Control device, control method, and program

Abstract

To provide a control device capable of achieving control with less delay while suppressing an influence due to noise and defects of input information.SOLUTION: A vehicle control device 1 is provided with a target control amount calculation part 11 for calculating a target control amount to a control amount determined as a control object on the basis of time series data of input information. The target control amount calculation part 11 includes an input information acquisition part 2 for acquiring input information at different times of M points (M is an integer of two or more) for a period from the current time to a prescribed time before, control amount predictors 30, ..., 3M-1 for calculating a prediction control amount on the basis of the input information, and a weighted average calculator 4 for calculating a weighted average value of M prediction control amounts calculated by the control amount predictors. The i-th (i is an integer between 0 and M-1) control amount predictor 3i calculates a prediction control amount by using a control amount prediction model for associating input information from the current time to a prediction time τi before with a control amount at the current time.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a control device, a control method, and a program, and more particularly to a control device, a control method, and a program for controlling a control amount determined for a control object. [Background technology]

[0002] In recent years, efforts to provide access to sustainable transportation systems that take into consideration vulnerable transport participants have become more active. To achieve this, we are focusing on research and development into autonomous driving technology to further improve traffic safety and convenience.

[0003] For example, Patent Document 1 describes an autonomous driving technology in which a control device automatically controls the steering angle without relying on steering operation by the driver. The control device described in Patent Document 1 determines a target steering angle for the steering angle of the vehicle so that the vehicle travels along an arc passing through the current position of the vehicle and a target point set on a target route. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2021-126925 Summary of the Invention [Problem to be solved by the invention]

[0005] In many autonomous driving technologies, a target control amount (target steering angle in the example of Patent Document 1) for a control amount is calculated based on information acquired by external sensors such as cameras and radars mounted on the vehicle. However, since the information acquired by such external sensors contains observation noise and defects, the target control amount may fluctuate due to the influence of the noise and defects of the external sensors.

[0006] Furthermore, an FIR filter is known as a method for smoothing such vibrations of the target control amount. However, although smoothing can be achieved by applying an FIR filter, a delay also occurs (see FIG. 4 described later).

[0007] The present invention aims to provide a control device, control method, and program that can achieve control with little delay while suppressing the effects of noise and missing information in input, and ultimately to contribute to the development of sustainable transportation systems. [Means for solving the problem]

[0008] (1) A control device according to the present invention (for example, vehicle control devices 1, 1A, 1B described later) calculates a target control amount for a control amount determined for a control object (for example, a braking device 7, a power plant 8, and an electric power steering device 9 described later) based on input information, and operates the control object based on the target control amount. The control device includes: target control amount calculation means (for example, target control amount calculation units 11, 11A, 11B described later) that calculates the target control amount based on time-series data of the input information; and automatic operation means (for example, motor drive unit 12 described later) that operates the control object based on the target control amount. The target control amount calculation means includes input information acquisition means (for example, input information acquisition unit 2 described later) that acquires the input information at M different times (M is an integer of 2 or more) between the current time and a predetermined time ago; and M control amount prediction means (for example, control amount predictors 30, 31, ..., 3 described later) that calculate predicted control amounts based on the input information acquired by the input information acquisition means. M-1 ), and weighted average calculation means (for example, weighted average calculators 4 and 4A described later) that calculates a weighted average of the M predicted control amounts calculated by the M control amount prediction means as the target control amount, wherein the ith (i is an integer between 0 and M-1) control amount prediction means calculates the predicted control amount by using a control amount prediction model that associates the input information i hours before the current time with the control amount at the current time.

[0009] (2) In this case, it is preferable that the control quantity prediction model is constructed by machine learning using input sample data, which is time series data of the input information, and ideal output data, which is time series data of ideal control quantities for the input sample data, as training data.

[0010] (3) In this case, it is preferable that the i-th controlled variable prediction model is constructed using the training data that is a set of the input sample data and the ideal output data that is i hours ahead of the input sample data.

[0011] (4) In this case, it is preferable that the weighted average calculation means sets the i-th weight for the predicted control amount calculated by the i-th control amount prediction means to a value greater than the j-th weight for the predicted control amount calculated by the j-th control amount prediction means (j is an integer greater than i).

[0012] (5) In this case, it is preferable that the weighted average calculation means sets the k-th weight value so that it decreases exponentially with respect to the value k (k is an integer between 0 and M-1).

[0013] (6) In this case, it is preferable that the target control amount calculation means further includes a reliability acquisition means (e.g., a reliability acquisition unit 2A described below) that acquires the reliability of the input information i hours before the current time, and that the weighted average calculation means sets the value of the i-th weight based on the i-th reliability acquired by the reliability acquisition means.

[0014] (7) In this case, it is preferable that the controlled object is a steering mechanism of the vehicle (for example, the electric power steering device 9 described below), the control amount is a steering angle by the steering mechanism, and the input information includes external environment information regarding the surroundings of the vehicle.

[0015] (8) In this case, it is preferable that the controlled object is a vehicle driving device (e.g., a power plant 8 described below), the controlled variable is the driving force of the driving device, and the input information includes external environmental information relating to the surroundings of the vehicle.

[0016] (9) In this case, it is preferable that the controlled object is a braking device of a vehicle (for example, braking device 7 described below), the controlled variable is the braking force by the braking device, and the input information includes external environmental information regarding the surroundings of the vehicle.

[0017] (10) In this case, it is preferable that the control device further includes a learning means for learning the control quantity prediction model based on time series data of the control quantity and the input information during manual driving in which the driver of the vehicle is the main operator of the controlled object.

[0018] (11) A control method according to the present invention is a method for controlling a control variable set for a control object by a computer, and includes the steps of: acquiring input information at M different times (M is an integer of 2 or more) between the current time and a predetermined time ago; calculating M predicted control variables based on the input information at the M different times; calculating a weighted average of the M predicted control variables as a target control variable for the control variable; and operating the control object based on the target control variable, wherein the step of calculating the predicted control variable calculates the i-th predicted control variable by using a control variable prediction model that associates the input information at i hours (i is an integer between 0 and M-1) before the current time with the control variable at the current time.

