JPH05238402A - Road condition recognizing device for vehicle - Google Patents

Abstract

PURPOSE:Output Z of a neural network NN shows cant condition of a travelling road, because the above-mentioned neural network NN memorizes heart results of the cant condition of the travelling road as a coupling coefficient and a threshold vehicle. CONSTITUTION:An average computing units 41-44 equalize car speed V, lateral acceleration G, handle steering angle thetaf and handle torque T that are detected by a car speed sensor, a lateral acceleration sensor, a handle steering angle sensor and a handle torque sensor, and are fed time sequentially during the fixed time. The memorial network NN consits of an input layer 45, an intermediate layer 46, and an output layer 47, and carries out neural network computation based on respective signals from the average computing units 41-44.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vehicle road condition recognizing device which is mounted on a vehicle and recognizes an inclined state (a cant state) of a traveling road surface in a vehicle width direction.

[0002]

2. Description of the Related Art Conventionally, for example, JP-A-3-12561.
As shown in Japanese Patent Publication No. 5, when the steering angle of the steering wheel is a predetermined value or less and the vehicle speed is a predetermined value or more, the vehicle heights of the left and right wheels are adjusted according to the lateral acceleration of the vehicle detected by the lateral acceleration sensor. As a result, there is one in which the vehicle body is kept horizontal even if the road surface is canted. That is, in this device, according to the magnitude of the lateral acceleration,
The cant state of the road surface is recognized.

[0003]

In the above-mentioned conventional device, the steering angle and the vehicle speed are taken into consideration, but these are used only for the condition of judging the cant state of the road surface. However, the cant state of the road surface is determined only by the magnitude of lateral acceleration. However, when the vehicle is traveling on a cant road, the magnitude of this lateral acceleration changes depending on the vehicle speed and the operating state of the steering wheel. I can't recognize it. The present invention has been made to solve the above problem, and an object thereof is to provide a vehicle road condition recognition device for accurately recognizing a cant state of a traveling road surface.

[0004]

In order to achieve the above object, the structural features of the present invention include a vehicle speed sensor for detecting a vehicle speed, a lateral acceleration sensor for detecting a lateral acceleration of the vehicle, and a steering wheel operation state. The steering wheel sensor that detects the vehicle speed, the lateral acceleration of the vehicle, and the result of learning the cant state of the road surface based on the operating state of the steering wheel are stored, and the vehicle speed, the lateral acceleration of the vehicle, and the steering wheel detected by the sensors are stored. And a neural network means for judging the cant state of the road surface by a neural network calculation using the learning result based on the operation state.

[0005]

In the present invention constructed as described above, the neural network means is based on the vehicle speed detected by the vehicle speed sensor, the lateral acceleration detected by the lateral acceleration sensor, and the operation state of the steering wheel detected by the steering wheel sensor. Then, the cant state of the traveling road surface is determined by the neural network calculation using the learning result stored in the network means. In this case, the learning result stored in the neural network means is obtained by previously learning various cant states of the road surface based on the vehicle speed, the lateral acceleration of the vehicle, and the operating state of the steering wheel.
The neural network means determines not only the same cant state as the learned one but also the cant state other than the same learning result.

[0006]

As can be understood from the above description of the operation,
According to the present invention, the cant state of the road surface other than the learning result can be determined, and at the same time, the vehicle speed and the operating state of the steering wheel are taken into consideration in the determination, so that the cant state of the traveling road surface is accurately recognized. To be done.

[0007]

【Example】

a. First Embodiment First, a first embodiment in which a vehicle road condition recognition device according to the present invention is applied to a four-wheel steering vehicle will be described with reference to the drawings. FIG. 1 schematically shows this four-wheel steering vehicle. There is. This four-wheel steering vehicle has a front wheel steering mechanism 10 that steers the left and right front wheels FW1 and FW2, a rear wheel steering mechanism 20 that steers the left and right rear wheels RW1 and RW2, and an electric control that electrically controls the rear wheel steering mechanism 20. And a device 30.

The front wheel steering mechanism 10 is provided with a steering shaft 12 having a steering handle 11 fixed to the upper end thereof, and the lower end of the coaxial 12 penetrates into a steering gear box 13.
The steering gear box 13 supports the rack bar 14 so as to be displaceable in the axial direction and connects the lower end portion of the steering shaft 12 to the rack bar 14, and is an amount determined by a predetermined steering gear ratio with respect to the rotation of the steering wheel 11. Only the rack bar 14 is displaced in the axial direction. Rack bar 14
Left and right front wheels FW1 and FW2 are connected to both ends of the left and right wheels via tie rods 15a and 15b and knuckle arms 16a and 16b, and the left and right front wheels FW1 and FW2 are steered according to the axial displacement of the rack bar 14.

