Full-stack S-DOVS: Autonomous Navigation in Complete Real-World Dynamic Scenarios

Abstract

Autonomous navigation in dynamic environments is a nowadays unsolved challenge. Several approaches have been proposed to solve it, but they either have a low success rate, do not consider robot kinodynamic constraints or are not able to navigate through big scenarios where the known map information is needed. In this work, a previously existing planner, the Strategy-based Dynamic Object Velocity Space, S-DOVS, is modified and adapted to be included in a full navigation stack, with a localization system, an obstacle tracker and a global planner. The result is a system that is able to navigate successfully in real-world scenarios, where it may face complex challenges as dynamic obstacles or replanning. The final work is exhaustively tested in simulation and in a ground robot.