[0019] (12) A program according to the present invention causes a computer to control a control variable set for a control object, and causes the computer to execute the following steps: acquiring input information at M different times (M is an integer of 2 or more) between the current time and a predetermined time ago; calculating M predicted control variables based on the input information at the M different times; calculating a weighted average of the M predicted control variables as a target control variable for the control variable; and operating the control object based on the target control variable; and the step of calculating the predicted control variable causes the computer to calculate the i-th predicted control variable by using a control variable prediction model that associates the input information at i hours (i is an integer between 0 and M-1) before the current time with the control variable at the current time. [Effects of the Invention]

[0020] (1) In the present invention, the target control amount calculation means calculates a target control amount for a control amount set for a control object based on time-series data of input information, and the automatic operation means operates the control object based on the calculated target control amount. The target control amount calculation means also includes input information acquisition means for acquiring input information at M different times between the current time and a predetermined time ago, M control amount prediction means for calculating predicted control amounts based on the input information at each of the M times, and weighted average calculation means for calculating a weighted average of the M predicted control amounts calculated by these means as the target control amount. According to the present invention, by calculating the weighted average of the M predicted control amounts as the target control amount, it is possible to suppress the effects of noise and deficiencies contained in the input information and calculate a target control amount with little vibration. Furthermore, in the present invention, the ith (i is an integer between 0 and M-1) control variable prediction means among a total of M control variable prediction means calculates a predicted control variable by using a control variable prediction model that associates input information at a time i hours before the current time (i.e., if the current time is defined as point 0, the ith time counting backward from the current time) with the control variable at the current time. In other words, each control variable prediction means calculates a control variable at the same current time as a predicted control variable based on input information at different times. Therefore, according to the present invention, it is possible to achieve control with little delay while suppressing the effects of noise and deficiencies contained in the input information, which can ultimately contribute to the development of sustainable transportation systems.

[0021] (2) In the present invention, each controlled variable prediction means calculates a predicted controlled variable from input information by using a controlled variable prediction model constructed by machine learning using input sample data (time-series data of input information) and ideal output data (time-series data of ideal controlled variables for the input sample data) as training data. Therefore, according to the present invention, it is possible to realize ideal control with little delay while suppressing the effects of noise and defects contained in the input information.

[0022] (3) In the present invention, the i-th controlled variable prediction means calculates a predicted controlled variable at the current time from input information i hours before the current time by using the i-th controlled variable prediction model constructed from teacher data that is a set of input sample data and ideal output data that is i hours ahead of the input sample data. Therefore, according to the present invention, it is possible to realize ideal control with little delay while suppressing the effects of noise and defects contained in the input information.

[0023] (4) In the present invention, the weighted average calculation means sets the ith weight for the predicted control amount calculated by the ith control amount prediction means to a value greater than the jth weight for the predicted control amount calculated by the jth (j is an integer greater than i) control amount prediction means. In other words, the weighted average calculation means sets the weight for the ith predicted control amount calculated based on input information i hours before the current time to a value greater than the weight for the jth predicted control amount calculated based on input information j hours before that. Generally, the closer the time of the input information is to the current time, the higher the prediction accuracy of the predicted control amount tends to be. Therefore, by setting the weight to a larger value in accordance with the increase in prediction accuracy, it is possible to calculate a highly accurate target control amount.

[0024] (5) In the present invention, the weighted average calculation means sets the k-th weight value so that it decreases exponentially with respect to the value k (k is an integer between 0 and M-1). This makes it possible to calculate the target control amount with high accuracy through simple calculations.

[0025] (6) In the present invention, the reliability acquisition means acquires the reliability of input information i hours before the current time, and the weighted average calculation means sets the i-th weight value based on the acquired i-th reliability. This allows for the calculation of highly accurate target control variables that reflect the reliability of the input information at each time.

[0026] (7) In the present invention, the target control amount calculation means calculates the target control amount for the steering angle by the steering mechanism according to the above-described procedure based on time-series data of input information including external environment information around the vehicle, and the automatic operation means operates the steering mechanism based on the calculated target control amount. Therefore, according to the present invention, it is possible to achieve steering control with little delay while suppressing the effects of noise and defects contained in the external environment information.

[0027] (8) In the present invention, the target control amount calculation means calculates the target control amount for the driving force of the driving drive device according to the above-described procedure based on time-series data of input information including external environment information about the vehicle, and the automatic operation means operates the driving device based on the calculated target control amount. Thus, according to the present invention, it is possible to achieve driving force control with little delay while suppressing the effects of noise and defects contained in the external environment information.

[0028] (9) In the present invention, the target control amount calculation means calculates the target control amount for the braking force by the braking device according to the above-described procedure based on time-series data of input information including external environment information around the vehicle, and the automatic operation means operates the braking device based on the calculated target control amount. Therefore, according to the present invention, it is possible to realize braking control with little delay while suppressing the effects of noise and defects contained in the external environment information.

[0029] (10) In the present invention, the learning means learns the control variable prediction model based on time-series data of control variables and input information during manual driving in which the driver of the vehicle is the main operator of the controlled object. Thus, according to the present invention, the input / output characteristics of the control variable prediction model can be changed in accordance with changes over time in the characteristics of the driver of the vehicle and the characteristics of external sensors for acquiring external information.

[0030] (11) The control method according to the present invention has the same effects as the control device described above.

[0031] (12) The program according to the present invention provides the same effects as the control device described above. [Brief explanation of the drawings]

[0032] [Figure 1] 1 is a diagram schematically illustrating the configuration of a vehicle control device according to a first embodiment of the present invention and a vehicle equipped with the same. [Figure 2] FIG. 2 is a diagram showing the configuration of an automatic steering control module in the vehicle control device. [Figure 3] 10 is a diagram comparing the target control amount calculated by the target control amount calculation unit among Comparative Example 1 (top), Comparative Example 2 (middle), and this embodiment (bottom). [Figure 4] FIG. 2 is a diagram schematically illustrating the configuration of teacher data. [Figure 5] FIG. 1 is a diagram schematically illustrating the configuration of training data when M controlled variable prediction models are constructed by machine learning. [Figure 6] FIG. 6 is a diagram showing the configuration of an automatic steering control module in a vehicle control device according to a second embodiment of the present invention. [Figure 7] FIG. 10 is a diagram showing the configuration of an automatic steering control module in a vehicle control device according to a third embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0033] First Embodiment A vehicle control device according to a first embodiment of the present invention will be described below with reference to the drawings.