The rear wheel steering mechanism 20 includes an electrically controlled actuator 21, which drives a relay rod 22 axially displaceable in the axial direction. Tie rods 23a are provided at both ends of the relay rod 22,
The left and right rear wheels RW1 and RW2 are connected via the 23b and the knuckle arms 24a and 24b, and the left and right rear wheels RW are connected.
1 and RW2 are steered according to the axial displacement of the relay rod 22.

The electric control unit 30 includes a vehicle speed sensor 31, a lateral acceleration sensor 32, a steering wheel steering angle sensor 33, a steering wheel torque sensor 34, a yaw rate sensor 35 and a rear wheel steering angle sensor 36. The vehicle speed sensor 31 detects the vehicle speed V by measuring the rotation speed of the output shaft of a transmission (not shown), and outputs a detection signal indicating the vehicle speed V. The lateral acceleration sensor 32 detects the lateral acceleration G of the vehicle by measuring the widthwise acceleration of the vehicle body, and outputs a detection signal representing the lateral acceleration G. Steering wheel angle sensor 3
3 detects the steering wheel steering angle θf by measuring the rotation angle of the steering shaft 12, and outputs a detection signal indicating the steering wheel steering angle θf. The steering wheel torque sensor 34 is used for the steering shaft 12.
By measuring the torque acting on the steering wheel 11, the driver detects the steering wheel torque T applied to the steering wheel 11 and outputs a detection signal indicating the steering wheel torque T. The steering wheel steering angle sensor 33 and the steering wheel torque sensor 34 correspond to the steering wheel sensor for detecting the operating state of the steering wheel of the present invention. Yaw rate sensor 35
Detects the yaw rate γ by measuring the rotational angular velocity of the vehicle body about the vertical axis, and outputs a detection signal representing the yaw rate γ. The rear wheel steering angle sensor 36 is the relay rod 22.
The rear-wheel steering angle θr is detected by measuring the displacement amount of and the detection signal indicating the steering angle θr is output. The lateral acceleration G, the steering wheel steering angle θf, the steering wheel torque T, the yaw rate γ, and the rear wheel steering angle θr are positive in the right direction and negative in the left direction.

These sensors 31 to 36 are connected to a microcomputer 37. The microcomputer 37 is composed of a CPU, a ROM, a RAM, an I / O, a timer, etc., and executes a program corresponding to the flowchart of FIG. 2 stored in the ROM every predetermined time, and a cant state of the traveling road surface by a neural network operation. Is detected, and the rear wheels are steered according to the detection result. A drive circuit 38 is connected to the microcomputer 37, and the circuit 38 drives the actuator 21 according to a control signal from the computer 37.

Here, the neural network operation realized by the execution of the program will be described. FIG. 7 is a block diagram showing functions realized by executing the program. Average calculator 41
Reference numerals 44 to 44 represent vehicle speed V, lateral acceleration G, steering wheel angle θf, and steering wheel torque T, which are input from vehicle speed sensor 31, lateral acceleration sensor 32, steering wheel steering angle sensor 33, and torque sensor 34 over a predetermined time. _AV , G _AV , θf
_AV and T _AV are respectively calculated, and each average value V _AV , G _AV , θf _AV ,
Input data ID1 of neural network NN to T _AV
To ID4 are supplied to the input units 45-1 to 45-4 forming the input layer 45 of the network NN, respectively.

The neural network NN comprises an input layer 45 consisting of the input units 45-1 to 45-4, an intermediate layer 46 consisting of intermediate units 46-1 to 46-m, and a single output unit 47-1. The output layer 47 includes the input data IDi (i = 1 to 4), output values Xi of each input unit (i = 1 to 4), and intermediate units 46-1 to 46.
-m output value Yj (j = 1 to m) and output unit 47
The respective relationships of the output value Z of -1 are defined by the following mathematical expressions 1 to 3.

