Elements of Autonomous Self-Reconfigurable Robots

Abstract

In this thesis, we study several central elements of autonomous self-reconfigurable modular robots. Unlike conventional robots such robots are: i) Modular, since robots are assembled from numerous robotic modules. ii) Reconfigurable, since the modules can be combined in a variety of ways. iii) Self-reconfigurable, since the modules themselves are able to change how they are combined. iv) Autonomous, since robots control themselves without human guidance. Such robots are attractive to study since they in theory have several desirable characteristics, such as versatility, reliability and cheapness. In practice however, it is challenging to realize such characteristics since state-of-the-art systems and solutions suffer from several inherent technical and theoretical problems and limitations. In this thesis, we address these challenges by exploring four central elements of autonomous self-reconfigurable modular robots: design, scalability, self-reconfiguration and adaptation. <br/><br/>The first element we consider is the design of systems, modules, robots, and behaviors. We introduce a number of design principles that will guide our designs throughout the thesis. The design principles advocate simple, extendable, heterogeneous systems, where the robot's behavior emerges from autonomous modules controlled in a distributed fashion. The second element considered is scalability in terms of size and number of modules. We study the interdependence between morphology, module size and behavior and observe how none of these aspects can be studied in isolation. To facilitate scalability we propose a module organization inspired by the anatomy of biological organisms, which allows reuse of module structures and control from one robot to the next. The third considered element is the process of self-reconfiguration. To fulfill the goals of scalability and fault-tolerance, we propose a distributed strategy based on meta-modules that emerge from the structure of other modules. The fourth and final element considered is adaptation, which we study in the context of locomotion. Our approach is distributed by having each module learn its own function in isolation from other modules. We study how adaptive, configuration independent and fault-tolerant collective behaviors emerge at the level of the robot.<br/><br/>