MODULAR PRECISION ASSEMBLY SYSTEMS http://www.cs.cmu.edu/~msl The Microdynamic Systems Laboratory in the Robotics Institute at Carnegie Mellon University has embarked on a new project to improve the way automated assembly systems are designed, programmed, deployed, and operated. We are concerned with the creation of new hardware and software technologies and methods for automated assembly of precision high-value products such as magnetic storage devices, palmtop and wearable computers, and other high-density equipment. Our developing Agile Assembly Architecture (AAA) supports the creation of miniature assembly factories (minifactories) built from small modular robotic components. The goals are to substantially reduce design, deployment, and product changeover times, to greatly improve quality levels by using sensor-moderated precision motion, and to achieve new levels of manufacturing system portability by shrinking the sizes of typical assembly systems from large room size to tabletop size. We believe these goals can only be achieved through careful integration of hardware and software tools in a manner heretofore unseen in the robotics and automation community. We are building a distributed system of tightly integrated mechanical/computational agents endowed not only with information about their own capabilities but also with the ability to appreciate their role in the factory as a whole and negotiate with their peers in order to participate in flexible factory level cooperation. AAA will use defined procedures, protocols, and well-structured agent autonomy to simplify the process of designing and programming large high-precision assembly systems [1]. The unified design and programming tools of AAA will allow a user to select agents over the Internet and program them in a simulated factory environment. AAA will take advantage of agents' self-knowledge and ability to explore their environments to make the transition between simulation and reality as painless and seamless as possible. To facilitate the design and operation of distributed systems of this type, each minifactory agent will be capable of representing itself to its peers and providing detailed models (both geometric and behavioral) for use in its simulation. To support a suitably extensible simulation environment, we expect to make use of distributed models and processing by relying on agents to represent their own behavior during simulation [2]. This mitigates the need to develop necessarily restricted models for agent behavior and allows for simplified integration of new types of agents into AAA. Figure 1 illustrates how a designer might develop a specific minifactory for a particular family of related products. The designer is shown interacting with a comprehensive modeling and simulation environment. Functional models of the minifactory components are distributed: they reside within each minifactory module at the geographic location of the module vendor company. By accessing and incorporating data from each on-line module remotely (by Internet), the modeler/simulator has only up to date and reliable information instead of catalog or faxed data sheet information which might not reflect the currently available module product. During the creation of the minifactory design, the designer is free to select and simulate a wide variety of configurations using modules from many different sources to arrive at a final design (virtual minifactory). The designer may also make use of various tools such as automated assembly sequence planners, schedulers, and layout tools. Many other steps will be required successfully program, deploy, and operate a complete minifactory. Figure 2 is a representative sketch of the minifactory hardware we are developing [3]. Products move through the factory on the backs of couriers based on closed loop planar linear motors interacting with overhead devices which place parts and perform other operations. Our hope is that AAA and minifactory can provide viable high-precision assembly alternatives for industries that would benefit from drastically reduced factory design and deployment time. More generally we project that the resulting class of factory automation will be suitable for use in the production of short-lead-time products consisting of moderate-sized high-precision parts and assemblies. We see this distributed approach to factory design, modeling and control affording a number of advantages including modularity, robustness, efficiency, and simplicity. Acknowledgements This research is supported in part by the National Science Foundation under grants CDA-9503992, DMI-9523156, and DMI-9527190, and by the CMU Engineering Design Research Center and CMU Data Storage Systems Center. Dr. Ralph Hollis The Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213 USA E-mail: rhollis@cs.cmu.edu References [1] R. L. Hollis and A. Quaid, "An Architecture for Agile Assembly," Proc. Am. Soc. of Precision Engineering, 10th Annual Mtg., Austin, TX, October 15-19, 1995. [2] A. A. Rizzi, J. Gowdy, and R. L. Hollis, "Agile Assembly Architecture: An Agent-Based Approach to Modular Precision Assembly Systems," IEEE Int'l Conf. on Robotics and Automation, Albuquerque, April 20-25, 1997. [3] A. Quaid and R. L. Hollis, "Cooperative 2-DOF Robots for Precision Assembly," IEEE Int'l Conf. on Robotics and Automation, Minneapolis, April 22-28, 1996. Figure Captions Figure 1. Minifactory Design Scenario Figure 2. Operating Minifactory