A Viewpoint Planning and Navigation Algorithm for Mobile Robots using Depth Images

In this paper, we illustrate a viewpoint planning and local navigation algorithm for mobile robot exploration using 3D perception. Laser scans and stereo camera data are read and composed into a single point cloud. The latest acquired point clouds are registered and used to detect relevant objects and obstacles in space. To demonstrate the capability of the proposed perception system, a viewpoint planning application has been developed. The algorithm selects a goal object, computes a viewpoint to observe the region occluded by such object, and plans a path to reach the desired configuration. The navigation method executes the computed path and reaches the goal. Experiments have assessed the capability of the system to plan using three-dimensional occupancy information and to find new viewpoints in cluttered environments.