LAUSR

Autonomous mobile robots perform many tasks, such as grasping and inspection, that may require complete models of three-dimensional (3-D) objects in the environment. If little or no knowledge about an object is known a priori, the robot must take sensor measurements from strategically determined viewpoints in order to reconstruct a 3-D model of the object. We propose an autonomous object reconstruction approach for mobile robots that is very general, with no assumptions about object shape or size, such as a bounding box or predetermined set of candidate viewpoints. A probabilistic, volumetric method for determining the optimal next-best view is developed based on a partial model of a 3-D object of unknown shape and size. The proposed method integrates an object probability characteristic to determine sensor views that incrementally reconstruct a 3-D model of the object. Experiments in simulation and on a real-world robot validate the work and compare it to the state of the art.