Adaptive Tactical Behaviour Planner for Autonomous Ground Vehicle

The ability of autonomous vehicles to successfully replace human drivers depends on their capability to plan safe, efficient and usable paths in dynamically evolving traffic scenarios. This challenge gets more difficult when the autonomous vehicle has to drive through complex scenarios such as intersections which demand interactive behaviour between vehicles. Many autonomous vehicle demonstrations over the last few decades have highlighted the limitations in the current state-of-the-art of path planning solutions. They have been found to be inefficient and sometimes unsafe when tackling interactively demanding scenarios. The generic path planning solutions consists of three planners, a "global path planner", a "behaviour planner" and a "local path planner. In this paper we establish that the "behaviour planner" is the limitation of a successful path planning solution, after reviewing the individual planners and the associated solutions. In this paper a new adaptive tactical behaviour planner is proposed to overcome the limitations. This planner is motivated by how expert human drivers behave in interactive scenarios, and is made up of a three module architecture. The paper describes the individual modules, and also highlights how they play a part in the overall behaviour selection for the autonomous vehicle. The paper is concluded by a discussion on how this proposed planner generates safe and efficient behaviours in complex dynamic traffic scenarios by considering a case of a roundabout not controlled by traffic signals.