Distributed Architecture for Intelligent Robotic Assembly Part I: Design and Multimodal Learning

摘要

In this chapter we describe the general framework design for a distributed architecture to integrate multiple sensorial capabilities within an intelligent manufacturing system for assembly. We will formally define the model of the M2ARTMAP multimodal architecture. We present some simulated results of the model using a public domain multimodal data base. Initial findings have indicated the suitability to employ the M2ARTMAP model in intelligent systems when three or less modalities are involved. Taking into account these results and the M2ARTMAP’s modularity it was decided to integrate this model using the CORBA (Component Object Request Broker Architecture) middleware to develop robotic assembly tasks that includes contact force sensing, an invariant object recognition system and natural language processing. This chapter introduces the overall system in the intelligent cell and the task planner (SIEM) and the object recognition system (SIRIO) are described in detail in part II and Part III. Design and experiments of the distributed architecture using a Local Area network (LAN) and an industrial robot KUKA KR-15 to fusion different modalities in a common task are first described. The modalities are: vision, contact force sensing (tactile), and natural language processing using context free grammars. The vision modality and force sensing are implemented based on a FuzzyARTMAP neural network and a main coordinator. The testbed for the application and a general distributed environment using CORBA as a middleware is described. Later several learning simulations using M2ARTMAP, are presented (Lopez-Juarez & Ordaz-Hernandez, 2005). The main distributed objects are described in detail, the experimentation is presented and the results analyzed. Finally, future work is proposed.