| In this thesis, techniques for fault tolerance for a massive two-dimensional array of proximity sensors have been studied. When such sensors are crucial for system operation, it is important to have fault tolerance means that would assure system robustness. Sensor arrays designed for robot operation in an uncertain environment present a good example of such systems. As such arrays become ever larger in complex automation systems, the need for sensor fault tolerance becomes evident. The target system considered in this thesis is a modularized sensitive skin developed in the context of robot motion planning in uncertain environments. Such tasks, e.g. in underwater or construction site applications, are among the most daunting problems in robotics. Namely, one way to avoid obstacles without assuming a known environment is to make use of a sensitive skin that covers the whole body of the robot, along with proper sensor-based motion planning algorithms. In this work, this is accomplished via modularized sensitive skin patches that can be connected in various ways, and a hierarchical control architecture that can be easily modified for various system configurations. In the implemented system, the skin contains about 1000 infrared sensors; it can be easily expanded with minor changes. In addition to the hardware design, control software and software for sensor-based motion planning and collision avoidance have been developed. To operate such a complex system in real time, an effective fault tolerance system is necessary. Namely, it is desirable that the system would function properly even if some of its sensors were fully or partially inoperative. To achieve this goal, various types of faults that can appear in skin sensor system have been identified and classified. Then, fault tolerance algorithms have been developed, tested, and demonstrated experimentally. The approach used for fault tolerance makes use of redundancy obtained either from manipulator dynamics or from virtual sensors formed by sensor grouping. A number of fault models have been studied, and relevant fault recovery algorithms have been developed and tested. |
