Abstract
In this work, we present a lightweight pipeline for robust behavioral cloningof a human driver using end-to-end imitation learning. The proposed pipelinewas employed to train and deploy three distinct driving behavior models onto asimulated vehicle. The training phase comprised of data collection, balancing,augmentation, preprocessing and training a neural network, following which, thetrained model was deployed onto the ego vehicle to predict steering commandsbased on the feed from an onboard camera. A novel coupled control law wasformulated to generate longitudinal control commands on-the-go based on thepredicted steering angle and other parameters such as actual speed of the egovehicle and the prescribed constraints for speed and steering. We analyzedcomputational efficiency of the pipeline and evaluated robustness of thetrained models through exhaustive experimentation during the deployment phase.We also compared our approach against state-of-the-art implementation in orderto comment on its validity.