Designing For System Intelligence Sesh Commuri 1 , Yushan Li 1 , Dean Hougen 2 , Rafael Fierro 3 1 School of Elec. & Comp. Engg., University of Oklahoma, Norman, OK 73072 2 School of Comp. Science, University of Oklahoma, Norman, OK 73072 3 School of Elec. & Comp. Engg., Oklahoma State University, Stillwater, OK 74078 Abstract: This paper presents a hierarchical architecture for the realization of teams of intelligent autonomous ground vehicles called Adaptation and Learning at All Levels (AL 2 ). Requirements on system intelligence are translated into requirements on the system hardware and software. Recent trends in the hardware-software co-design and hardware reconfiguration that enable the realization of these requirements are then introduced and the design methodology is discussed. Examples are provided to illustrate the design methodology. Copyright © 2004 IFAC Keywords: Architectures, intelligent systems, autonomous mobile vehicle. 1. INTRODUCTION In recent years, mobile robots have evolved from the humble origins of Autonomous Ground Vehicles (AGVs) in manufacturing applications to complex multi-vehicle robotic teams that are expected to coordinate and function cooperatively to satisfy specified objectives. The high cost associated with the acquisition and deployment of mobile robots motivates the development of low cost multi-robot teams that can function cooperatively to achieve specified goals. As the performance requirements get more stringent and the application realm become more diverse, embedding “Intelligence” in the mobile robots becomes critical to the realization of Intelligent Autonomous Vehicles. While a number of techniques were proposed in literature for the control and coordination of multi-robot teams, the use of these systems is constrained due to the inflexible hardware and software implementations. Designing systems that are intelligent is substantially more complex than just the integration of smart controllers or transducers (IEEE 1451.1, 1451.2, 1999a,b). Embedding intelligence into systems requires a new design paradigm that takes into account the hardware and software complexities involved in the design of embedded systems (Albus et al. 2000a,b). The design must address issues such as: y intelligent sensor selection and sensor fusion, y fault detection and accommodation, y dynamic reconfigurability and high levels of reliability under all operating conditions, y hierarchical architecture for implementing hybrid controllers, y sophisticated algorithms for implementing adaptive behaviours, y remote operation and coordination with other controllers, and y provision for system evolution, and paths for the migration of capabilities. The design must also address real-time and non real- time issues in sensor fusion, communication between network nodes, and the automation of the decision making process. Many methods have been suggested for controlling multi-robot systems (Cao et al., 1997, Dudek et al. 1996). Some of these use a centralized planning approach to make efficient use of resource (Bonasso et al., 1997, Schreckenghost et al., 1998). In contrast, the requirement for loss tolerance has lead many researchers to consider distributed systems (Balch et al. 1998, Parker et al. 1998). More recently, attention has been paid to design architectures that combine the advantages of both centralized and distributed approaches (Simmons et al., 2002). While the selection of the right architecture plays an important role in the design of the system, the ultimate performance depends on the software and the algorithms implemented therein. In recent years, adaptation and learning have been proved to be an effective way to enable robots to carry out complex IAV2004 - PREPRINTS 5th IFAC/EURON Symposium on Intelligent Autonomous Vehicles Instituto Superior Técnico, Lisboa, Portugal July 5-7, 2004