Poster: An Efficient Road Detection Framework for Supporting Autonomous Driving in Rural Areas

Cloud networking infrastructure and edge computing have made autonomous driving in metropolitan and urban areas feasible in the near future. Unfortunately, it is currently not possible to deploy safe and reliable autonomous vehicles in rural areas due to the stark contrast in the environments. Underdeveloped areas lack reliable telecommunication infrastructure for vehicle-to-infrastructure communication. In addition, rural roads are typically in poor condition compared to urban roads. Vehicles failing to detect where the road is can lead to detrimental effects when driving in a real-time scenario. In this paper, we propose a solution to address autonomous driving on rural roads. Autonomous vehicles may be usable in rural areas without telecommunication services by loading computation aboard the vehicle on an edge device. We deploy a MobileSAM model on practical edge device hardware to segment where the road is on a front-facing camera feed. In order to ensure the model is fast and reliable for real-time use, a promising solution is to optimize the model using Apache Tensor Virtual Machine (TVM), an open-source framework for loading deep learning models on any hardware.