Vehicle autonomy using cooperative perception for mobility-on-demand systems

Abstract

A holy grail of research in urban transportation systems is to increase throughput of people while minimizing the requirement of building additional road and rail networks. The promising new paradigm of Mobility-on-Demand (MoD), where shared personal transportation vehicles provide necessary service, is fast becoming a viable and preferred alternative to the traditional framework of having either public fixed route service or privately owned vehicles. This chapter looks at innovative approaches that enable an MoD system to be economical in development, robust in operation, efficient and sustainable in deployment. We develop algorithms to efficiently use the combination of prior information with minimalistic sensing with only a single 2-D LIDAR, utilizing vehicle odometry and prior road information which may not have accurate metric information but is topologically consistent. Additionally, we use the rich capability of cooperative perception, which can far extend perception range without expensive long-range sensors, by exchanging local perception information with other vehicles or infrastructure via wireless communications. The augmented perception capability enables a vehicle to see the oncoming traffic situation ahead even beyond human line-of-sight and field-of-view, which thereby contributes to traffic flow efficiency and safety improvement through long-term perspective driving, e.g., early obstacle detection and avoidance, and early lane changing. This chapter develops these ideas, presents results in demonstration and provides insights of the motion and operation planning for a case study of a fully autonomous vehicle deployment to the general public.