System and method for generating and controlling driving paths in autonomous vehicle

1. A method of generating and controlling a driving path in an autonomous vehicle, the method comprising:

acquiring sensing data for an outside of the autonomous vehicle through at least one sensing module coupled to the autonomous vehicle;

generating a plurality of driving paths corresponding to the driving intention information by using the sensing data, driving intention information, and target speed information as input data for a first artificial neural network;

evaluating a degree of risk of each of the plurality of driving paths from object information recognized on the basis of the sensing information;

determining steering angle information for the autonomous vehicle by using the driving path having a lowest degree of risk and the target speed information as input data for a second artificial neural network; and

controlling a steering angle of the autonomous vehicle on the basis of the determined steering angle information,

wherein the generating of the plurality of driving paths corresponding to the driving intention information includes:

extracting a feature vector using the sensing data as input data of a convolution neural network (CNN)-based feature extraction algorithm; and

while using the feature vector, the target speed information, the driving intention information, and a noise vector as input data of a recurrent neural network (RNN)-based path generation algorithm, repeatedly varying the noise vector to generate the plurality of driving paths.

2. The method of claim 1 , further comprising:

collecting pieces of training-purpose sensing data for the first and second artificial neural networks during actual driving of a user from the at least one sensing module of the autonomous vehicle or a learning data collecting vehicle;

assigning each of the pieces of collected training-purpose sensing data with a corresponding piece of driving state information of the autonomous vehicle;

constructing training data using each of the pieces of training-purpose sensing data assigned with the driving state information; and

training the first and second artificial neural networks on the basis of the training data,

wherein the training data includes the training-purpose sensing data, speed information, true value driving intention information, true value steering angle information, and actual path information for a time t.

3. The method of claim 2 , wherein the constructing of the training data using each piece of training-purpose sensing data assigned with the driving state information includes:

calculating a transformation matrix using a yaw value, a pitch value, a roll value, a height value, and a latitude and longitude value of the autonomous vehicle or the learning data collecting vehicle acquired from the training-purpose sensing data;

converting a global position of the autonomous vehicle or the learning data collecting vehicle after the time t into an egocentric coordinate system using the transformation matrix; and

acquiring the actual path information on the basis of the converted egocentric coordinate system.

4. The method of claim 3 , wherein the constructing of the training data using each of the pieces of training-purpose sensing assigned with the driving state information includes:

for each position in the actual path information, detecting a corresponding shortest position for the autonomous vehicle or the learning data collecting vehicle in each piece of the training-purpose sensing data assigned with the driving state information;

acquiring the driving state information corresponding to the detected position as the driving intention information in the actual path information; and

selecting the true value driving intention information that represents the actual path information among pieces of driving intention information of the actual path information.

5. The method of claim 4 , wherein the selecting of the true value driving intention information that represents the actual path information among the pieces of driving intention information of the actual path information includes:

selecting a maximum number of pieces of driving intention information among L pieces of driving intention information in the actual path information or driving intention information at a last position as the true value driving intention information.

6. The method of claim 2 , wherein the driving state information includes information corresponding to the driving intention information, and

the driving intention information and the driving state information include one of straight driving, a left turn, a right turn, a U-turn, a lane change to left, a lane change to right, avoidance, a left direction, and a right direction.

7. The method of claim 2 , further comprising, wherein the at least one sensing module configured to sense the training data includes a plurality of sensing modules, performing time-series synchronization on the training-purpose sensing data acquired from each of the plurality of sensing modules.

8. The method of claim 2 , wherein the training of the first and second artificial neural networks on the basis of the training data includes training the first artificial neural network to reduce an error between a plurality of driving paths, which are pieces of output data of the first artificial neural network, and the actual path information.

9. The method of claim 2 , wherein the training of the first and second artificial neural networks on the basis of the training data includes:

setting the target speed information as an initial hidden state vector of the second artificial neural network based on a recurrent neural network (RNN);

sequentially inputting each position in the actual path information and a hidden state vector corresponding to each position to the second artificial neural network;

acquiring a hidden state vector at an end of the second artificial neural network as steering angle information; and

training the second artificial neural network to reduce an error between the acquired steering angle information and the true value steering angle information.

10. The method of claim 1 , wherein the repeatedly varying of the noise vector to generate the plurality of driving paths includes:

inputting a I st position in a driving path and a hidden state vector corresponding to the I st position as input data of the RNN-based path generation algorithm;

acquiring an I+1 st position in the driving path on the basis of an I+1 st hidden state vector acquired as an output for the input and the feature vector; and

repeatedly varying the noise vector with respect to the driving path to generate the plurality of driving paths.

