JPH07110712A - Steering angle controller - Google Patents

Abstract

PURPOSE:To stably perform the control in accordance with the change of the curvature of a road or the vehicle speed and to reduce the calculation volume and the memory capacity. CONSTITUTION:A means 10 which measures a vehicle speed V and a device 30 which measures a vehicle yaw angle (y), a road curvature Cr, and an extent epsilonor transverse deviation as the extent of positional deviation of the vehicle from a reference line of the road are provided. A device 24 multiplies the extent epsilonof transverse deviation and its time differential value by gains and adds multiplication results and outputs the result. A device 22 multiplies the road curvature Cr by a gain and outputs the result. A device 18 outputs a learning signal which determines the gain to be used in a device 20 and is defined by the difference between the command curvature calculated by a means 18A and the running curvature calculated by a means 18B. The device 20 consists of a neural network and uses the learning signal to calculate and output the extent of compensation which reduces the extent epsilon of transverse deviation which is generated because proper gains cannot be set in the device 24.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a steering angle control device,
More specifically, the present invention relates to a steering angle control device that controls a steering angle of a traveling body such as an automatic piloting device that acts on behalf of a human being, an emergency collision avoidance device, and a mobile robot such as an automatic guided vehicle.

[0002]

2. Description of the Related Art Conventionally, a lateral deviation amount with respect to a traveling direction of a vehicle is determined based on a road edge or a center line based on information obtained by image-processing an image of a road (embedded in the ground. The amount of lateral deviation can be similarly obtained by using an electric wire or magnet), and the lateral deviation amount, the time differential value of the lateral deviation amount, or the time integration value of the lateral deviation amount is multiplied by the gain and added. A method is known in which the steering amount is determined by adding the amount (feedback control amount by the lateral deviation amount) to the amount obtained by multiplying the road curvature by the gain (feedforward control amount by the road curvature) to control the steering angle. .

Further, a method of performing multiple regression analysis from the steering amount actually steered by the driver on the basis of the lateral distance information with respect to the traveling direction from the road edge, and determining the steering amount using the obtained regression equation. Is known (Japanese Patent Application Laid-Open No. 4-30
4502). In this method, it is possible to cope with changes in road curvature and vehicle speed by updating the coefficient of the regression equation during traveling.

[0004]

However, in the conventional method of combining the feedback control amount based on the lateral deviation amount and the feedforward control amount based on the road curvature,
Since it is necessary to obtain an appropriate gain according to the magnitude of the road curvature and the vehicle speed, it is necessary to conduct a traveling experiment combining the road curvature and the vehicle speed, which requires a great deal of effort to adjust the gain, which is not realistic.

Further, in the method using the conventional regression equation, in order to adapt to the change of the road curvature and the vehicle speed, it is necessary to perform the multiple regression analysis during traveling, so that the calculation cost is increased and the obtained result is obtained. Since the regression coefficients are tabulated, the memory capacity also becomes large. Further, in the method using the regression equation, since the feedback control is performed, the control may become unstable depending on the selection of the coefficient.

The present invention has been made to solve the above-mentioned problems, and enables stable control adaptable to changes in the traveling speed, such as the curvature of the traveling path such as road curvature and the vehicle speed, and the amount of calculation and memory. An object of the present invention is to provide a steering angle control device that can reduce the capacity.

[0007]

In order to achieve the above-mentioned object, the present invention, as shown in FIG. 1, includes a positional deviation amount of a traveling body from a reference line in a traveling road on which the traveling body travels and a curvature of the traveling road. And a measuring means a for measuring the state quantity related to the lateral motion of the traveling body, a speed measuring means b for measuring the traveling speed of the traveling body, and a target for calculating a target steering angle based on the curvature of the traveling road and the positional deviation amount. A value calculating means c, a command curvature calculating means d for calculating a command curvature based on the curvature of the traveling path and a positional deviation amount, and a curvature of the traveling locus based on the curvature of the traveling path, the traveling speed and the state quantity. Running curvature calculation means e
And a learning signal calculation means f for outputting the deviation between the command curvature and the curvature of the traveling locus as a learning signal, and a neural network, and inputs the traveling speed and the curvature of the traveling path, and performs an arithmetic operation using the learning signal. , Compensation amount calculation means g for calculating a compensation amount for feedforward control
And a control means h for controlling the traveling body based on the target steering angle and the compensation amount.

