Learning vision based autonomous lateral vehicle control without supervision

Abstract Supervised deep learning methods using image data as input have shown promising results in the context of vehicle control. However, these supervised two main disadvantages: 1) They require a copious amount labeled training data, which is difficult and expensive to collect. 2) Such models do not perform well, when situations that are distribution set encountered. This includes deviations from designated driving behavior. We therefore provide framework mitigate problems merely an unlabeled sequence images. Visual Odometry first used determine trajectory. Model Predictive Control (MPC) then uses this trajectory implicitly infer steering labels. Meanwhile, synthesized images at deviated trajectories included for enhanced robustness neural network model. Experimental demonstrate performance our par with requiring additional collection or supervision. Code supplementary information available here: https://github.com/idilsulo/nn-driving