process according to the second embodiment.

[0099] First, the location information acquisition unit 1501 acquires location information of the current location of vehicle 1 from the location information sensor 31 (S301). Next, the driving history determination unit 1502 determines whether driving history data 1511 is stored in the storage unit 1510 and whether the sprung mass vertical acceleration of vehicle 1 is greater than or equal to the second threshold (S302). If driving history data 1511 is not stored in the storage unit 1510, or if the sprung mass vertical acceleration of vehicle 1 is less than the second threshold (S302), the process ends.

[0100] On the other hand, if the memory unit 1510 stores driving history data 1511 and the vertical acceleration of the vehicle 1 is greater than or equal to the second threshold (S302: Yes), the control parameter learning model 1411 receives data related to road surface vibration from the forward data output unit 1505 (S303). As a result, the control parameter learning model 1411 outputs a control gain.

[0101] The subsequent processes, from determining whether to proceed with inference or learning, to processing in inference mode (S103) and processing in learning mode (S104-S107), are the same as in the first embodiment.

[0102] (Overview) As described above, the vehicle control device 514 according to this embodiment includes a position information acquisition unit 1501 that acquires position information of the vehicle 1, and a storage unit 1510 that stores driving history data 1511 which associates the position information, vehicle speed, and data related to road surface vibration acquired from the vehicle behavior data. The driving history data 1511 stores the position information of the road surface, the vehicle speed, and data related to road surface vibration when the vehicle 1 is driving on a road surface for which there is no driving history, and the control parameter learning model 1411 takes the data related to road surface vibration as input and outputs a control gain (control parameter). When vehicle 1 is traveling to a location stored in the travel history data 1511, indicated by location information, the learning unit 1405 determines that the difference between the current vehicle speed acquired by the vehicle behavior data processing unit 1401 and the vehicle speed associated with the location information of the current travel location in the travel history data 1511 stored in the storage unit 1510 is less than a first threshold, and uses the control parameter learning model 1411 to determine the control gain (control parameter) from the road surface vibration data associated with the location information of the current travel location in the travel history data 1511.

[0103] Therefore, in this embodiment, since the learning process is performed using road surface vibration data stored in the driving history data 1511, it has the same effects as the first embodiment, and it is not necessary to provide an imaging unit 15c for imaging the road surface ahead on the vehicle 1, and the learning process can be performed by providing a position information sensor 31 such as a navigation system and a position information acquisition unit 1501 on the vehicle 1. Furthermore, according to this embodiment, it is not necessary to learn to cover the optimal output value for a given position, and efficient learning can be achieved. Moreover, according to this embodiment, it is also possible to use learning results from other locations.

[0104] Furthermore, in the vehicle control device 514 according to this embodiment, when the vehicle is traveling to a location stored in the travel history data 1511 indicated by location information, if the difference between the current vehicle speed acquired by the vehicle behavior data processing unit 1401 and the vehicle speed associated with the location information of the current travel location in the travel history data 1511 stored in the storage unit 1510 exceeds a first threshold, the learning unit 1405 corrects the road surface vibration data associated with the location information of the current travel location in the travel history data 1511 based on the current vehicle speed, and uses the control parameter learning model 1411 to determine the control gain (control parameter) from the corrected road surface vibration data.

[0105] In other words, in this embodiment, if there is a significant difference between the current vehicle speed and the vehicle speed associated with the position information of the current driving location in the driving history data 1511, the road surface vibration data associated with the position information of the current driving location in the driving history data 1511 is corrected based on the current vehicle speed, and then the control parameter learning model 1411 is trained. Therefore, according to this embodiment, highly robust, or in other words, highly generalizable learning can be performed, and more optimal vehicle control can be achieved.

[0106] Furthermore, in the vehicle control device 514 according to this embodiment, when the vehicle 1 travels to a location indicated by location information, the memory processing unit 1504 overwrites and updates the travel history data 1511 in the memory unit 1510 with road surface vibration data obtained by traveling to the same location information each time the vehicle travels to that location information. Therefore, according to this embodiment, it is possible to always store the latest travel history data 1511 by reflecting the changes in the road surface and the vehicle 1 over time in the travel history data 1511. In addition, in this embodiment, since the travel history data can always be kept up-to-date in this way, more robust (in other words, more generalizable) learning can be performed, and more optimal vehicle control can be achieved.

[0107] Furthermore, in the vehicle control device 514 according to this embodiment, the vehicle behavior data includes the sprung mass vertical acceleration of the vehicle 1. The memory processing unit 1504 stores the driving history data 1511 in the memory unit 1510 by associating the position information with the road surface vibration data only at locations where the sprung mass vertical acceleration exceeds a second threshold when the vehicle 1 is driving on a road surface at a location where there is no driving history. The learning unit 1405 learns the control parameter learning model 1411 when the vehicle 1 is driving on a location already stored in the driving history data 1511. Therefore, according to this embodiment, the storage capacity of the memory unit 1510 that stores the driving history data 1511 can be reduced.

