Robotic Haptic Exploration of Shape and Symmetry

This thesis presents research on the use of symmetric models during haptic exploration procedures that have the objective of determining an object’s shape. These haptic exploration techniques, and their subsequent determination of a surface’s geometric properties, are crucial to allow robots to interact with a greater variety of objects, especially as the field of robotics transitions into unstructured environments. Symmetry is an extremely frequent shape property, especially in man-made objects, and it provides shape information that becomes useful in grasping and manipulation tasks, as well as enriching shape information for the aforementioned haptic exploration tasks. In this work, we present an improvement to Gaussian Process-driven exploration tasks. This method allows to describe symmetry to obtain a more precise shape estimation during active exploration, and can even be discovered in real time during the exploration procedure itself. This work involved the creation of a custom software resource to perform Gaussian Process regression with the addition of symmetries, and include a novel method of representing rotational symmetries. These novel models were then used in shape exploration procedures of 2D and 3D surfaces, both in a simulated environment and in an actual robotic task, using a series of custom-made contact sensors. These procedures are able to discover symmetry of each particular object in real time. This property can also be exploited, resulting in shape estimations that have a lower surface error and uncertainty. Additionally, exploration experiments that use these symmetry-finding procedures also require a lower total number of physical contacts and take less time to finish.