[0034] FIG. 1 is a diagram schematically illustrating the configuration of a vehicle control device 1 according to this embodiment and a vehicle V equipped with the same. The upper part of FIG. 1 shows a plan view of the vehicle V, and the lower part of FIG. 1 shows a side view. Note that the following description will be given assuming that the vehicle V is a so-called right-hand drive four-wheel vehicle in which the driver's seat is located on the right side in the vehicle width direction as viewed along the direction of travel, but the present invention is not limited to this. The vehicle V may also be a so-called left-hand drive four-wheel vehicle in which the driver's seat is located on the left side in the vehicle width direction as viewed along the direction of travel.

[0035] The vehicle V is equipped with an electric power steering device 9 as a steering mechanism that steers the left and right front wheels Wf, a power plant 8 as a driving device that generates a driving force to rotate the front wheels Wf, which are the driving wheels of the vehicle V, a braking device 7 that generates a braking force to stop the rotation of the front wheels Wf and rear wheels Wr, a sensor unit 6 provided on the vehicle body, and a vehicle control device 1 that controls the electric power steering device 9, the power plant 8, and the braking device 7 based on the detection signal of the sensor unit 6 and driving operations by the driver (e.g., steering operations, acceleration / deceleration operations, braking operations, etc.).

[0036] The electric power steering device 9 includes a gearbox 93 that connects a pinion shaft 92 extending from a steering wheel 91 that receives steering operations by the driver to the left and right front wheels Wf, an electric motor 94 provided in the gearbox 93, and a steering sensor 95 that detects the steering angle of the steering wheel 91.

[0037] The gearbox 93 includes a rack shaft extending along the vehicle width direction and meshing with the pinion shaft 92, tie rods connecting both ends of the rack shaft to the left and right front wheels Wf, and converts the rotational movement of the steering wheel 91 caused by the driver's steering operation into movement along the vehicle width direction, thereby steering the left and right front wheels Wf in the direction of travel. The electric motor 94 rotates in response to a control signal output from the vehicle control device 1, and generates a driving force to assist the driver's steering operation or to automatically steer the front wheels Wf without the driver's steering operation. The steering sensor 95 detects the steering angle of the steering wheel 91 and sends a signal corresponding to the detected value to the vehicle control device 1.

[0038] The power plant 8 is a driving force generating source that generates a driving force for rotating the front wheels Wf to move the vehicle V forward or backward along the traveling direction in response to acceleration / deceleration operation of an accelerator pedal (not shown) by the driver and control signals output from the vehicle control device 1. In the following, a case will be described in which the power plant 8 is a drive motor that generates driving force by consuming electric power supplied from a high-voltage battery, a fuel cell stack, or the like (not shown), but the present invention is not limited to this. The power plant 8 may also be an engine that generates driving force by consuming fuel stored in a fuel tank (not shown), or a transmission that changes the speed of the engine output and transmits it to the front wheels Wf.

[0039] The braking device 7 includes a disc brake device that generates braking force to slow down or stop the rotation of each wheel Wf, Wr by tightening discs provided on the axles of each wheel Wf, Wr mainly when driving, based on braking operation of the brake pedal (not shown) by the driver or control signals output from the vehicle control device 1, and a parking brake that generates braking force to keep the rotation of each wheel Wr, Wf stopped mainly when parking.

[0040] The sensor unit 6 includes a camera unit 61, multiple (e.g., five) lidar units 62a, 62b, 62c, 62d, and 62e, multiple (e.g., five) radar units 63a, 63b, 63c, 63d, and 63e, a vehicle body sensor 64, and an external environment recognition device 65.

[0041] The camera unit 61 is a camera that captures images in front of the vehicle V. The camera unit 61 is attached, for example, to a position on the inside of the cabin of the roof of the vehicle V, near the front window. Images captured by the camera unit 61 are transmitted to the external environment recognition device 65.

[0042] The lidar units 62a to 62e are light detection and ranging (LIDAR) devices that detect objects around the vehicle V by measuring scattered light from the objects in response to pulsed laser irradiation. The first lidar unit 62a is provided in the right corner of the front of the vehicle V as viewed along the direction of travel, and detects objects in front of and slightly to the right of the vehicle V. The second lidar unit 62b is provided in the left corner of the front of the vehicle V as viewed along the direction of travel, and detects objects in front of and slightly to the left of the vehicle V. The third lidar unit 62c is provided in the center of the rear of the vehicle V in the vehicle width direction, and detects objects behind the vehicle V. The fourth lidar unit 62d is provided on the rear side of the right side of the vehicle V, and detects objects behind the vehicle V on the right side. The fifth lidar unit 62e is provided on the rear side of the left side of the vehicle V, and detects objects behind the vehicle V on the left side. Detection signals from these lidar units 62a to 62e are transmitted to an external environment recognition device 65.

[0043] The radar units 63a to 63e are millimeter-wave radars that detect objects around the vehicle V by measuring waves reflected from the objects in response to millimeter-wave irradiation. The first radar unit 63a is provided in the right corner of the front of the vehicle V as viewed in the direction of travel, and detects objects slightly forward and to the right around the vehicle V. The second radar unit 63b is provided in the left corner of the front of the vehicle V as viewed in the direction of travel, and detects objects slightly forward and to the left around the vehicle V. The third radar unit 63c is provided in the center of the front of the vehicle V in the vehicle width direction, and detects objects in front around the vehicle V. The fourth radar unit 63d is provided in the right corner of the rear of the vehicle V as viewed in the direction of travel, and detects objects slightly rearward and to the right around the vehicle V. The fifth radar unit 63e is provided in the left corner of the rear of the vehicle V as viewed in the direction of travel, and detects objects slightly rearward and to the left around the vehicle V. Detection signals from these radar units 63a to 63e are transmitted to the external environment recognition device 65.

