End-to-end autonomous vehicle lateral control with deep learning

Ferenc, Csanád; Zoldy, Mate [Zöldy, Máté (Belsőégésű motoro...), szerző] Gépjárműtechnológia Tanszék (BME / KJK)

Angol nyelvű Tudományos Konferenciaközlemény (Egyéb konferenciaközlemény)
    Azonosítók
    • MTMT: 32464222
    An end-to-end procedure predicting decisions by applying deep learning techniques to simulate and reproduce human driving behaviors from camera images is one of the wellknown methods, with the highest prediction performance, for developing autonomous vehicles. Convolutional neural networks (CNNs) are frequently used in the controlling processes of steering angles and conventionally forward-facing cameras are used to generate lateral error cases. In the present paper, the authors propose an end-to-end lateral control method based on previouscontinuous steering angle information learning of a deep convolutional neural network (DCNN) by considering the ego vehicle's dynamics to predict human driver decisions, but without neglecting the longitudinal control cases such as acceleration and deceleration.
    Hivatkozás stílusok: IEEEACMAPAChicagoHarvardCSLMásolásNyomtatás
    2022-01-26 21:51