[0014]

[Equation 1] Xi = f (ω0i · IDi−θ0i)

[0015]

[Equation 2]

[0016]

[Equation 3]

Each of these coupling coefficients ω0i, ω1ij, ω2j and each threshold value θ0i, θ1j, θ2 is a constant representing a learning result in advance, and the function f (X) is a function such as a step function or a sigmoid function. Yes, these coupling coefficients ω0i, ω1ij, ω2
j, the threshold values θ0i, θ1j, θ2 and the function f (X) are stored in advance in the ROM of the microcomputer 37.

Next, the pre-learning will be described. First, the road surface condition recognizing device in which the neural network NN is configured by software or hardware as in the above embodiment is installed in the vehicle. However, in this neural network NN, the respective coupling coefficients ω0i, ω1ij, ω2
R capable of changing and storing j and the respective threshold values θ0i, θ1j, θ2
A memory device such as AM is prepared. Then, this vehicle is experimentally run on a cant road and a flat road, and the neural network NN is made to learn the road surface recognition state, and the learning result is used as the coupling coefficient ω0i, ω1ij, ω2j.
And the threshold values θ0i, θ1j, and θ2 are stored in the memory device.

This learning will be described with reference to the following specific examples. The vehicle is driven on a left cant road (a cant road descending leftward in the vehicle width direction) at a predetermined angle (for example, 3%). In this state, the average values V _AV , G _AV , θf _AV , and T _AV of the vehicle speed V, the lateral acceleration G, the steering wheel steering angle θf, and the steering wheel torque T are input, and “1” is input to the subtractor 48 as a teacher signal. Then, the deviation Δe between the teacher signal “1” and the output value Z is fed back to the neural network NN as shown by the broken line in FIG. Then, each coupling coefficient is calculated by using the back propagation method so that the deviation Δe is minimized.
ω0i, ω1ij, ω2j and the respective thresholds θ0i, θ1j, θ2 are sequentially updated. The vehicle is driven on a right cant road (a cant road descending to the right in the width direction of the vehicle) at a predetermined angle (for example, 3%). In this case, "-1" is supplied to the subtractor 48 as a teacher signal, and each coupling coefficient is calculated by using the back propagation method so that the deviation Δe is minimized in the same manner as described above.
ω0i, ω1ij, ω2j and the respective thresholds θ0i, θ1j, θ2 are sequentially updated. Drive the vehicle on a flat road. In this case, "0" is supplied as a teacher signal to the subtractor 48, and each coupling coefficient ω0i, ω1ij, ω2j and the respective thresholds θ0i, θ1j, θ2 are sequentially updated. It should be noted that the number of such learning cases is not limited to the above, and the more the better, the better.

Next, the operation of the first embodiment constructed as above will be described. Microcomputer 3 while the vehicle is running
7 executes the program including steps 100 to 107 of FIG. 2 at predetermined time intervals by the action of the built-in timer. In this program, after starting at step 100, at step 101, the vehicle speed sensor 31, the lateral acceleration sensor 32, the steering wheel steering angle sensor 33, the steering wheel torque sensor 34, the yaw rate sensor 35, and the rear wheel steering angle sensor 3 are detected.
The detection signals representing vehicle speed V, lateral acceleration G, steering wheel steering angle θf, steering wheel torque T, yaw rate γ, and rear wheel steering angle θr are input from 6. Next, at step 102, a "neural network calculation routine" is executed.

The details of this "neural network operation routine" are shown in FIG. 3, and after the start in step 110, the time series data V _{0 to} V _{1 is} entered in step 111.
_{n-1, G 0 ~G n} -1, and updates the _{θf 0 ~θf n-1, T} 0 ~T n-1. These time series data V _{0 to} V _n-1 , G _{0 to} G _n-1 , θ
f _{0 to} θf _n-1 and T _{0 to} T _n-1 are n pieces of sampling data of the input vehicle speed V, lateral acceleration G, steering wheel steering angle θf and steering wheel torque T, which are traced back from the present to the past. In the updating process, the oldest n-th data is deleted and the other data are sequentially shifted, and the newly input vehicle speed V, lateral acceleration G, steering wheel steering angle θf and steering wheel torque T are the latest data. V ₀ , G
Registered and stored as ₀ , θf ₀ , and T ₀ . After this updating process, in step 112, the time series data V _{0 to} V _n-1 , G _{0 to} G _n-1 , θf _{0 to} θf _n-1 , T _{0 to} T _{n- are} calculated based on the following _equations 4 to 7. ₁ of the average value _{V AV, G AV, θf AV} , with calculating the T _AV, the respective average value _{V AV, G AV, θf AV} , and the input data ID1~ID4 neural network calculates the T _AV.

