Automatic Modeling for Modular Reconfigurable Robotic Systems: Theory and Practice

Abstract

A modular reconfigurable robot consists of a collection of individual link and joint components that can be assembled into a number of different robot geometries. Compared to a conventional industrial robot with fixed geometry, such a system can provide flexibility to the user to cope with a wide spectrum of tasks through proper selection and reconfiguration of a large inventory of functional components. Several prototyping systems have been demonstrated in various research institutions (Cohen et al. 1992; Fukuda & Nakagawa 1988; Schmitz, et al. 1988; Wurst 1986). Applications of modular systems have been proposed in rapid deployable robot systems for hazardous material handling (Paredis et al. 1995), in space stationed autonomous systems (Ambrose 1995), and in manufacturing systems (Chen 2000; 2001). In the control and simulation of a modular reconfigurable robot system, precise kinematic and dynamic models of the robot are necessary. However, classical kinematic and dynamic modelling techniques for robot manipulators are meant for robot with fixed geometry. These models have to be derived manually and individually stored in the robot controller prior to simulating and controlling the robot. Commercial robot simulation software usually provides end users with a library of predefined models of existing robots. The models of any new robot not in the library have to be derived exclusively from the given parameters and commands in the package. For a modular robot system built upon standard modular components, the possible robot geometries and degrees of freedom become huge. As shown by Chen (1994), the number of robot-assembly configurations grows exponentially when the module set becomes large and the module design becomes complicated. To derive all of these models and store them as library functions require not only tremendous effort but also very large amount of disk storage space. In such cases, it is impractical and almost impossible to obtain the kinematic and dynamic models of a robot based on the fixed-geometry approach. Hence, there is a need to develop an automatic model-generation technique for modular robot applications.