[0044] The vehicle body sensor 64 transmits a signal corresponding to the motion state of the vehicle body, such as the vehicle speed and acceleration of the vehicle V, to the external environment recognition device 65.

[0045] The external environment recognition device 65 is a computer that performs sensor fusion processing on the detection results of the camera unit 61, the lidar units 62a to 62e, the radar units 63a to 63e, and the vehicle body sensor 64, thereby recognizing external environment information about the surroundings of the vehicle V (positions, distances, and relative speeds of obstacles and other vehicles, as well as the types and positions of lane markings, etc.) and vehicle state information about the motion state of the vehicle V (vehicle speed, direction of travel, acceleration, etc.), and further evaluating the reliability of the recognition results. The external environment recognition device 65 transmits information about the recognition results, information about the reliability of the recognition results, etc. to the vehicle control device 1.

[0046] The vehicle control device 1 is a computer configured with hardware such as a processing means such as a CPU, auxiliary storage means such as an HDD or SSD storing various programs, and main storage means such as a RAM for storing data temporarily required for the processing means to execute the programs. With this hardware configuration, the vehicle control device 1 is configured with an automatic steering control module that controls the electric power steering device 9 as a control object, an automatic driving control module that controls the power plant 8 as a control object, and an automatic braking control module that controls the braking device 7 as a control object.

[0047] The automatic steering control module of the vehicle control device 1 calculates a target control amount for a control amount (e.g., steering angle) set for the electric power steering device 9 based on input information acquired from the external environment recognition device 65, and operates the electric power steering device 9 so that the control amount in the electric power steering device 9 becomes the target control amount.

[0048] The automatic driving control module of the vehicle control device 1 calculates a target control amount for a control amount (e.g., driving force) set for the power plant 8 based on input information acquired from the external environment recognition device 65, and operates the power plant 8 so that the control amount in the power plant 8 becomes the target control amount.

[0049] The automatic braking control module of the vehicle control device 1 calculates a target control amount for a control amount (e.g., braking force) set for the braking device 7 based on input information acquired from the external environment recognition device 65, and operates the braking device 7 so that the control amount in the braking device 7 becomes the target control amount.

[0050] Fig. 2 is a diagram showing the configuration of an automatic steering control module related to control of the electric power steering device 9 in the vehicle control device 1. Note that the configurations of the automatic driving control module and the automatic braking control module are almost the same as the configuration of the automatic steering control module shown in Fig. 2, so detailed explanations will be omitted.

[0051] The vehicle control device 1 calculates a target control amount (hereinafter referred to as "u") for a control amount (hereinafter referred to as "u") output from a steering sensor 95 of the electric power steering device 9. cmd The target control amount calculation unit 11 calculates the target control amount u based on input information (hereinafter referred to as "x") transmitted from the external environment recognition device 65. cmd and a motor drive unit 12 that operates the electric motor 94 of the electric power steering device 9 based on the above.

[0052] In the following, variables at time t are written in parentheses. That is, the input information, the controlled variable, and the target controlled variable at time t are x(t), u(t), and u(t), respectively. cmd It is written as (t).

[0053] The motor drive unit 12 calculates the target control amount u calculated by the target control amount calculation unit 11 according to the procedure described below. cmdand the control amount u output from the steering sensor 95 match, the electric motor 94 is operated (that is, the drive current to the electric motor 94 is adjusted) based on a known feedback control algorithm.

[0054] The target control amount calculation unit 11 includes an input information acquisition unit 2 and a plurality of control amount predictors 30, 31, . . . , 3 M-1 and a weighted average calculator 4.

[0055] The input information acquisition unit 2 acquires time-series data of input information x, which is information necessary for controlling the control target, from the recognition result of the external environment recognition device 65. Note that the input information x acquired by the input information acquisition unit 2 is N (N is an integer of 2 or more) input variables (x1, x2, ..., x), such as lane marking position information, the distance to the vehicle in front, and the relative speed with respect to the vehicle in front, as shown in the following formula (1). N ) is an N-dimensional vector composed of

number

[0056] The input information acquisition unit 2 acquires input information at M different times (M is an integer of 2 or more, hereinafter also referred to as "number of samples") from the external environment recognition device 65 between the current time and a predetermined time ago. That is, when the current time is "t" and the sampling time is represented as "Δt", the input information acquisition unit 2 acquires input information at M different times [t-τ0, t-τ1, t-τ2, ..., t-τ] between the current time t and a predetermined time Δt × (M-1). M-1 ] in the input information [x(t-τ0),x(t-τ1),x(t-τ2),…,x(t-τ M-1 In the following, if "i" is any integer between 0 and M-1, the difference between the current time t and the time t-Δt×i i points before the current time t is called "τ i " and is also simply called i-hours.

[0057] As shown in FIG. 2, the input information acquisition unit 2 includes control variable predictors 30, 31, 32, ..., 33, the number of which is equal to the number M of samples in the input information acquisition unit 2.M-1 are connected to each of the control variable predictors 30, 31, 32, ..., 3 M-1 are the different times [t-τ0, t-τ1, t-τ2, ..., t-τ M-1 ] in the input information [x(t-τ0),x(t-τ1),x(t-τ2),…,x(t-τ M-1 )], the predicted controlled variable [u τ0 (t),u τ1 (t),u τ2 (t),…,u τM-1 (t)] is calculated.

[0058] In the following, the input information x(t-τ i ) based on the predicted control variable u τi Control variable predictor 3 that calculates (t) i is called the i-th control variable predictor, and this i-th control variable predictor 3 i The predicted control amount u calculated by τi (t) is called the i-th predicted controlled variable. In the following, the i-th controlled variable predictor 3 i Input x(t-τ i ) and output u τi (t) and the time difference τ i is also called the predicted time.

