Inference and Planning at Occluded Intersections for Urban Autonomous Driving
Xuhe Zhao,Chaojie Zhang,Jun Wang,Yanting Zhu
DOI: https://doi.org/10.1109/tiv.2024.3409726
IF: 8.2
2024-01-01
IEEE Transactions on Intelligent Vehicles
Abstract:Occlusions significantly impact the safety of autonomous vehicles. Intersections with multiple occlusions and multiple observed vehicles are common in urban scenarios. This paper proposes a solution for occlusion inference and planning in the above scenarios. This framework introduces a novel model of potential traffic, enabling a unified representation of uncertainty. The occlusion inference is based on the inverse planning. It uses behaviors of social vehicles to infer the potential traffic. When there are more than one social vehicles, evidence theory is used to fuse the inferences from multiple sources. Simulations are conducted in different intersection scenarios, demonstrating the improvement of traveling efficiency.
What problem does this paper attempt to address?