[0008]

The measuring means a of the present invention measures the amount of positional deviation of the traveling body from the reference line in the traveling path on which the traveling body travels, the curvature of the traveling path, and the state quantity related to the lateral movement of the traveling body. As the measuring means a, a measuring device that includes an image device and a computing device and measures the amount of positional deviation, the curvature of the traveling path, and the state amount related to the lateral motion of the traveling body by image processing and computation may be used. Further, a separate measuring device of a position deviation measuring means a1 for measuring a position deviation amount, a state quantity measuring means a2 for measuring a state quantity related to the lateral motion of the traveling body, and a traveling road curvature measuring means a3 for measuring a curvature of the traveling road is provided. You may use. Further, as the state quantity relating to the lateral movement of the vehicle, a yaw angle, a yaw rate, a lateral displacement, a lateral speed, a lateral acceleration, etc. can be used.

The speed measuring means b measures the running speed of the running body. The target value calculation means c calculates the target steering angle based on the curvature of the road and the amount of positional deviation. The command curvature calculation means d calculates the command curvature based on the curvature of the road and the amount of positional deviation. The traveling curvature calculation means e is a curvature of the traveling path,
The curvature of the traveling locus is calculated based on the traveling speed and the state quantity. The learning signal calculation means f outputs the deviation between the command curvature and the curvature of the traveling locus as a learning signal. The compensation amount calculation means g is composed of a neural network, receives the traveling speed and the curvature of the traveling path, and calculates the compensation amount for feedforward control by calculation using the learning signal. Then, the control means h controls the traveling body based on the target steering angle and the compensation amount.

Since learning is performed in this manner, stable control adapted to changes in the traveling speed and the curvature of the traveling path becomes possible. Further, since the neural network is used, the calculation amount and memory capacity can be reduced.

[0011]

As described above, according to the present invention,
Even if the target steering angle is not properly set, the optimum steering compensation amount can be obtained by learning by adapting to changes in the curvature of the road and the traveling speed, and stable control can be performed.
Since the feedforward compensation amount is obtained by the neural network, it is possible to reduce the calculation amount and the memory capacity.

[0012]

Embodiments of the present invention will now be described in detail with reference to the drawings. In the present embodiment, the present invention is applied to a steering angle control device that controls the steering angle of a vehicle.

As shown in FIG. 2, the steering angle control system of the present embodiment comprises a vehicle speed measuring means 10 for measuring a vehicle speed V, a vehicle body yaw angle y, a road curvature Cr and a position deviation of the vehicle from a reference line on the road. And a measuring device 30 for measuring a lateral deviation amount ε which is a quantity. This measuring device 30 processes the image captured by the image capturing device and the image capturing device that captures the road ahead of the vehicle in the traveling direction, and based on the following equation (1), the vehicle body yaw angle y, the road curvature Cr, and the lateral deviation. Quantity ε
And a computing device that computes. When this measuring device 30 is represented by functional blocks, a vehicle body yaw angle measuring means 12 for measuring a vehicle body yaw angle y, a road curvature measuring means 14 for measuring a road curvature Cr, and a lateral deviation amount measuring means 16 for measuring a lateral deviation amount ε. Can be expressed as

The measuring device 30 corresponds to a vehicle body yaw angle measuring means 12, a road curvature measuring means 14 and a lateral deviation amount measuring means for respectively measuring the vehicle body yaw angle y, the road curvature Cr and the lateral deviation amount ε. A separate meter may be used.

As shown in the geometrical relationship diagram between the vehicle and the road in FIG. 3, the vehicle body yaw angle y is an angle formed by the vehicle traveling direction and the vehicle direction, and the lateral deviation amount ε is the center of the road as a reference line. The road curvature Cr indicates the distance between the vehicle center and the vehicle center (the position where the measuring device is attached), and the road curvature Cr indicates the vehicle center curvature.
Note that a road edge or the like can be used as the reference line.

Further, the steering angle control system of this embodiment carries out feedforward control by means of a proportional derivative computing unit 24 as a feedback controller for performing feedback control by the lateral deviation amount ε and by road curvature Cr as the curvature of the running road. Proportional computing device 2 as a feedforward controller
2, a feedforward compensation amount calculation device 20, and a learning signal generator 18 for generating a learning signal. In the present embodiment, the feedback control based on the lateral deviation amount is the proportional derivative control using the proportional derivative calculating device 24. Also,
The feedforward compensation amount calculation device 20 is composed of a three-layer perceptron type neural network simulating a human neural circuit.

