Model-based tactile object recognition using pseudo-random encoding.

Abstract

An original model-based tactile 3-D object recognition system is presented. The development of this system was motivated by the need for a "blind" back-up solution to maintain the object recognition capability of the Space Shuttle robot manipulator when the sun light blinds the video-cameras normally used for object recognition. Tactile object recognition requires the identification of the explored object as well as the recovery of its 3-D position and orientation. Such a complex task calls for a robotic tactile sensing system whose parameters (spatial resolution, probing compliance, tactile image processing) are beyond those usually offered by the emergent tactile sensing technology. The development of the tactile sensing system described in this thesis has resulted in a number of contributions: (i) novel tab-shaped elastic overlay which reduces the cross-talk errors in the tactile sensor, (ii) instrumented passive compliant wrist for more efficient object exploration, and (iii) 2-D correlation method for the integration of the local tactile images in a global image of the explored object surface. The original model-based tactile object recognition paradigm in this thesis can be summarized as follows: "Given a set of 3-D objects having their surfaces Braille-like embossed with terms of a Pseudo-Random Array (PRA) for which is a priori known how the unfolded faces of geometrical models of these objects are mapped into the PRA plane, and given the tactile image of the explored object surfaces, determine the identity, position, and orientation of that object." Tactile object recognition is thus reduced to the recognition of a small set of symbols embossed on object surfaces. The use of multi-valued (instead of the usual binary) pseudo-random encoding allows to reduce the number of embossed symbols which actually are needed to recognized in the tactile image. An original contribution is made by the development of a Pseudo-Random Multi-Valued Product Array (PRMVPA) encoding and related pseudo-random/natural code conversion algorithm. While it is inherently limited to a given set of embossed objects, the described "blind" method allows for a simpler and faster 3-D object recognition using solely tactile sensing. Such an object recognition method has potential applications in controlled environments such as the Space Shuttle, nuclear stations, underwater maintenance of the oil platforms and pipelines.