Hybrid Optimization Method for Safe Autonomous Navigation under Uncertainty

A fundamental difficulty that autonomous driving systems face in dynamic environments is uncertainty in the motion of the other traffic participants. Although predictive control algorithms have proven to be able to handle such scenarios, the multimodal motion of agents often increases the risk in traffic environments, leading to unsmooth motion and sometimes the failure of such risk assessment and management algorithms. This paper proposes to tackle such problems using a hybrid optimization method that finds a global safe trajectory with respect to the surrounding agents using a Fusion of stochastic Predictive Inter-Distance Profile (F-sPIDP) before following the safe path using a combination of stochastic reference points and an uncertainty-aware local trajectory optimizer. The results show that the proposed method guarantees safety even with varying levels of uncertainties.