FIG. 5 shows the structure of a three-layer perceptron type neural network. This neural network is composed of a plurality of input layers, a plurality of hidden layers (middle layers), and one output layer. The road curvature Cr and the product of the road curvature Cr and the vehicle speed V are input to each unit of the input layer. The road curvature Cr and the product of the road curvature Cr and the vehicle speed V are input to the straight road (road curvature Cr =
This is to prevent the output of the neural network from changing due to the change in vehicle speed during traveling in 0). The input / output function of each unit of the hidden layer is a sigmoid function saturated with -1 and 1 (for example, f (x) = [1-exp (-x)] /
[1 + exp (-x)]). The input / output function of the unit in the output layer is a linear function.

The learning signal generator 18 operates the vehicle based on the road curvature Cr, the vehicle speed V, and the vehicle body yaw angle y, and the vehicle curvature Cr, the vehicle curvature V, and the vehicle body yaw angle y. A traveling curvature calculating means 18B for calculating the curvature of the trajectory (traveling curvature) is provided. The learning signal generator 18 includes a vehicle speed measuring unit 10, a vehicle body yaw angle measuring unit 12 of the measuring unit 30, and a road curvature measuring unit 1.
4 and the lateral deviation amount measuring means 16 are connected. Vehicle curvature measuring means 10 and road curvature measuring means 14 of the measuring device 30
Are connected to the feedforward compensation amount calculation device 20.

The road curvature measuring means 14 of the measuring device 30 is
The lateral deviation amount measuring means 16 of the measuring device 30 is connected to the proportional operation device 22, and is connected to the proportional derivative operation device 24. The proportional calculation device 22 and the proportional differential calculation device 24 are connected to the adder 26, and the adder 26 and the feedforward compensation amount calculation device 20 output the steering amount signal.
8 is connected. Then, the steering angle of the vehicle is controlled by the steering amount signal.

Next, the operation of this embodiment will be described. First, FIG. 4 schematically shows a road image obtained by the image processing device constituting the measuring device 30. L1, L2, and L3 each represent an actual distance (not a length on the image) from the vehicle center. Further, ε1, ε2, ε3 are real distances L1, L2, L3.
Is the actual distance from the road edge.

Lateral deviation amount ε, vehicle body yaw angle y and road curvature C
r is L1, L2, L3, ε obtained by image processing
It is calculated by the following equation based on 1, ε2, ε3.

[0022]

[Equation 1]

In the proportional derivative calculating device 24, the lateral deviation amount ε is input, and the time differential value of the input lateral deviation amount ε is calculated, and the lateral deviation amount ε and the time differential value of the lateral deviation amount ε are respectively calculated. It is multiplied by an appropriate gain, added together, and output.
The output outpd of the proportional / differential operation device 24 is expressed by the following equation.

Outpd = Kp · ε + Kd · dε / dt (2) Here, d / dt represents a time derivative, and Kp and Kd represent constant gains.

If an appropriate gain is set, the output value of the proportional / differential arithmetic unit 24 becomes the steering amount so that the lateral deviation amount ε becomes 0, that is, the vehicle runs on the center of the road. .

In the proportional calculation device 22, the road curvature Cr is input, multiplied by an appropriate gain, and output. The output outc of the proportional calculation device 22 is given by the following equation.

Outc = Kc · Cr (3) Here, Kc is a constant gain.

If the gain is properly set, the output value of the proportional calculation device 22 becomes a steering amount peculiar to the vehicle according to the curvature of the road.

Therefore, ou output from the adder 26
tpd + outc is a target steering angle for the vehicle to travel in the center of the road according to the curvature of the road.

Feedforward compensation amount calculation device 20
Is a lateral deviation amount ε that occurs because an appropriate gain cannot be set in the proportional operation device 22 and the proportional derivative operation device 24.
Outputs the compensation amount for reducing.

This feedforward compensation amount calculation device 2
The output outnn of 0 is represented by the function F as follows.

Outnn = F (V, Cr, W) (4) Here, W is the weight (weight) of the neural network adjusted by the learning signal output from the learning signal generator 18.

The final steering amount δ output from the adder 28 is given by the following expression (5) from the expressions (2), (3) and (4).