[0022]

[Expression 4] ID1 = V _AV = (V ₀ + V ₁ + ... + V _n-1 ) / n

[0023]

[Equation 5] ID2 = G _AV = (G ₀ + G ₁ + ... + G _n-1 ) / n

[0024]

[Equation 6] ID3 = θf _AV = (θf ₀ + θf ₁ + ... + θf _n-1 ) / n

[0025]

## EQU7 ## ID4 = T _AV = (T ₀ + T ₁ + ... + T _n-1 ) / n The processes of these steps 111 and 112 correspond to the functions of the average calculators 41 to 44 in FIG.

Next, in steps 113 to 115, these input data ID1 to ID4 and the respective coupling coefficients ω0i and ω1i stored in the ROM of the microcomputer 37.
Using j, ω2j and the thresholds θ0i, θ1j, θ2,
The output values Xi (i = 1 to 4), Yj (j = 1 to m), and Z of the input layer, the intermediate layer, and the output layer, which form the neural network operation, are calculated by executing the operations of Equations 1 to 3 above. To do. The processes of steps 112 to 115 correspond to the input layer 45, the intermediate layer 46, and the output layer 47 of FIG.

In this case, the coupling coefficients ω0i, ω1ij, ω2j and the threshold values θ0i, θ1j, θ2 are determined based on the vehicle speed V, the lateral acceleration G, the steering angle θf and the steering wheel torque T on the cant state of the road surface. It is a result of learning, and each output value Z of the neural network operation is “1” when the traveling road surface is in the left cant state of 3%, “0” when the traveling road surface is flat, and 3% when the traveling road surface is 3%. When it is in the right cant state, it becomes "-1". Further, when the traveling road surface is in the cant state of less than 3%, the output value Z is interpolated by the neural network calculation and the output value Z is "0" to
It becomes a value between "1" or "0" to "-1". As a result, the cant state of the traveling path is output as a continuous quantity.

After the completion of the "neural network operation routine", the map provided in the ROM of the microcomputer 37 in step 103 of FIG. 2 (FIG. 4, FIG. 4).
By referring to 5), the coefficient K ₁ corresponding to the vehicle speed V,
K ₂ is determined and the same ROM is read at step 104.
The correction value Az corresponding to the output value Z is determined by referring to the map (FIG. 6) separately provided in FIG. Next, in step 105, the target rear wheel steering is performed by executing the calculation of the following equation 8 using the input steering angle θf and yaw rate γ and the determined coefficients K ₁ and K ₂ and the correction value Az. Calculate the angle θr *.

[0029]

[Equation 8] θr * = K ₁ θf + K ₂ γ + Az Next, in step 106, the target rear wheel steering angle θ calculated as described above.
A control signal representing a deviation θr * −θr obtained by subtracting the detected rear wheel steering angle θr from r * is output to the drive circuit 38. The drive circuit 38 outputs a drive signal corresponding to the control signal to the actuator 21.
The actuator 21 displaces the relay rod 22 leftward or rightward by an amount corresponding to the deviation θr * −θr. As a result, the left and right rear wheels RW1 and RW2 are steered so that the deviation θr * −θr becomes “0”, that is, the rear wheels RW1 and RW2 are steered to the target rear wheel steering angle θr *.

As a result of steering the left and right rear wheels RW1, RW2 to the target rear wheel steering angle θr * in this way, the coefficient K ₁ is negative if the vehicle speed V is low (see FIG. 4), and the target rear wheels Since the steering angle θr * has a positive and negative sign opposite to the steering wheel angle θf due to the influence of the first term K ₁ θf in the above equation 8,
The left and right rear wheels RW1 and RW2 are steered in reverse phase with respect to the left and right front wheels FW1 and FW2, and the small turning performance of the vehicle at low speed traveling is improved. If the vehicle speed V is high, the coefficient K
₂ is positive (see FIG. 5), and the target rear wheel steering angle θr * becomes a value in which the positive and negative signs are matched with the yaw rate γ due to the influence of the second term K ₂ γ in the above equation 8. Left and right rear wheels RW
1 and RW2 are steered in the direction in which the yaw rate γ is suppressed, so that the running stability of the vehicle during high speed running becomes good.

