Safety is one of the most significant challenges for autonomous driving. However, autonomous vehicles need also to be as efficient as possible while ensuring safety. In this paper we present an approach that helps the vehicle to drive efficiently in scenarios with occlusions, while being safe and comfortable. We quantity the visibility risk (VR) representing the collision risk with possible hidden obstacles in occlusions and anticipate the predictive VR by forecasting the scene in the short-term. The predictive VR can be integrated into the cost functional of arbitrary cost-based planning approach. By doing this, the vehicle is motivated to maintain a better view on the Region of Interest (ROI) on the map, which further opens up the scope of available behaviours and thus pursuing more efficient behaviour becomes possible. The proposed method is evaluated with a proof-of-concept planner decomposed into lateral (e.g. paths) and longitudinal directions (e.g. accelerations) in one overtaking scenario on a curved two-way street with dynamic occlusions and one intersection scenario with static occlusion, which shows promising results in vehicle driving efficiency.
Generating Efficient Behaviour with Predictive Visibility Risk for Scenarios with Occlusions
2020-09-20
1127145 byte
Conference paper
Electronic Resource
English
PEDESTRIAN COLLISION AVOIDANCE SYSTEM FOR SCENARIOS WITH OCCLUSIONS
British Library Conference Proceedings | 2019
|A POMDP MANEUVER PLANNER FOR OCCLUSIONS IN URBAN SCENARIOS
British Library Conference Proceedings | 2019
|