Tolerância a falhas adaptativa para robôs móveis com arquitetura híbrida

Abstract

This work presents a novel architectural design methodology, which enables weaving fault-tolerance into hybrid architecture control frameworks. The beneficial aspects fostered by fault tolerance greatly surpass the overhead in project development. Dataflow processing paradigm is based on functions and abstract data elements, providing a simple, yet powerful, description power. The simple structure allows the inclusion of redundant information and processing elements within the control. Redundancies elicited by dataflow automatically enable alternative configurations of available functional blocks. The methodology also gives the designer the freedom to define requirements and restrictions for both control performance and reliability. A control graph contains all the information necessary to optimize the robot resources usage and also captures all the necessary adaptations for the system. This graph may be automatically generated combining the hybrid control description and the dataflow information. The thesis shows the results from a prototype of hybrid architecture for a Nomad robot. The resulting fault tolerant hybrid architecture provides real time execution of control and fault recovery. Experiments were carried out to demonstrate the gain in reliability versus the added overhead.