Detecting and transporting objects by pushing-only approach

Abstract

Due to laser scanners’ quality of information, there is a wide spectrum of applica- tions for these sensors in indoor and outdoor, both structured and unstructured environments. Most works present their own case-specific modelling strategies, which are often similar, although presenting no unity or consensus among them. In light of this lack of formalism this dissertation presents an analytical approach for identification and localization of objects using laser sensors. Firstly, the contribution lies in formally defining a laser sensor measurements and their representation, the identification of objects, their main properties and their location in a scene. Sec- ondly, this work presents handling box-shape objects combining mapping, searching, and path planning techniques. Laser scanner data are used to build up a 2D map, which aids the objects’ identification in the scene. Thirdly and most important, the dissertation aims a robust algorithm for pushing objects, from random positions to a final destination. Our main contribution is the route recovery strategy, which reacts whenever the box transportation starts going out of the planned one. It provides robustness to objects rotation and slipping during their displacements, thus guarantees all of them are correctly delivered. Besides, a topological map is created by Voronoi Graph in order to avoid collisions and Dijkstra’s algorithm finds the optimal route. Then, Bézier curves provides suitable paths taking into account the position of the robot, objects and final destination. Finally, simulations are run in V-REP + Matlab, and real experiments validate the proposal, which demonstrates quite efficient for environments without occlusion of the objects to be transported. Keywords: Reactive Object’s Control. Voronoi Graph. Bezier curves. Dijkstra’s algorithm.