In this paper we address an important issue in the area of qualitative reasoning, that is, how to create qualitative models of mechanical devices that incorporate the effects of the detailed geometric structure of its components. The model we propose allows envisioning and prediction of behaviour without having to deal directly with the complexity of the detailed geometry of the parts. We claim that configuration spaces are an appropriate symbolic representation for reasoning about the kinematic behaviour of mechanical devices. We show that the kinematic behaviour of a mechanism can be precisely determined by first finding the relative motions of pairs of objects and then composing these motions. For a subclass of mechanisms, (fixed axis mechanisms), we show that a simplified version of the composition operation can be used to obtain the overall behaviour, and we outline a constraint propagation, label inferencing algorithm to produce a region diagram. This diagram constitutes a total qualitative envisionment of the mechanism's behaviour. Finally, given a sequence of input motions and a region diagram, we indicate how to predict the behaviour of the mechanism.
Bibliographical noteFunding Information:
Having a computer understand how devices such as boilers and amplifiers work has been the topic of much recent research in Qualitative Physics I 3, a subfield of AI. The two main goals of this research are to find computational methods that produce explanations on how a device works, and predictions on its behaviour. These explanations and predictions should be similar to those given by a person, namely they should identify important states of the device (such as 'on' or 'off', 'engaged' or 'disengaged'), and transitions between these states. They should thus focus on the qualitative aspect of changes inside a device, rather than on the precise values of the devices' parameters. Qualitative distinctions between states provide us with the necessary insight to understand how a device works. Further motivation of the work in Qualitative Physics is to find computationally efficient methods to predict the behaviour of a device. Traditional quantitative methods (such as differential analysis) are generally complex and expensive, and their full generality is unnecessary to explain the functioning of common devices such as heaters or boilers. In addition, the solutions produced by quantitative methods require careful interpretation to Author's current address: IBM T. J. Watson Research Center, PO Box 704, Yorktown Heights, NY 10598, USA. This work was partially supported by a National ScienceF oundation grant under contract DCR-8603758 and by the Defense Advanced Research Projects Agencyu nder contract N00014-85-K-0163 from the Office of Naval Research. Paper accepted March 1988. Discussion ends July 1989.
- configuration space
- qualitative reasoning
- spatial reasoning
- symbolic reasoning