An Integrated Decision and Motion Planning Framework for Automated Driving on Highway

Decision-making and motion planning are two core functionalities of intelligent vehicles. A novel integrated decision-making and planning framework is presented for real-time and safe driving on highway under complex scenarios. This integrated framework decomposes the complicated coupling between the environment and controlled vehicle by a dynamic safety domain composed of two layers. The outer layer is designed to integrate multiple driving objectives and subtasks as a safety boundary. It describes the interactions between the ego vehicle and surrounding vehicles, lane lines, and speed limit as potential fields, to ensure good real-time and safety performances simultaneously. The inner layer is designed by using the dynamics of the safety boundary to obtain the planning solution analytically for better computation efficiency. With this framework, a two-dimensional risk model is set up to solve the safe driving trajectory directly without the time-consuming interactive process. The proposed approach was tested and compared with other methods to validate its feasibility and effectiveness. The results show that it can make decision and planning safely, efficiently and comfortably under the dynamic traffic scenarios.

I. Introduction

Intelligent vehicle (IV) has great potential to improve traffic safety and efficiency, and even completely change the transportation mode. Based on the environmental perception results, the decision-making and motion planning module can generate a control sequence of IV [1]. This study focuses on the design of the latter one, for which there are mainly two typical frameworks to combine different driving tasks [2]: (1) Hierarchical framework; (2) And integrated one. The hierarchical framework splits the decision and motion planning function of whole driving tasks into many submodules, e.g., behavior selection, path planning, control and etc. On the contrary, the integrated one generates the desired control sequence directly.