Drive with Quantified Risk: A Probabilistic Approach for Occlusion-Aware Trajectory Planning

Driving through traffic scenes with occlusions safely, efficiently and comfortably remains a huge challenge for autonomous vehicles (AVs), Blind spots are ubiquitous, from where emergencies may occur at any time, making AV hardly evade without sufficient precaution. In this study, we refer to the concept of phantom obstacles, and give them various aspects of uncertainties to ensure safety when facing driving scenarios with occlusions. A comprehensive lateral and longitudinal trajectory planning method is proposed based on stochastic model predictive control (SMPC) to mimic human-like precautionary driving styles. Furthermore, based on the cause analysis of deadlock situation, a quantified risks approach is proposed for occlusion-aware trajectory planning to formulate a non-conservative and deadlock-free driving strategy. Simulations for different objects in various typical scenarios demonstrate the safety, efficiency and universality of planning results.