Intention-aware Decision Making in Urban Lane Change Scenario for Autonomous Driving

Abstract

Autonomous vehicles need to face human-driving vehicles with their uncertain intentions in dynamic urban environment. Thus it leads to a challenging decision-making problem. In this paper, we focus on solving this problem in lane driving situation including performing lane changing or lane keeping maneuvers. A general POMDP model is formulated to represent autonomous driving decision-making process, and several approximations are applied to reduce the complexity of solving POMDP model. Firstly, we proposed a maneuver-based decomposition method to represent the possible candidate policies using path and velocity profiles in policy generation process. Secondly, a deterministic machine learning model is built to recognize human-driven vehicles’ driving intentions. Then, a situation prediction model is proposed to calculate the possible future actions of other vehicles considering cooperative driving behaviors. Finally, we build a multi-objective reward function to evaluation each policy. In addition, we test our methods in realistic simulation software. The experimental results show that our algorithm could perform lane keeping or lane changing maneuvers successfully.