Further, if the traveling road surface is in the left cant state, the output value Z of the neural network operation is "0" ...
The value is between "1", and the correction value Az is determined to be a negative value as shown in FIG. As a result, the target rear wheel steering angle θ
r * is corrected to the negative side due to the influence of the third term Az in the equation (8). Therefore, in this case, the left and right rear wheels RW1, RW2
Is corrected and steered to the left and acts so as to cancel the left cant state to realize the same steering wheel operation state as when the vehicle is traveling on a flat road. If the road surface is in the right cant state, the output value Z of the neural network calculation
Becomes a value between "0" and "-1", and the correction value Az is determined to be a positive value as shown in FIG. As a result, the target rear wheel steering angle θr * is corrected to the positive side due to the influence of the third term Az in the equation (8). Therefore, in this case, the left and right rear wheels R
W1 and RW2 are corrected and steered in the right direction, and act to cancel the right cant state to realize the same steering wheel operation state as when the vehicle is traveling on a flat road.

B. Second Embodiment Next, a second embodiment in which the vehicle road condition recognition device according to the present invention is applied to a vehicle whose height can be adjusted will be described with reference to the drawings. FIG. 8 schematically shows this vehicle. There is. This vehicle includes suspension devices 51 to 54 that support the vehicle body at the positions of the respective wheels FW1, FW2, RW1, RW2 and that can adjust the height (vehicle height) of the vehicle body, and each suspension device 51. An electric control device 60 for electrically controlling the devices 54 to 54 is provided.

The electric control unit 60 comprises a vehicle speed sensor 61, a lateral acceleration sensor 62, a steering wheel steering angle sensor 63, a steering wheel torque sensor 64 and vehicle height sensors 65 to 68. The vehicle speed sensor 61, the lateral acceleration sensor 62, the steering wheel steering angle sensor 63, and the steering wheel torque sensor 64 have the same configurations as those in the first embodiment. Vehicle height sensor 65-
Reference numeral 68 is incorporated in each of the suspension devices 51 to 54, detects the vehicle heights H _FL , H _FR , H _RL , and H _RR at the positions of the wheels FW1, FW2, RW1, RW2, respectively, and the vehicle heights H _FL , H A detection signal indicating _FR , H _RL , H _RR is output.

These sensors 61 to 68 are connected to the microcomputer 71. The microcomputer 71 has a CPU and an RO as in the case of the first embodiment.
A program including an M, a RAM, an I / O, a timer and the like, which is stored in the ROM every predetermined time and corresponds to the flowchart of FIG. 9, is executed to detect the cant state of the traveling road surface by a neural network operation. The vehicle heights H _FL , H _FR , H _RL and H _RR at the positions of the respective wheels FW1, FW2, RW1 and RW2 are adjusted according to the detection result. Driving circuits 72 to 75 are connected to the microcomputer 71, and the circuits 72 to 75 drive and control the suspension devices 51 to 54 in accordance with control signals from the computer 71.

Next, the operation of the second embodiment constructed as above will be described. Microcomputer 7 while the vehicle is running
1 executes the program including steps 200 to 205 of FIG. 9 at predetermined time intervals by the action of the built-in timer. In this program, after the start in step 200, the vehicle speed V, lateral acceleration G, steering wheel from the vehicle speed sensor 61, lateral acceleration sensor 62, steering wheel steering angle sensor 63, steering wheel torque sensor 64, and vehicle height sensors 65 to 68 in step 201. The detection signals representing the steering angle θf, the steering wheel torque T and the vehicle heights H _FL , H _FR , H _RL and H _RR are input. Next, in step 202, the same "neural network calculation routine" of FIG. 3 as in the first embodiment is executed,
An output value Z representing the cant state of the road is calculated.

After the completion of this "neural network operation routine", in step 203 of FIG. 9, the left wheel FW1 corresponding to the output value Z is referred to by referring to the map (FIG. 10) provided in the ROM of the microcomputer 77. R
By determining the target vehicle heights H _FL *, H _RL * for W1 and referring to the map (FIG. 11) provided in the ROM, the target vehicle heights H _FR *, H _RR for the right wheels FW2, RW2 are obtained. * Decide. Next, at step 205, the detected vehicle heights H _FL , H _FR , H _RL , and H _RR are respectively detected from the determined target vehicle heights H _FL *, H _FR *, H _RL *, and H _RR *. Each deviation H _FL * − subtracted
Control signals representing H _FL , H _FR * -H _FR , H _RL * -H _RL , and H _RR * -H _RR are output to the drive circuits 72 to 75, respectively. The drive circuits 72 to 75 output drive signals corresponding to the control signals to the suspension devices 51 to 54, and the suspension devices 51 to 54 output the deviations H _FL * −H _FL , H _FR * −.
_{H FR, H RL * -H RL} , H RR * -H RR amounts only the wheels corresponding to FW1, FW2, RW1, displacing the vehicle height corresponding to RW2 position. As a result, the vehicle height at each position is set to the target vehicle height H _FL *, H _FR *, H _RL *, H _RR *.