11. A system for generating and controlling a driving path in an autonomous vehicle, the system comprising:

at least one sensing module coupled to the autonomous vehicle and configured to acquire sensing data for an outside of the autonomous vehicle;

a memory in which a program for generating a plurality of driving paths for the autonomous vehicle on the basis of the sensing data and determining steering angle information corresponding to one of the generated plurality of driving paths is stored; and

a processor configured to execute the program stored in the memory,

wherein the processor executes the program to:

use the sensing data, driving intention information set and target speed information as input data for a first artificial neural network to generate a plurality of driving paths corresponding to the driving intention information; and

evaluate a degree of risk of each of the plurality of driving paths from object information recognized on the basis of the sensing information; and

use the driving path having a lowest degree of risk and the target speed information as input data for a second artificial neural network to determine steering angle information for the autonomous vehicle to control a steering angle of the autonomous vehicle,

wherein the processor is configured to:

extract a feature vector by using the sensing data as input data of a convolution neural network (CNN)-based feature extraction algorithm; and

while using the feature vector, the target speed information, the driving intention information, and a noise vector as input data of a recurrent neural network (RNN)-based path generation algorithm, repeatedly vary the noise vector to generate the plurality of driving paths.

12. The system of claim 11 , wherein the processor is configured to:

input a I st position in a driving path and a hidden state vector corresponding to the I st position as input data of the RNN-based path generation algorithm;

acquire an I+1 st position in the driving path on the basis of an I+1 st hidden state vector acquired as an output for the input and the feature vector; and

repeatedly vary the noise vector with respect to the driving path to generate the plurality of driving paths.

13. The system of claim 11 , wherein the processor is configured to:

collect pieces of training-purpose sensing data for the first and second artificial neural networks during actual driving of a user from the sensing module of the autonomous vehicle or a learning data collecting vehicle;

assign each of the pieces of collected training-purpose sensing data with a corresponding piece of driving state information of the autonomous vehicle;

construct training data using each of the pieces of training-purpose sensing data assigned with the driving state information; and

train the first and second artificial neural networks on the basis of the training data,

wherein the training data includes the training-purpose sensing data, speed information, true value driving intention information, true value steering angle information, and actual path information for a time t.

14. The system of claim 13 , wherein the processor is configured to:

calculate a transformation matrix using a yaw value, a pitch value, a roll value, a height value, and a latitude and longitude value of the autonomous vehicle or the learning data collecting vehicle acquired from the training-purpose sensing data;

convert a global position of the autonomous vehicle or the learning data collecting vehicle after the time t into an egocentric coordinate system using the transformation matrix; and

acquire the actual path information on the basis of the converted egocentric coordinate system.

15. The system of claim 14 , wherein the processor is configured to:

for each position in the actual path information, detect a corresponding shortest position for the autonomous vehicle or the learning data collecting vehicle in each piece of the training-purpose sensing data assigned with the driving state information;

acquire the driving state information corresponding to the detected position as the driving intention information in the actual path information; and

select the true value driving intention information that represents the actual path information among pieces of driving intention information of the actual path information.

16. The system of claim 15 , wherein the processor is configured to select a maximum number of pieces of driving intention information among L pieces of driving intention information in the actual path information or driving intention information at a last position as the true value driving intention information.

17. A system for generating and controlling a driving path in an autonomous vehicle, the system comprising:

at least one sensing module coupled to the autonomous vehicle and configured to acquire sensing data for an outside of the autonomous vehicle;

a memory in which a program for generating a plurality of driving paths for the autonomous vehicle on the basis of the sensing data and determining steering angle information corresponding to one of the generated plurality of driving paths is stored; and

a processor configured to execute the program stored in the memory,

wherein the processor executes the program to:

use the sensing data, driving intention information set and target speed information as input data for a first artificial neural network to generate a plurality of driving paths corresponding to the driving intention information; and

evaluate a degree of risk of each of the plurality of driving paths from object information recognized on the basis of the sensing information; and

use the driving path having a lowest degree of risk and the target speed information as input data for a second artificial neural network to determine steering angle information for the autonomous vehicle to control a steering angle of the autonomous vehicle,

wherein the processor is configured to:

set the target speed information as an initial hidden state vector of the second artificial neural network based on a recurrent neural network (RNN), sequentially input each position in an actual path information and a hidden state vector corresponding to each position to the second artificial neural network, acquire a hidden state vector at an end of the second artificial neural network as steering angle information, and train the second artificial neural network to reduce an error between the acquired steering angle information and a true value steering angle information.