Dynamic Window-based Obstacle Avoidance in the Presence of Moving Obstacles

This paper proposes a reactive control method for mobile robots in the presence of moving obstacles. The method is based on the dynamic window algorithm and extends it in order to avoid moving obstacles efficiently. Firstly, the future collision is detected based on the generalized velocity obstacles. Secondly, input value can be varied within the prediction horizon. This means that better input can be selected such that the robot starts passing maneuver ealier. However, this causes the dimension of the search space larger. In order to reduce the computation time, GDS (gradually dense-sparse) discretization and randomized sampling method similar to RRT is adopted. By means of these extensions, the robot can avoid moving obstacles with simpler cost function and reasonable computation time. Performance of the proposed method is evaluated by numerical simulations.