Maneuver를 통한 차량의 차선 단위 경로 예측

Abstract

For the path planning of autonomous vehicle, the path prediction of neighboring vehicles should be done. Since there are many possibilities in vehicle path prediction when given the same situation, the prediction should be given in various ways. This study suggests a deep learning network that predicts the lane-level trajectory of target vehicle considering surrounding vehicles. The proposed network employed Encoder-Decoder structure and applied Convolutional Social Pooling method to consider the interaction with neighboring vehicles. Also, after the additional network estimates the maneuver, the maneuver is given as a condition to Decoder. Therefore, the network can generate the prediction based on the maneuver. The proposed network has been verified through highD(Highway Drone) open dataset.