This paper presents collision avoidance and local motion planning modules for a mobile robot equipped with a depth camera. In this paper, we identify some limitations of the existing neural controller, and then we propose the extensions which improve the behavior of the robot. We show that the knowledge about control history is crucial to efficiently avoid collisions with the obstacles if the robot is equipped with a narrow field of view camera. We propose the architecture which utilizes CNN-based neural modules to plan the local motion of the robot. Finally, we provide the results of the experimental verification on the real robot.
Milena Molska, Dominik Belter, Convolutional Neural Network-based Local Obstacle Avoidance for a Mobile Robot, 2021
Ещё видео!