[0059] That is, the 0th controlled variable predictor 30 calculates the predicted value of the controlled variable at the current time t as the 0th predicted controlled variable u based on the input information x(t) at the current time t. τ0 The first controlled variable predictor 31 calculates the predicted value of the controlled variable at the current time t as the first predicted controlled variable u(t) based on input information x(t-τ1) from the current time t to the prediction time τ1 before. τ1 The second controlled variable predictor 32 calculates the predicted value of the controlled variable at the current time t as the second predicted controlled variable u(t) based on the input information x(t-τ2) from the current time t to the prediction time τ2 before. τ2 (t) is calculated as the (M-1)th control variable predictor 3 M-1 is the predicted time τ from the current time t. M-1 Input information x(t-τM-1 ), the predicted value of the controlled variable at the current time t is calculated as the (M-1)th predicted controlled variable u τM-1 Calculated as (t).

[0060] These M controlled variable predictors 30, 31, 32, ..., 3 M-1 Each of the i-th control variable predictors 3 has a control variable prediction model that associates input information from the current time to the prediction time before with the control variable at the current time, and calculates the predicted control variable by using this control variable prediction model. i is the predicted time τ from the current time t. i Previous input information x(t-τ i ) and the control amount u(t) at the current time t. By using this i-th control amount prediction model, the input information x(t-τ i ) to the i-th predicted controlled variable u τi Calculate (t).

[0061] In the following, the input / output characteristics of the i-th controlled variable prediction model are expressed as the input information x(t-τ i ) as explanatory variables and the predicted control amount u τi It is expressed by a function f with (t) as the objective variable. τi " is a parameter that characterizes the input / output characteristics of the i-th controlled variable prediction model. In other words, as will be explained later, when the controlled variable prediction model is constructed by a neural network, "A τi " corresponds to multiple weighting coefficients that characterize the input / output characteristics of the neural network.

number

[0062] The weighted average calculator 4 calculates the weighted average of M controlled variable predictors 30, 31, 32, ..., 33 as shown in the following formula (3). M-1 The M predicted control variables [u τ0 (t),u τ1 (t),u τ2 (t),…,uτM-1 (t)] is the weighted average of the target control variable u cmd (t) and input to the motor drive unit 12.

number

[0063] Here, in the above formula (3), "w k " is the kth (k is an integer between 0 and M-1) predicted controlled variable u τk (t), which is also referred to as the kth weight. k The shorter the predicted control amount u τk (t) tends to be highly reliable. Therefore, the weighted average calculator 4 calculates the i-th weight w i The jth (j is an integer greater than i) weight w j In other words, the weighted average calculator 4 calculates the predicted control amount u τk Weight w for (t) k , the predicted time τ k More specifically, the weighted average calculator 4 calculates the k-th weight w so that it decreases exponentially with respect to the value k, as shown in the following equations (4-1) to (4-3): k It is preferable to set the value of In the following equation (4-1), "β" is a positive time constant.

number

[0064] Next, a procedure for controlling the control amount determined for the control object by the vehicle control device 1 as described above will be described. First, the input information acquisition unit 2 performs the control process for the predetermined time τ M-1 Input information [x(t-τ0),x(t-τ1),x(t-τ2),…,x(t-τ M-1 Next, M controlled variable predictors 30, 31, 32, ..., 3 M-1is the input information [x(t-τ0), x(t-τ1), x(t-τ2), ..., x(t-τ M-1 )] based on M predicted control variables [u τ0 (t),u τ1 (t),u τ2 (t),…,u τM-1 More specifically, the i-th control variable predictor 3 i , the predicted time τ from the current time t i Previous input information x(t-τ i ) and the control variable u(t) at the current time t, the i-th predicted control variable u τi Calculate (t).

[0065] Next, the weighted average calculator 4 calculates the weighted average of the control variables 30, 31, 32, ..., 3 M-1 The M predicted control variables [u τ0 (t),u τ1 (t),u τ2 (t),…,u τM-1 (t)] is the weighted average of the target control variable u cmd The motor drive unit 12 calculates the target control amount u cmd The control object is operated based on a known feedback control algorithm so that the control amount u(t) matches the control amount u(t) output from the steering sensor 95. The vehicle control device 1 controls the control object by repeatedly executing the above steps at predetermined time intervals (for example, sampling time Δt).

[0066] Next, the target control amount u is calculated by the target control amount calculation unit 11 as described above. cmd The effect achieved when calculating the target control amount u(t) will be described in comparison with Comparative Examples 1 and 2. Here, the target control amount calculation unit in Comparative Example 1 is a unit that calculates the target control amount u(t) based only on the input variable x(t). cmd That is, the target control amount u (t) calculated by the target control amount calculation unit of Comparative Example 1 is cmdIn the target control amount calculation unit 11 shown in FIG. 2, (t) is the 0th predicted control amount u calculated by the 0th control amount predictor 30. τ0 The target control amount calculation unit of the second comparative example is a unit that calculates the target control amount u by applying an FIR filter with M samples to the output of the target control amount calculation unit of the first comparative example. cmd (t) means the time from the current time t to the time τ i The output of Comparative Example 1 in the previous (i is an integer between 0 and M-1) is y τi (t), the target control amount u cmd (t) is expressed by the following equation (5).

number

[0067] FIG. 3 shows the u calculated by the target control amount calculation unit. cmd 3 is a diagram comparing Comparative Example 1 (top), Comparative Example 2 (middle), and this embodiment (bottom). For reference, ideal control amounts, which will be described later, are shown by dashed lines in FIG.

[0068] First, the target control amount u cmd The time series data of the input information x that is used to calculate the target control amount u contains noise and missing data. Therefore, the target control amount u calculated in Comparative Example 1, which only uses input information at a single time, is cmd In this case, the influence of noise and missing data contained in the time series data of the input information x is directly reflected, and oscillations may occur as shown in the upper part of Figure 3.

[0069] The vibrations shown in the upper part of Figure 3 can be removed by applying an FIR filter as shown in the middle part of Figure 3. However, simply applying an FIR filter will drag along past information, resulting in a noticeable delay from the ideal control amount as shown in the middle part of Figure 3.

[0070] In contrast to this, according to this embodiment, the input information acquisition unit 2 and M controlled variable predictors 30, 31, 32, . . . , 3 M-1and the weighted average calculator 4 are used in combination to calculate the target control amount u cmd By calculating (t), it is possible to eliminate the effects of noise and defects contained in the time-series data of the input information x, while at the same time reducing the delay with respect to the ideal control amount.

