J.A. Carrasco-Ochoa et al. (Eds.): MCPR 2010, LNCS 6256, pp. 40–49, 2010. © Springer-Verlag Berlin Heidelberg 2010 Learning and Fast Object Recognition in Robot Skill Acquisition: A New Method I. Lopez-Juarez 1 , R. Rios-Cabrera 1 , M. Peña-Cabrera 2 , and R. Osorio-Comparan 2 1 Centro de Investigación y de Estudios Avanzados del I.P.N. (CINVESTAV) Ramos Arizpe. CP 25900. Coah. México {ismael.lopez,reyes.rios}@cinvestav.edu.mx 2 Instituto de Investigaciones en Matemáticas Aplicadas y en Sistemas, Universidad Nacional Autónoma de México (UNAM) Apdo. Postal 20-726, México DF, México {mario,roman}@leibniz.iimas.unam.mx Abstract. Invariant object recognition aims at recognising an object independ- ently of its position, scale and orientation. This is important in robot skill acquisition during grasping operations especially when working in unstruc- tured environments. In this paper we present an approach to aid the learning of manipulative skills on-line. We introduce and approach based on an ANN for object learning and recognition using a descriptive vector built on recurrent patterns. Experimental learning results using a fast camera are presented. Some simple parts (i.e. circular, squared and radiused-square) were used for comparing different connectionist models (Backpropagation, Perceptron and FuzzyARTMAP) and to select the appropriate model. Later during experi- ments, complex figures were learned using the chosen FuzzyARTMAP algo- rithm showing a 93.8% overall efficiency and 100% recognition rate with not so complex parts. Recognition times were lower than 1 ms, which clearly indi- cates the suitability of the approach to be implemented in robotic real-world operations. Keywords: ART Theory, Artificial Neural Networks, Invariant Object Recog- nition, Machine Vision, Robotics. 1 Introduction Grasping and assembly operations using industrial robots is currently based on the accuracy of the robot and the precise knowledge of the environment, i.e. information about the geometry of assembly parts and their localization in the workspace. Tech- niques are sought to provide self-adaptation for robots. This document reports a neural-based methodology for invariant object recognition applied to self-adapting industrial robots which can perform assembly tasks. New objects can also be learned quickly if certain clues are given to the learner, since the methodology uses only two on-line patterns for learning complex objects. The architecture is firstly trained with clues representing different objects that the robot is likely to encounter (and with others that represent complex objects) within the working space to form its initial