Occlusion-Aware Trajectory Planning With Quantified Risk Constraint for Deadlock Mitigation in Autonomous Driving

Driving through traffic scenes with occlusions safely, efficiently and comfortably remains a huge challenge for autonomous vehicles (AVs). The omnipresence of blind spots, where emergencies may arise unexpectedly, requires AVs to drive with adequate precautions. However, existing planning algorithms often resort to excessive conservatism that leads to navigation-disrupting deadlocks. To address these challenges, we adopt the concept of phantom obstacles, imparting them with uncertainties. At that point, we propose an integrated lateral and longitudinal trajectory planning method to emulate human-like precautionary driving styles, and adopt stochastic model predictive control (SMPC) to model various uncertainties of obstacles. Furthermore, we analyze deadlock causation mechanisms and present a novel quantified risk assessment method featuring an innovative adaptation of ST projection for occlusion-aware planning. This approach formulates a driving strategy that significantly reduces the likelihood of deadlock, enhancing the AV’s ability to navigate effectively in occluded scenarios. Case analysis and statistical research across diverse scenarios, featuring different objects in various typical settings, validate the safety, efficiency and comfort of the proposed planning results. These findings underscore the potential of our approach to advance the capabilities of autonomous vehicles in navigating complex and occluded driving environments.