Roundabout Crossing With Interval Occupancy and Virtual Instances of Road Users

Although autonomous vehicle technology has evolved significantly in recent years, the navigation of self-driving vehicles complex scenarios is still an open issue. One major challenges these conditions safe on roads to public traffic. The main issue interaction with regular traffic, as behaviors and intentions human-driven are hard predict understand. In this paper we propose a strategy allow safely cross multi-lane roundabout. Our approach uses High-Definition (HD) map at lane level future situation, harnessing concept virtual instances road users, which key anticipating situation roundabout that can be represented by graph loops. This presents methodology intervals representing occupancy vehicles, being widened reflect uncertainties localization. method avoids collisions guarantees no priority constraints violated during insertion maneuver. Moreover, does not provide overly cautious policy, i.e. , wait for long time before insertion. performance our was evaluated using SUMO simulation framework. To better evaluate complexity scenario, highly interactive flow generated real dynamic traffic data from INTERACTION dataset. We report tests carried out experimental test circuit. results show easy integrate into embedded system it allows roundabouts crossed high safety.