[0108] Furthermore, in the vehicle control device 514 according to this embodiment, the memory processing unit 1504 registers the vehicle behavior data, specifically the frequency-analyzed data, into the driving history data 1511. Therefore, according to this embodiment, the storage capacity of the memory unit 1510 that stores the driving history data 1511 can be reduced.

[0109] The vehicle control program executed by the vehicle control device 14 according to the above embodiment is provided pre-installed in a ROM 14b or the like.

[0110] The vehicle control programs executed by the vehicle control devices 14 and 514 according to the above embodiments may be configured to be provided as files in an installable or executable format, stored on a computer-readable recording medium such as a CD-ROM, flexible disk (FD), CD-R, or DVD (Digital Versatile Disk).

[0111] Furthermore, the vehicle control program executed by the vehicle control devices 14 and 514 according to the above embodiment may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Alternatively, the vehicle control program executed by the vehicle control device 14 according to the above embodiment may be provided or distributed via a network such as the Internet.

[0112] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents.

[0113] (Summary of this embodiment) The vehicle control device 14 according to this embodiment includes a forward data output unit 1406 that acquires data on road surface vibrations in front of the moving vehicle 1, a vehicle behavior data processing unit 1401 that acquires vehicle behavior data on the behavior of the moving vehicle 1, and a vehicle control unit 1402 that pre-learns a control parameter learning model 1411 that takes road surface vibration data as input and outputs control parameters for actuators 102, 103, 105, and 106 that control the behavior of the vehicle 1, and controls the actuators 102, 103, 105, and 106 based on the learned control parameter learning model 1411 and the vehicle behavior data.According to this embodiment, even when the vehicle 1 is traveling on a road surface different from that at the time of learning, the optimal control gain can be output, thereby enabling highly robust (in other words, highly generalizable) learning and more optimal vehicle control.

[0114] Furthermore, in the vehicle control device 14 according to this embodiment, the vehicle behavior data processing unit 1401 further acquires vehicle behavior data of the moving vehicle 1 by controlling actuators 102, 103, 105, and 106 based on control gains. The vehicle control device 14 also includes a reward calculation unit 1404 that calculates a reward value from the acquired vehicle behavior data using a predetermined reward function, and a learning unit 1405 performs reinforcement learning on the control parameter learning model 1411 so as to maximize the reward value. According to this embodiment, more robust learning can be performed, and more optimal vehicle control can be achieved.

[0115] Furthermore, in the vehicle control device 14 according to this embodiment, the forward data output unit 1406 determines the condition of the road surface in front of the vehicle 1 based on the image captured by the imaging unit 15c, which is capable of capturing images in front of the vehicle 1, and obtains data related to road surface vibration based on the road surface condition and the vehicle speed, which is the speed of the vehicle included in the vehicle behavior data. According to this embodiment, more robust learning can be performed, and more optimal vehicle control can be achieved.

[0116] Furthermore, in the vehicle control device 14 according to this embodiment, the control parameter learning model 1411 is pre-learned using road surface vibration data, control parameters output using that data as input, and reward values ​​obtained when driving on that road surface, and is further learned in the user's operating environment. According to this embodiment, when a user drives the vehicle 1, it can be driven with the optimal control gain (control parameters) from the beginning, and learning can be performed to suit the user's environment.

[0117] The vehicle control device 514 according to this embodiment includes a storage unit 1510 that stores driving history data 1511 which associates location information with road surface vibration data obtained from vehicle behavior data. When the vehicle 1 travels on a road surface at a location where there is no driving history, the location information of that road surface and the road surface vibration data are stored in association with each other in the driving history data 1511. When the vehicle 1 travels on a location stored in the driving history data 1511 indicated by the location information, the learning unit 1405 inputs the stored road surface vibration data ahead into the control parameter learning model 1411 and outputs control parameters. In this embodiment, it is not necessary to provide an imaging unit 15c for imaging the road surface ahead on the vehicle 1, and learning processing can be performed by providing a location information sensor 31 such as a navigation system and a location information acquisition unit 1501 on the vehicle 1. Furthermore, according to this embodiment, it is not necessary to learn to cover the optimal output value for a given location, and efficient learning can be achieved. Moreover, according to this embodiment, it is also possible to utilize learning results from other locations.

[0118] In the vehicle control device 514 according to this embodiment, when the vehicle 1 is traveling to a location stored in the travel history data 1511 indicated by location information, the learning unit 1405 determines that the difference between the current vehicle speed acquired by the vehicle behavior data processing unit 1401 and the vehicle speed associated with the location information of the current travel location in the travel history data 1511 stored in the storage unit 1510 is less than a first threshold, and uses the control parameter learning model 1411 to determine control parameters from the road surface vibration data associated with the location information of the current travel location in the travel history data 1511. In this embodiment, it is not necessary to provide the vehicle 1 with an imaging unit 15c that images the road surface ahead, and learning processing can be performed by providing the vehicle 1 with a location information sensor 31 such as a navigation system and a location information acquisition unit 1501. Furthermore, according to this embodiment, it is not necessary to learn to cover the optimal output values ​​for a given location, and efficient learning can be achieved. Moreover, according to this embodiment, it is also possible to utilize learning results from other locations.