[0071] Next, M controlled variable predictors 30, 31, 32, ..., 3 M-1 This section explains the procedure for constructing M controlled variable prediction models using a neural network.

[0072] First, the operator prepares training data required to construct M controlled variable prediction models by machine learning, as shown in Fig. 4. As shown in Fig. 4, the training data is time series data [x(t0), x(t1), x(t2), ..., x(t L-3 ),x(t L-2 ),x(t L-1 )] (hereinafter, the time series data of the input information of the number of samples L is also referred to as "input sample data"), and the ideal control amount u of the number of samples L for the input sample data of the number of samples L ideal The time series data [u ideal (t0),u ideal (t1),u ideal (t2),…,u ideal (t L-3 ),u ideal (t L-2 ),u ideal (t L-1 )] (Hereinafter, the time series data of the ideal control amount of the sample number L will also be referred to as "ideal output data"). Here, the ideal control amount u ideal corresponds to the control amount that the target control amount calculation unit 11 attempts to realize.

[0073] For the input sample data and ideal output data of such sample number L, actual driving data obtained when an exemplary driver who can manually achieve ideal driving operations manually drives the vehicle V for a predetermined period of time can be used. More specifically, the input sample data of sample number L can be time series data of input information x acquired by the input information acquisition unit 2 when the exemplary driver manually drives the vehicle V for a predetermined period of time. Furthermore, the ideal output data for this input sample data can be time series data of the control amount u manually achieved by the exemplary driver for the same period as the period during which the input sample data was acquired.

[0074] Next, the operator constructs M controlled variable prediction models in order by machine learning using the input sample data and ideal output data prepared by the above procedure as training data. For example, the i-th controlled variable prediction model is constructed by input sample data of sample number L and prediction time τ i (=t i It is constructed from Li sets of training data, each set consisting of ideal output data advanced by t −t0.

[0075] That is, the 0th controlled variable prediction model, whose prediction time is 0, is calculated based on the input sample data [x(t0), x(t1), ..., x(t L-2 ),x(t L-1 )] and the ideal output data [u ideal (t0),u ideal (t1),…,u ideal (t L-2 ),u ideal (t L-1 )] and L sets of training data. The first control variable prediction model with a prediction time of τ1 (= t1 - t0) is constructed using the input sample data [x(t0), x(t1), ..., x(t L-3 ),x(t L-2 )] and the ideal output data [uideal (t1),u ideal (t2),…,u ideal (t L-2 ),u ideal (t L-1 )] and L-1 sets of training data are constructed. The prediction time is τ i (=t i The i-th controlled variable prediction model, which is the i-th controlled variable prediction model, is calculated based on the input sample data [x(t0), x(t1), ..., x(t L-2-i ),x(t L-1-i )] and the time is predicted for this input sample data as τ i The ideal output data [u ideal (t i ),u ideal (t i+1 ),…,u ideal (t L-2 ),u ideal (t L-1 )] and Li sets of training data.

[0076] The weighting coefficient A of the neural network that characterizes the input / output characteristics of the i-th controlled variable prediction model is τi is the error function E(A τi ) is determined according to a known algorithm so as to minimize

number

[0077] The vehicle control device 1 according to this embodiment provides the following effects. (1) The target control amount calculation unit 11 calculates a target control amount u for a control amount u determined for a control object based on time-series data of input information x. cmd The motor drive unit 12 calculates the target control amount u cmd The target control amount calculation unit 11 operates the control target based on the target control amount. MInput information [x(t-τ0),x(t-τ1),x(t-τ2),…,x(t-τ M-1 )], and a predicted control amount [u τ0 (t),u τ1 (t),u τ2 (t),…,u τM-1 M control variable predictors 30, 31, 32, ..., 3 M-1 and the M predicted control variables [u τ0 (t),u τ1 (t),u τ2 (t),…,u τM-1 (t)] is the weighted average of the target control variable u cmd (t). According to the vehicle control device 1, M predicted control variables [u τ0 (t),u τ1 (t),u τ2 (t),…,u τM-1 (t)] is the weighted average of the target control variable u cmd (t), the influence of noise and missing data in the time series data of the input information is suppressed, and the target control amount u with little vibration is obtained. cmd In the present invention, a total of M controlled variable predictors 30, 31, 32, ..., 3 M-1 Among them, the i-th control variable predictor 3 i is the predicted time τ from the current time t. i Input information x(t-τ i ) and the control amount u(t) at the current time t, the predicted control amount u τi (t) is calculated by each of the control amount predictors 30, 31, 32, ..., 3 M-1 is a control variable that predicts the control amount at the same current time based on input information at different times. τ0 (t),u τ1 (t),u τ2 (t),…,u τM-1(t)]. Therefore, the vehicle control device 1 can realize control with little delay while suppressing the effects of noise and defects contained in the time-series data of the input information, which can contribute to the development of a sustainable transportation system.

[0078] (2) In the vehicle control device 1, each of the control amount predictors 30, 31, 32, ..., 3 M-1 calculates the predicted control amount from the input information by using a control amount prediction model constructed by machine learning using input sample data (time series data of input information) and ideal output data (time series data of ideal control amount for the input sample data) as training data. Therefore, the vehicle control device 1 can realize ideal control with little delay while suppressing the effects of noise and defects contained in the time series data of the input information.

[0079] (3) In the vehicle control device 1, the i-th control variable predictor 3 i is the input sample data and the predicted time τ i By using the i-th controlled variable prediction model constructed from the training data set consisting of ideal output data advanced by τ i Previous input information x(t-τ i ) to the predicted control amount u at the current time t τi (t) is calculated. Therefore, the vehicle control device 1 can realize ideal control with little delay while suppressing the influence of noise and defects contained in the time-series data of the input information.

