Deep Imitation Learning for End-to-End Autonomous Driving: Integration of Carla and OpenAI Gym

Traditional autonomous driving systems, structured with modular architectures, have achieved significant results in autonomous driving. However, delays and errors within each module can lead to degraded system performance. To address this issue, recent research has focused on end-to-end autonomous driving systems. However, the lack of standardized architectures or development environments presents limitations in developing and verifying new algorithms. In this paper, an architecture is designed to facilitate the smooth development of end-to-end autonomous driving technology. To achieve this, an integrated environment combining the Carla simulator and OpenAI gym is configured. Additionally, to enhance the dataset quality, image and vehicle state information over time is sequentially accumulated to construct the dataset, and the driver's point of view (PoV) information is added using virtual reality (VR). Moreover, real-time expert feedback is applied to increase learning efficiency, making it easier to develop deep imitation learning models. To verify the validity of the proposed system, validation is performed in various environments, and the results are analyzed.