Search and planning under incomplete information : a study using Bridge card play

• Main Author: Frank, Ian
• Published: University of Edinburgh 1996
• Subjects: 510
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.651086

This thesis addresses the task of problem-solving in domains with incomplete information and multiple agents with opposing goals. In particular, we describe FINESSE - a system that forms plans for the problem of declarer play in Bridge. Firstly, we examine the problem of search. We formalise a reduced form of incomplete information games in which equilibrium point strategies can be identified, and identify two specific problems that can affect algorithms in such domains. In Bridge, we show that the reduced form corresponds to the typical model analysed in expert texts, and examine search algorithms which overcome the problems we have identified. Secondly, we look at how planning algorithms can be made to cope with the difficulties of such domains. This calls for the development of new techniques for representing uncertainty and actions with disjunctive effects, for coping with an opposition, and for reasoning about compound actions. The original motivation for this work came from the field of mathematical reasoning. In Edinburgh, the technique of proof-planning has been used to control the search for the proofs in theorems. The defining feature of this paradigm is that it restricts the available options at any stage of planning to pre-determined set of possibilities. By applying proof-planning techniques to the domain of Bridge, FINESSE becomes capable of constructing plans that look ahead right to the end of the tree of possible actions. Against human players it is therefore less prone to the short-sightedness that afflicts the traditional look-ahead approach. This allows it to solve problems that are beyond the range of previous systems. A side-effect of formalising the allowed ways to play the game is that we arrive at a deeper understanding of the game itself. The high-level nature of the objects in FINESSE's plans and the use of a qualitative uncertainty representation language give the system the ability to produce meaningful textual explanations for its actions.