Door Handle Detection Modelling for Servosila Engineer Robot in Gazebo Simulator

Abstract

For a typical indoor environment, a seamless navigation could be achieved only under the condition that a robot is capable to interact with standard obstacles of a human-centered environment, including a successful negotiation with various types of doors. In this paper the Russian crawler robot Servosila Engineer was used for modelling a door handle detection procedure within the Gazebo simulator. The detection was based on YOLO v3, which was connected to the Gazebo model of the robot using the Robot Operating System. Influences of a door color and a door transparency, a type of a door handle, a scene illumination and a distance from the robot to the door on successful detection rate were studied within a virtual room of the Gazebo simulator. An optimal distance to a door and a minimal luminance level were established empirically via the virtual experiments.