In this case, the traveling road surface is in the left cant state and the output value Z of the neural network operation is "0".
If the value is between "1", the target vehicle heights H _FL *, H _RL * at the positions of the left wheels FW1, RW1 are determined to be positive values and the right wheels FW2, RW2 are set as shown in FIG. The target vehicle height H _FR *, H _RR * at the position is determined to be a negative value, as shown in FIG. Therefore, the left part of the vehicle body is controlled to be raised and the right part of the vehicle body is controlled to be lowered, so that the vehicle body of the vehicle on the road surface in the left cant state is kept horizontal. Further, when the traveling road surface is in the right cant state and the output value Z of the neural network calculation becomes a value between "0" and "-1", the left wheel FW
1, the target vehicle heights H _FL *, H _RL * at the RW1 position are determined to be negative values as shown in FIG. 10, and the right wheel FW2,
The target vehicle heights H _FR * and H _RR * at the RW2 position are determined to be positive values, as shown in FIG. Therefore, while the left part of the vehicle body is controlled to descend, the right part of the vehicle body is controlled to rise,
The vehicle body of the vehicle on the right cant is kept horizontal.

C. Modifications In the first and second embodiments, the neural network calculation is performed by the microcomputers 37 and 71.
Although it is realized as software by the program processing of, the neural network operation may be configured by a hard circuit as shown in the functional block diagram of FIG.

[Brief description of drawings]

FIG. 1 is an overall schematic diagram of a four-wheel steering vehicle to which a vehicle road condition recognition device according to a first embodiment of the present invention is applied.

2 is a flowchart of a program executed by the microcomputer of FIG.

FIG. 3 is a flowchart showing details of a “neural network calculation routine” of FIG.

FIG. 4 is a change characteristic diagram of a coefficient K ₁ with respect to a vehicle speed V.

FIG. 5 is a change characteristic diagram of a coefficient K ₂ with respect to a vehicle speed V.

[FIG. 6] Output value Z by neural network calculation
It is a change characteristic view of the correction value Az with respect to.

7 is a functional block diagram equivalently showing a neural network operation executed by the microcomputer of FIG.

FIG. 8 is an overall schematic view of a vehicle including a suspension device to which the vehicle road condition recognition device according to the second embodiment of the present invention is applied.

9 is a flowchart of a program executed by the microcomputer of FIG.

FIG. 10 is a change characteristic diagram of left wheel target vehicle height values H _{FL *,} H _{RL *} with respect to an output value Z calculated by a neural network calculation.

FIG. 11 is a change characteristic diagram of target vehicle height values H _{FR *,} H _{RR *} for the right wheel with respect to an output value Z calculated by a neural network calculation.

[Explanation of symbols]

10 ... Front wheel steering mechanism, 20 ... Rear wheel steering mechanism, 30, 60
... Electric control device, 31,61 ... Vehicle speed sensor, 32,62
... lateral acceleration sensor, 33, 63 ... steering wheel steering angle sensor,
34, 64 ... Steering wheel torque sensor, 35 ... Yaw rate sensor, 36 ... Rear wheel steering angle sensor, 37, 71 ... Microcomputer, NN ... Neural network, 45 ...
Input layer, 46 ... Intermediate layer, 47 ... Output layer, 51-54 ... Suspension device, 65-68 ... Vehicle height sensor.

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁵ Identification code Office reference number FI technical display location B62D 119: 00 133: 00

Claims (1)

[Claims] 1. A vehicle speed sensor for detecting a vehicle speed, a lateral acceleration sensor for detecting a lateral acceleration of a vehicle, a steering wheel sensor for detecting an operating state of a steering wheel, and a vehicle speed, a lateral acceleration of a vehicle and an operating state of a steering wheel. I remember the result of learning the cant state of the road surface,
Neural network means for determining the cant state of the road surface by a neural network calculation using the learning result based on the vehicle speed detected by each sensor, the lateral acceleration of the vehicle, and the operating state of the steering wheel. Vehicle road condition recognition device.