Δ = outpd + outc + outnn (5) The learning signal for determining the gain (weight because it is a neural network) used in the feedforward compensation amount calculation device 20 is calculated by the command curvature calculation means 18A. It is defined by the difference between the command curvature and the traveling curvature calculated by the traveling curvature calculating means 18B. The learning signal error will be described.

First, since the command curvature is obtained by adding the road curvature Cr to a value obtained by multiplying the lateral deviation amount ε and the time differential value of the lateral deviation amount ε by an appropriate gain and adding them, the command curvature Ccom is It is represented by a formula.

Ccom = Cr + K1 · ε + K2 · dε / dt (6) Here, K1 and K2 are constant gains.

For the running curvature, add the average value of the road curvature Cr during one control cycle and the value obtained by dividing the amount of change in the vehicle body yaw angle y during one control cycle by the running distance of one control cycle. Required by. Therefore, the running curvature Ctra is
When the average value of the road curvature Cr is Crmean, the minute traveling distance is ΔL, and the variation amount of the vehicle body yaw angle y when traveling the minute traveling distance ΔL is Δy, it is expressed by the following equation. This minute distance ΔL is represented by the product of the vehicle speed V and one control cycle.

Ctra = Crmean + Δy / ΔL (7) The learning signal error is the command curvature Ccom and the traveling curvature C.
Since it is a deviation from tra, the following equation is obtained.

Error = Ccom-Ctra (8) Road curvature Cr input to the input layer of the neural network that constitutes the feedforward compensation amount calculation device 20.
The respective signals of the product of the road curvature Cr and the vehicle speed V propagate through the hidden layer toward the output layer. At this time, the output X _i of the i-th unit of the hidden layer and the output Y _i of the i-th unit of the output layer are respectively expressed by the following equations.

X _i = f (Σ _j W _ij × I _j ) ... (9) Y _i = f (Σ _j Z _ij × X _j ) ... (10) where f is each hidden layer Unit input / output function (f
(X) = [1-exp (-x)] / [1 + exp (-
x)]), W _ij is the weight between the i-th unit of the hidden layer and the j-th unit of the input layer, I _j is the output of the j-th unit of the input layer, and Z _ij is the i-th unit of the output layer and the hidden layer. This is the weight between the jth unit.

A backpropagation algorithm is used to update the weight of each layer of the neural network. One half of the square of the learning signal error output from the learning signal generator 18 is E (= er
If ror ^2/2) and defined by the steepest descent method (one of the optimization method), weight updating value each, as follows.

ΔZ _ij = −η · ∂E / ∂Z _ij (11) ΔW _ij = −η · ∂E / ∂W _ij (12) where η is a positive constant learning It is a coefficient. The weight update values are specifically expressed as follows.

ΔZ _ij = η · error · X _j (13) ΔW _ij = η · (Σ _k Z _ki · error) (1-X _i ² ) · I _j / 2 (14) Here Where k is taken for all output units.

In the feedforward compensation amount calculation device 20, the weight correction amount is obtained and the weight is updated. If the value of E converges to the minimum value by updating the weights, the command curvature and the running curvature match, and the vehicle will run on the center of the road.
The neural network is trained until is the minimum value.

Next, the result of verifying the effect of learning by computer simulation will be described. However, in the computer simulation of the present embodiment, the above (6)
~ (8) in place of the equation (6) '~ (8)' below, where t is the sampling time and ΔT is the sampling interval (1 control cycle).

Ccom [t] = Cr [t] + K1 · ε [t] + K2 (ε [t] −ε [t-1]) / ΔT (6) ′ Ctra [t] = (Cr [t ] + Cr [t−1]) / 2+ (y [t] −y [t−1]) / V · ΔT (7) ′ error [t] = Ccom [t−1] −Ctra [t ] (8) 'As shown in FIG. 6, the vehicle traveled on a circular orbit with a radius of 50 m and made a right turn. Further, the maximum value of the vehicle speed V was set to 40 km / h (about 11.1 m / s), and the sampling interval was set to 0.4 seconds. Also, L1, L2, L
3 is 10 m, 15 m, 20 m, Kp, Kd, Kc are -0.265, -0.583, 53.0, K1, K, respectively.
2 was set to -0.005 and -0.011 respectively. further,
The number of units in the input layer of the neural network is 2, the number of units in the hidden layer is 6, the number of units in the output layer is 1, and the learning coefficient η is 0.6.

