Reliable, robust, and comprehensive risk assessment framework for urban autonomous driving

Abstract

Abstract Urban autonomous driving is both complex and dangerous. A multitude of road types, road users, and potential traffic rule violations all make it challenging for an autonomous vehicle to safely navigate an urban environment. To tackle this issue, in this paper, we propose a risk assessment framework capable of reliably predicting and assessing the possibility of collision in urban environments involving the concurrent occurrence of various road types, road users, and traffic rule violations. The proposed framework reliably predicts and assesses the possibility of collision by means of long-term motion prediction–based risk identification, unified risk measures, and probabilistic risk reasoning under a distributed reasoning architecture. The framework is tested and evaluated through real-world testing conducted not only on private urban test roads under a great variety of driving scenarios but also on public urban roads under real traffic conditions. Experimental results show the performance of the framework to be sufficiently reliable for urban autonomous driving in terms of risk assessment.