| A major factor which has limited the application of robots in industrial settings has been the lack of robust sensing and control algorithms for detection and prevention of collision which could cause injury to humans and damage to an expensive robot. In this dissertation, a new approach is presented for an industrial robot safety system that uses the combined technologies of neural networks and fuzzy logic to accomplish real-time sensory data fusion and decision making. A three level safety architecture is presented which consists of the sensing level, the integration level, and the safety decision-making level. The integration level is implemented by a set of neural networks and the decision-making level is a fuzzy-logic controller.; The unique aspects of this research are the development of (1) a new hybrid approach combining neural networks and fuzzy logic for sensory data fusion, (2) a partitioning algorithm for sensory elements in a hybrid control architecture, and (3) a new robot safety control system architecture using the hybrid sensory fusion approach. This dissertation also includes an implementation of the new robot safety architecture, and a comparison to other safety architecture approaches. This new approach to sensor data fusion (or sensor integration) will have applications beyond the area of robot safety, such as machine vision, military applications (i.e. classification of radar targets), speech recognition, and fault diagnosis. |
