Collision-Free Dynamic Window Approach for Moving Obstacles

Abstract Dynamic window approach (DWA) is known as collision-free for static obstacles with proper design. Under static operation conditions, if there is no admissible velocity within the velocity dynamic window, the agent reduces the velocity as much as possible. As such, the agent is collision-free as long as the projected time is large enough to make the agent fully stop. Existing work of the DWA considering moving obstacles demands for trajectory prediction of moving obstacles so that admissible velocities of the agent can be determined. However, there is no guarantee for any trajectory prediction method that the predicted trajectory is accurate even considering prediction confidence. Hence, collisions are reported for the DWA considering moving obstacles. This paper proposes collision-free DWA for moving obstacles through determination of the mobility of moving obstacles within each control cycle. Algorithms are proposed to efficiently determine admissible velocity of the agent considering the mobility of moving obstacles. Collision-free results are demonstrated through two case studies including an open space navigation and a merging lane navigation through the comparisons with the DWA based on the trajectory prediction.