[0119] Furthermore, in the vehicle control device 514 according to this embodiment, when the vehicle 1 travels to a location stored in the travel history data 1511 indicated by location information, if the difference between the current vehicle speed acquired by the vehicle behavior data processing unit 1401 and the vehicle speed associated with the location information of the current travel location in the travel history data 1511 stored in the storage unit 1510 exceeds a first threshold, the learning unit 1405 corrects the road surface vibration data associated with the location information of the current travel location in the travel history data 1511 based on the current vehicle speed, and uses the control parameter learning model 1411 to determine the control gain from the corrected road surface vibration data. According to this embodiment, it is possible to perform highly robust, or in other words, highly generalizable, learning, and to perform more optimal vehicle control.

[0120] Furthermore, in the vehicle control device 514 according to this embodiment, when the vehicle 1 travels to a location indicated by location information, the memory processing unit 1504 overwrites and updates the travel history data 1511 in the memory unit 1510 with road surface vibration data obtained by traveling to the same location information each time the vehicle travels to that location information. According to this embodiment, it is possible to always store the latest travel history data 1511 by reflecting the changes in the road surface and the vehicle 1 over time in the travel history data 1511. In addition, in this embodiment, since the travel history data can always be kept up-to-date in this way, more robust (in other words, more generalizable) learning can be performed, and more optimal vehicle control can be achieved.

[0121] Furthermore, the vehicle control device 514 according to this embodiment further includes a storage processing unit 1504 that stores driving history data 1511 in the storage unit 1510 by associating position information with road surface vibration data only at locations where the vehicle behavior data exceeds a predetermined threshold (second threshold) when the vehicle 1 is driving on a road surface at a location where there is no driving history. The learning unit 1405 learns the control parameter learning model 1411 when the vehicle 1 is driving on a location where the driving history data 1511 has been stored. According to this embodiment, the storage capacity of the storage unit 1510 that stores the driving history data 1511 can be reduced.

[0122] Furthermore, in the vehicle control device 514 according to this embodiment, the memory processing unit 1504 stores data related to road surface vibration, specifically data obtained by frequency analysis, in the driving history data 1511. According to this embodiment, the storage capacity of the memory unit 1510 that stores the driving history data 1511 can be reduced.

[0123] (Note) In the vehicle control device according to the above embodiment, data regarding forward road surface vibrations is acquired by a camera or position information.

[0124] In the vehicle control device according to the above embodiment, the vehicle behavior data is the vertical acceleration of the vehicle. [Explanation of Symbols]

[0125] 1...Vehicle, 14...ECU (Vehicle Control Unit), 15, 15a, 15c...Imaging Unit, 30...Suspension System, 31...Position Information Sensor, 102, 103, 105, 106...Actuator, 1401...Vehicle Behavior Data Processing Unit, 1402...Vehicle Control Unit, 1403...Learning Inference Switching Unit, 1404...Reward Calculation Unit, 1405...Learning Unit, 1406, 1505...Forward Data Output Unit, 1411...Control Parameter Learning Model, 1501...Position Information Acquisition Unit, 1502...Driving History Determination Unit, 1503...Memory Determination Unit, 1510...Memory Unit, 1511...Driving History Data.

Claims

1. A vehicle control device installed in a vehicle, A forward data output unit that acquires data on road surface vibrations in front of the vehicle while it is in motion, A vehicle behavior data processing unit that acquires vehicle behavior data relating to the behavior of the vehicle while it is in motion, Based on the data relating to the forward road surface vibration, a control parameter learning model is pre-trained to take the road surface vibration data as input and output control parameters for an actuator that controls the behavior of the vehicle. Based on the control parameters output from the trained control parameter learning model and the vehicle behavior data, a vehicle control unit controls the actuator. A vehicle control device equipped with the following features.

2. The control parameter learning model is pre-trained using data on the road surface vibrations ahead, and is further trained in the user's environment. The vehicle control device according to claim 1.

3. A location information acquisition unit that acquires location information of the aforementioned vehicle, The system includes a storage unit that stores driving history data that associates the aforementioned location information with the road surface vibration data obtained from the vehicle behavior data, The aforementioned driving history data stores, in association with location information of the road surface and data related to the road surface vibration when the vehicle travels on a road surface in a location where there is no prior driving history. The aforementioned vehicle control device is When the vehicle travels to a location stored in the driving history data indicated by the location information, the learning unit inputs the stored data regarding the road surface vibration ahead to the control parameter learning model and outputs the control parameters. The vehicle control device according to claim 1, further comprising:

4. The system further includes a storage processing unit that, when the vehicle is traveling on a road surface in a location where there is no record of the vehicle's travel history, stores the travel history data in the storage unit by associating the location information with the road surface vibration data only at locations where the vehicle's behavior data exceeds a predetermined threshold, The learning unit learns the control parameter learning model when the vehicle travels to a location stored in the travel history data. The vehicle control device according to claim 3.

Images