[0080] (4) In the vehicle control device 1, the weighted average calculator 4 calculates the i-th control variable predictor 3 i The predicted control amount u calculated by τi (t) the i-th weight w i jth (j is an integer greater than i) controlled variable predictor 3 j The predicted control amount u calculated by τj (t) jth weight w jIn other words, the weighted average calculator 4 calculates the weighted average from the current time t to the predicted time τ i Previous input information x(t-τ i ) the i-th predicted controlled variable u τi Weight w for (t) i and further past prediction time τ j Previous input information x(t-τ j ) the jth predicted controlled variable u τj Weight w for (t) j Generally, the closer the time of the input information is to the current time, the higher the prediction accuracy of the predicted controlled variable tends to be. Therefore, by setting the weight to a larger value in accordance with the increase in prediction accuracy, it is possible to obtain a highly accurate target controlled variable u cmd can be calculated.

[0081] (5) In the vehicle control device 1, the weighted average calculator 4 calculates the k-th weight w so that it exponentially decreases with respect to the value k (k is an integer between 0 and M-1). k This allows for a highly accurate target control variable u cmd can be calculated.

[0082] (6) In the vehicle control device 1, the target control amount calculation unit 11 calculates the target control amount u for the steering angle by the electric power steering device 9 according to the above-described procedure based on the time-series data of the input information including the external environment information around the vehicle. cmd The motor drive unit 12 calculates the target control amount u cmd The electric power steering device 9 is operated based on the above information. Therefore, the vehicle control device 1 can realize steering control with little delay while suppressing the influence of noise and defects contained in the external world information.

[0083] (7) In the vehicle control device 1, the target control amount calculation means configured in the automatic driving control module calculates the target control amount for the driving force of the power plant 8 according to the above-described procedure based on time-series data of input information including external environment information around the vehicle, and the automatic operation means operates the power plant 8 based on the calculated target control amount. Therefore, the vehicle control device 1 can achieve driving force control with little delay while suppressing the effects of noise and defects contained in the external environment information.

[0084] (8) In the vehicle control device 1, the target control amount calculation means configured in the automatic braking control module calculates the target control amount for the braking force by the braking device 7 using the above-mentioned procedure based on time-series data of input information including external environment information around the vehicle, and the automatic operation means operates the braking device 7 based on the calculated target control amount. Therefore, the vehicle control device 1 can achieve braking control with little delay while suppressing the effects of noise and defects contained in the external environment information.

[0085] Second Embodiment Next, a vehicle control device according to a second embodiment of the present invention will be described with reference to the drawings. In the following description, the same components as those in the vehicle control device 1 according to the first embodiment will be denoted by the same reference numerals, and detailed description thereof will be omitted.

[0086] Fig. 6 is a diagram showing the configuration of an automatic steering control module related to control of the electric power steering device 9 in the vehicle control device 1A according to this embodiment. Note that the configurations of the automatic driving control module and the automatic braking control module are almost the same as the configuration of the automatic steering control module shown in Fig. 6, so detailed explanations will be omitted.

[0087] As shown in FIG. 6, the target control variable calculation unit 11A includes an input information acquisition unit 2, a reliability acquisition unit 2A, and M control variable predictors 30, 31, . . . , 3 M-1 and a weighted average calculator 4A.

[0088] The reliability acquisition unit 2A acquires time series data of reliability c for the time series data of input information x acquired by the input information acquisition unit 2 from the external environment recognition device 65. More specifically, the reliability acquisition unit 2A acquires the time series data of reliability c for the M points of input information [x(t-τ0), x(t-τ1), x(t-τ2), ..., x(t-τ M-1 )] and the reliability for each [c(t-τ0),c(t-τ1),c(t-τ2),...,c(t-τ M-1 )] is obtained. That is, the reliability c(t-τ0) is the reliability for the input information x(t-τ0) at the same time, the reliability c(t-τ1) is the reliability for the input information x(t-τ1) at the same time, and the reliability c(t-τ M-1 ) is the input information x(t-τ M-1 ) is the degree of confidence in the

[0089] The weighted average calculator 4A calculates the weighted average of M controlled variable predictors 30, 31, 32, ..., 33 as shown in the following equations (7-1) and (7-2). M-1 The M predicted control variables [u τ0 (t),u τ1 (t),u τ2 (t),…,u τM-1 (t)] is obtained by the reliability obtaining unit 2A. M-1 )] is the weighted average value under the weight according to the target control amount u cmd (t) and input to the motor drive unit 12.

number

[0090] In the above formulas (7-1) and (7-2), “w k (c(t-τ k ))” is the kth predicted controlled variable u τk (t) is the kth weight for the kth input information x(t-τ k ) the kth confidence c(t-τ k ) is a function of the weighted average calculator 4A. k(c(t-τ k )) at the predicted time τ k It is preferable to set the weight w to a smaller value as the length of the k (c(t-τ k )) and the reliability c(t-τ k ) is smaller, in other words, the kth input information x(t-τ k ) is less reliable, the smaller the value should be set.

[0091] In addition to the effects (1) to (8) above, the vehicle control device 1A according to this embodiment has the following effects. (9) In the vehicle control device 1A, the reliability acquisition unit 2A calculates the predicted time τ i Previous input information x(t-τ i ) confidence in c(t-τ i ) and the weighted average calculator 4A calculates the i-th reliability c(t-τ i ) based on the i-th weight w i (c(t-τ i )) is set. This allows us to reflect the reliability c of the input information x at each time and obtain a highly accurate target control variable u cmd can be calculated.

[0092] <Third embodiment> Next, a vehicle control device according to a third embodiment of the present invention will be described with reference to the drawings. In the following description, the same components as those in the vehicle control device 1 according to the first embodiment will be denoted by the same reference numerals, and detailed description thereof will be omitted.

[0093] Fig. 7 is a diagram showing the configuration of an automatic steering control module related to control of the electric power steering device 9 in the vehicle control device 1B according to this embodiment. Note that the configurations of the automatic driving control module and the automatic braking control module are almost the same as the configuration of the automatic steering control module shown in Fig. 7, so detailed explanations will be omitted.

[0094] As shown in FIG. 7, the target control variable calculation unit 11B includes an input information acquisition unit 2, a learning device 2B, and M control variable predictors 30, 31, . . . , 3 M-1 and a weighted average calculator 4.

