Abstract:
|
Many mobile robotic systems produce a 3D model of their environment by sweeping it with
their RGBD (color and depth) cameras. In the case of the Advanrobot, an automatic inventory
wheeled robot for retail stores, such ability opens the door to a range of new possibilities such
as building a virtual store or helping retailers in measuring the impact of the placement of
products, furniture and their combination in the sales.
This work presents a solution to enable Advanrobot for building 3D models from its RGBD
camera measures. To do so, external tools have been used to handle navigation, localization,
mapping and point cloud combination so that the main development is focused on 3D exploration
to ensure 3D model completeness. As navigation is limited to 2 dimensions, the ground
projection of the camera point clouds is used as input for a frontier based exploration approach
consisting in a flexible framework able to extensively tune exploration decisions such as the
evaluation of frontiers, when should the robot decide a new navigation goal or how to choose
a proper goal when a frontier has already been chosen. Also, some additional tools have been
developed as spin-offs of the work such as a hole detector, a point cloud trimmer or a camera
calibrator.
The approach has been proved to be useful for automatic 3D mapping of the environment.
Exploration based mapping lead to similar results than teleoperated mapping. Each related
tool also solved its corresponding problem.
The results of this work may apply to any wheeled robot with a fixed RGBD camera. |