Motion Planning Approach Considering Localization Uncertainty

Localization plays an important role in autonomous driving since a high level of accuracy in vehicle localization is indispensable for a safe navigation. Most of the motion planning approaches in the literature assume negligible uncertainty in vehicle localization. However, the accuracy of localization systems can be low by design or even can drop depending on the environment in some cases. In these situations, the localization uncertainty can be taken into consideration in motion planning to increase the system reliability. Accordingly, this work presents two main contributions: (i) a probabilistic occupancy grid-based approach for localization uncertainty propagation, and (ii) a motion planning strategy that relies on such occupancy grid. Thus, the proposed motion planning solution for automated driving is able to generate safe human-like trajectories in real time while considering the localization uncertainty, ego-vehicle constrains and obstacles. In order to validate the proposed algorithms, several experiments have been conducted in a real environment.

I. Introduction

Automated driving requires methods to generalize unpredictable situations and reason in a timely manner in order to react safely even in complex urban situations. Generally, the knowledge about the environment is incomplete and the associated uncertainty is high, which affects motion planning. In view of this, two elements still need further substantial investigation: world modelling and motion planning from uncertain information.