Self-reconfiguring robots: designs, algorithms, and applications

Self-reconfiguring robots are robots composed of many physically connected modules which can change their structural configuration to support multiple functionalities. We claim that self-reconfiguring robots are more versatile, extensible, and fault tolerant than conventional fixed-architecture robots. In this thesis we discuss the scientific challenges of self-reconfiguring robots and their applications. We describe the design of our Molecule self-reconfiguring module and give results for our Molecule experiments. We also present planning algorithms tightly coupled to the Molecule hardware, as well as a generic approach for developing provably correct, distributed control algorithms which can be instantiated onto various hardware platforms. The Molecule is a robotic module capable of aggregating with identical modules to form dynamic three-dimensional structures. Several prototype designs are described. A planner is presented for developing relocation plans and an algorithm is described for merging multiple serial plans into a parallel plan. Control algorithms are proposed for Molecule structure locomotion, including planar translation, vertical stacking, and stair climbing. Experimental results are given for Molecule structure self-reconfiguration and for planning. design and analysis, permitting the development of performance guarantees. Algorithms consist of geometric rules based on the local environment and module internal state. We demonstrate provably correct locomotion algorithms, both with and without obstacles. We also present reconfiguration algorithms for self-assembly and an adaptive structure. Automated proving techniques are proposed which can verify rule set correctness. We describe several instantiations of our algorithms onto different hardware platforms, demonstrating that the benefits of our algorithmic performance guarantees can be applied to multiple hardware implementations.