Multipurpose Strategic Planning In the Game of Go Paper presentation Authors: Shui Hu and Paul E. Lehner Presentation by: Adi Botea
Outline Introduction Description of the Approach Example Conclusion
Introduction Go = strategy + tactics Current Go programs: –Valuable work in tactics BUT… –Lack of strategic focus So, this paper proposes: –Strategic & tactical planning in Go –A new approach on multipurpose moves
General Description of the Approach Static analysis to determine possible goals Dynamic look ahead to further estimate goal consequences Evaluate goals and goal interactions Select the best strategy (goal or set of goals) Perform detailed look ahead to find the best way to implement the plan
More Details on Approach Object hierarchies: –Stones, strings, chains, and groups Current Goal Structures (CGSs) –Hierarchies that characterize goals that may be pursued in a given position
Current Goal Structures
Static Goal Generation Example of strategic goal definition Example of tactical goal definition
Look Ahead Algorithm Similar to tactical search Goal Stacks for both players Given a hypothetical position:
Goal Achievability & Composition
Planning Procedure
Example - ORSIM
Example – ORSIM (cont.)
Conclusion & Results Multipurpose planning in Go –Includes strategy formalization –Catches interactions between moves Reasonable computational complexity Planner “is very strong on […] selected problems”