[0095] The learning device 2B predicts each of the control amount predictors 30, 31, ..., 32 based on the time series data of the control amount u and the input information x during manual driving in which the driver of the vehicle V is the operator of the electric power steering device 9, which is the control object. M-1 The control variable prediction model is learned.

[0096] More specifically, the learning device 2B calculates an error function e(A) defined by the following equation (8-1) based on time-series data such as the control amount u and the input information x acquired while the electric power steering device 9 is being manually operated by the driver. τi ) is calculated. In addition, the learning device 2B calculates the update amount (the second term on the right side of the formula (8-2)) defined by the gradient descent method as shown in the following formula (8-2). τi The i-th controlled variable prediction model is updated by successively adding to

number

[0097] Here, in the above formula (8-1), "u ideal " is an ideal control amount, and time series data of the control amount u manually realized by the driver can be used. Also, in the above equation (8-2), "η" indicates a learning rate and is set to a predetermined value.

[0098] In addition to the effects (1) to (8) above, the vehicle control device 1B according to this embodiment has the following effects. (10) In the vehicle control device 1B, the learning device 2B learns M control variable prediction models based on time-series data of the control variable u and input information x during manual driving in which the driver of the vehicle V is the main operator of the controlled object. Therefore, the vehicle control device 1B can change the input / output characteristics of the M control variable prediction models in accordance with the characteristics of the driver of the vehicle V and changes over time in the characteristics of the sensor unit 6 for acquiring external information.

[0099] Although one embodiment of the present invention has been described above, the present invention is not limited to this, and the detailed configuration may be modified as appropriate within the scope of the spirit of the present invention. [Explanation of symbols]

[0100] V...Vehicle 1, 1A, 1B... Vehicle control device (control device) 11, 11A, 11B...Target control amount calculation unit (target control amount calculation means) 2...Input information acquisition unit (input information acquisition means) 2A... Reliability acquisition unit (reliability acquisition means) 2B...Learning device (learning means) 30, 31, …, 3 M-1 ...Controlled variable predictor (controlled variable prediction means) 4,4A...Weighted average calculator (additional average calculation means) 12...Motor drive unit (automatic operation means) 6...Sensor unit 7...Braking device (control object) 8...Power plant (travel drive device, control object) 9...Electric power steering device (steering mechanism, control object)

Claims

1. A control device that calculates a target control amount for a control amount set for a control object based on input information, and operates the control object based on the target control amount, a target control amount calculation means for calculating the target control amount based on time-series data of the input information; an automatic operation means for operating the controlled object based on the target control amount, The target control amount calculation means an input information acquisition means for acquiring the input information at M different times (M is an integer of 2 or more) between the current time and a predetermined time ago; M control amount prediction means for calculating predicted control amounts based on the input information acquired by the input information acquisition means; a weighted average calculation means for calculating a weighted average of the M predicted control amounts calculated by the M controlled amount prediction means as the target controlled amount, The control device is characterized in that the i-th (i is an integer between 0 and M-1) controlled variable prediction means calculates the predicted controlled variable by using a controlled variable prediction model that associates the input information i hours before the current time with the controlled variable at the current time.

2. 2. The control device according to claim 1, wherein the controlled variable prediction model is constructed by machine learning using, as training data, input sample data, which is time-series data of the input information, and ideal output data, which is time-series data of ideal controlled variables corresponding to the input sample data.

3. 3. The control device according to claim 2, wherein the i-th controlled variable prediction model is constructed from the teacher data that is a set of the input sample data and the ideal output data that is i hours ahead of the input sample data.

4. 4. The control device according to claim 1, wherein the weighted average calculation means sets an i-th weight for the predicted control amount calculated by the i-th control amount prediction means to a value greater than the j-th weight for the predicted control amount calculated by the j-th control amount prediction means (j is an integer greater than i).

5. The control device according to claim 4, wherein the weighted average calculation means sets the value of the kth weight so that it exponentially decreases with respect to the value k (k is an integer between 0 and M-1).

6. the target control amount calculation means further includes a reliability acquisition means for acquiring reliability of the input information i hours before the current time, 6. The control device according to claim 5, wherein the weighted average calculation means sets the ith weight value based on the ith reliability acquired by the reliability acquisition means.

7. the controlled object is a steering mechanism of a vehicle, the control amount is a steering angle by the steering mechanism, The control device according to claim 1 , wherein the input information includes external environment information relating to the surroundings of the vehicle.

8. the controlled object is a vehicle driving device, the controlled variable is a driving force by the driving device; The control device according to claim 1 , wherein the input information includes external environment information relating to the surroundings of the vehicle.

9. the controlled object is a braking device of a vehicle, the controlled variable is a braking force applied by the braking device, The control device according to claim 1 , wherein the input information includes external environment information relating to the surroundings of the vehicle.

10. 10. The control device according to claim 7, further comprising: a learning means for learning the control amount prediction model based on time-series data of the control amount and the input information during manual driving in which a driver of the vehicle is an operating subject of the controlled object.

11. A control method for controlling a control amount determined for a control object by a computer, comprising: acquiring input information at M different times (M is an integer of 2 or more) between the current time and a predetermined time ago; calculating M predicted control variables based on the input information at M different times; calculating a weighted average of the M predicted controlled variables as a target controlled variable for the controlled variable; and manipulating the controlled object based on the target control amount, In the step of calculating the predicted control amount, the i-th predicted control amount is calculated by using a control amount prediction model that associates the input information i hours before the current time (i is an integer between 0 and M-1) with the control amount at the current time.

12. A program that causes a computer to control a control amount determined for a control object, The computer, acquiring input information at M different times (M is an integer of 2 or more) between the current time and a predetermined time ago; calculating M predicted control variables based on the input information at M different times; calculating a weighted average of the M predicted controlled variables as a target controlled variable for the controlled variable; and operating the controlled object based on the target control amount, In the step of calculating the predicted control amount, the program calculates the i-th predicted control amount by using a control amount prediction model that associates the input information i hours before the current time (i is an integer between 0 and M-1) with the control amount at the current time.

Images