Situation-Aware Drivable Space Estimation for Automated Driving

An automated vehicle (AV) must always have a correct representation of the drivable space to position itself accurately and operate safely. To determine space, current research focuses on single sources information, either using pre-computed high-definition maps, or mapping environment online with sensors such as LiDARs cameras. However, each these information can fail, some are too costly, maps could be outdated. In this work new method for situation-aware (SDS) estimation combining multiple is proposed, which also suitable AVs equipped inexpensive sensors. Depending situation, semantic sensed objects combined domain knowledge estimate drivability surrounding object (e.g. traffic light, another vehicle). These estimates modeled probabilistic graphs account uncertainty sources, an optimal spatial configuration their elements determined via graph-based simultaneous localization (SLAM). investigate robustness SDS towards potentially unreliable it has been tested in simulation real world data. Results different use cases straight roads, curved intersections) show considerable inputs, recovered allows accurate in-lane AV even extreme where no prior road network available.