Gradient descent dynamic window approach to the mobile robot autonomous navigation
Résumé
Avoiding obstacles is a key feature in a vehicle autonomous navigation methodology. The dynamic window approach (DWA), which has been proposed for several decades, has emerged as a responsive navigation methodology suitable reactive for obstacle avoidance. In the initial approach of the dynamic window obstacle avoidance strategy, the optimization of an objective function is realized with an exhaustive computation which can be costly in computational time. This is not useful in a real-time scenario where an autonomous vehicle needs to avoid obstacles in urban or road velocity conditions. The improved run time execution also makes for a less abrupt and more comfortable driving experience. In this paper, we revise the DWA methodology, implementing a new method that forces us to redefine the objective function differently, defining it as a loss function. In order to apply a gradient descent to optimize the convergence.
We use DWA to control our autonomous cars, but to verify the correctness of our optimization, we optimized DWA in its original context, i.e., to control a robot in an unknown environment with obstacles to visit given positions. We tested our approach, in simulation on ROS and on a real Turltebot robot.
Domaines
Automatique / Robotique
Origine : Fichiers produits par l'(les) auteur(s)