| The proposed research addresses the problem of part assembly. This process can be broken down into macro-assembly, bringing the part to a target, and micro-assembly, mating the part with a socket or receptacle. The distinguishing feature of the robotic control developed in this research is part assembly in unknown or uncertain environments, that is, environments that contain obstacles whose size and location are not initially known. To cope with such an environment requires that the control strategies have a high degree of artificial intelligence. This intelligence in large part depends upon the ability of the control strategy to improve with time, that is, to learn. One of the popular methods of incorporating learning is via neural networks. Intelligence, and to a lesser degree learning, can also be incorporated using fuzzy set theory and fuzzy control. The proposed research uses both neural networks and fuzzy control. One of the main goals of this research is to determine how best to combine, conventional control with other forms of control, such as neural networks and fuzzy control. Another main goal of the research is to explore and clarify the concept of learning in artificial intelligence. Another distinguishing feature of this research is the incorporation of a cost function, based on fuzzy entropy, to provide a measure of optimality of the control. The addition of the cost function proves very fruitful. For example, its presence enhances the learning capabilities of several of the algorithms. Two final goals of the research are to investigate how different assembly strategies can benefit from sensor fusion and to determine which of the strategies are most compatible with hierarchical control. |