FIG. 7 shows the result of computer simulation. In FIG. 7, the vertical axis represents the deviation from the road center,
The horizontal axis is the running time. When the sign of the vertical axis is positive, it is located right from the center with respect to the traveling direction, and when it is negative, it is located left. In the example from the start (0.00) to 150 seconds, the learning of the neural network is stopped. That is, the results of the conventional control, the feedback control by the lateral deviation amount ε, and the feedforward control by the road curvature Cr are shown.

As can be seen from FIG. 7, this vehicle runs about 0.8 m off the center of the road. After 150 seconds, the lateral deviation amount ε becomes 0.1 m due to the learning by the neural network. As described above, the deviation amount remains after the learning because of the calculation error by the computer simulation.

As described above, according to the present embodiment, the deviation from the road center can be reduced and the control performance can be improved by applying the learning by the neural network to the inconvenience caused by the conventional control. . In addition, the effect of applying the learning by the neural network was similarly observed in the change of the vehicle speed V. Further, after the learning is performed in this way, the learning is turned off, that is, a straight road or a left turn is performed with the learning coefficient η set to 0, and the generalization ability of the neural network is confirmed. The generalization ability eliminates the need for extra learning and reduces the calculation cost.

Therefore, even if the gain setting in the proportional computing device and the proportional differentiating device is improper, the optimum steering compensation amount can be obtained by learning by adapting to changes in road curvature and vehicle speed. Further, it can be expected that the neural network that realizes the feedforward compensation amount calculation device can reduce the memory capacity and omit the extra learning due to the generalization property, thereby reducing the calculation cost.

In the above embodiment, an example in which the vehicle body yaw angle, the road curvature and the lateral deviation amount are obtained by image processing has been described. However, the present invention is not limited to this, and other methods such as an electromagnetic induction method may be used. The vehicle body yaw angle, the road curvature and the lateral deviation amount may be obtained.

Also, an example in which the feedback control amount is calculated by proportional differentiation has been described, but the feedback control amount may be obtained by proportional control or proportional integral derivative control. Further, the command curvature calculation means may further add the product of the time integral value of the lateral deviation amount and the gain.

[Brief description of drawings]

FIG. 1 is a block diagram corresponding to the claims of the present invention.

FIG. 2 is a diagram showing a geometrical relationship between a vehicle and a road.

FIG. 3 is a block diagram of an embodiment of the present invention.

FIG. 4 is a schematic diagram of a road image used in this embodiment.

FIG. 5 is a configuration diagram of a three-layer perceptron type neural network according to the present embodiment.

FIG. 6 is a diagram showing a vehicle traveling course of the embodiment.

FIG. 7 is a graph of a computer simulation result used in an example.

[Explanation of symbols]

 10 vehicle speed measuring means 30 measuring device 12 vehicle body yaw angle measuring means 14 road curvature measuring means 16 lateral deviation amount measuring means

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Reference number within the agency FI Technical display location B62D 111: 00 137: 00 (72) Inventor Keiji Aoki 1 Toyota-cho, Toyota-shi, Aichi Toyota Motor Vehicle Incorporated (72) Inventor Akihide Tachibana, Toyota City, Aichi Prefecture 1 Toyota Toyota Motor Corporation (72) Inventor Toshihiko Suzuki 1, Toyota Town, Aichi Prefecture Toyota Motor Corporation (72) Inventor Tatsuaki Yokoyama 1 Toyota Town, Toyota City, Aichi Toyota Motor Co., Ltd.

Claims (1)

[Claims] 1. A measuring means for measuring a positional deviation amount of a traveling body from a reference line on a traveling path on which the traveling body travels, a curvature of the traveling road, and a state quantity related to a lateral motion of the traveling body, and a traveling speed of the traveling body. A speed measuring means for measuring a target value, a target value calculating means for calculating a target steering angle based on a curvature and a positional deviation amount of a traveling road, and a command curvature calculation for calculating a command curvature based on a curvature and a positional deviation amount of the traveling road Means, a running curvature calculating means for calculating the curvature of the running trajectory based on the curvature of the running path, the running speed and the state quantity; and a learning signal computing means for outputting a deviation between the command curvature and the curvature of the running trajectory as a learning signal. Compensation amount calculator for calculating a compensation amount for feedforward control by a calculation using the learning signal, which is composed of a neural network and inputs the traveling speed and the curvature of the traveling path. A steering angle control device including: a step; and a control unit that controls a traveling body based on the target steering angle and the compensation amount.