Integrated Decision Making and Planning Framework for Autonomous Vehicle considering Uncertain Prediction of Surrounding Vehicles
Chen Tang, Yuanzhi Liu, Hongyu Xiao, Lu Xiong
Abstract
Uncertainties in dynamical driving environments are crucial to the decision making and trajectory planning modules for autonomous vehicles. Without proper handling of such uncertainties, the decision result could be discontinuous, which causes jitters in the planned trajectory, and finally impacts driving safety and comfort. This paper proposes a decision-making and planning scheme based on Partially Observable Markov Decision Process (POMDP) to deal with the uncertainty in predicted trajectories, where future poses of surrounding vehicles are modelled using a multivariate Gaussian distribution. To better utilize decision results in the form of fine-grained discrete actions, construction of a feasible region constraint is also proposed to form an integrated decision and planning framework. By extending POMDP-determined actions considering uncertain trajectories of surrounding vehicles, the proposed scheme avoids jittery decisions and plans a smooth trajectory to reduce safety hazards and mitigate potential collisions.