Learning to win: case-based plan selection in a real-time strategy game
ICCBR'05 Proceedings of the 6th international conference on Case-Based Reasoning Research and Development
A case base planning approach for dialogue generation in digital movie design
ICCBR'11 Proceedings of the 19th international conference on Case-Based Reasoning Research and Development
| Hi-index | 0.00 |
Classical planning is useful because it generates a partially ordered set of steps to achieve a goal consistent with the planner''s world model. Situated activity is useful for agents which responds quickly to changes in the interactive world. Unfortunately, everyday-world activity cannot rely on the costly, brittle plans of a theorem proving, nor can it resort to the stateless decisions of a situated automaton to cope with delayed feedback from the world. The only way for an agent to carry out long-term tasks in complex, dynamic worlds is to build a planning system that retrieves competent plans quickly, yet recover when its expectations fail. Furthermore, when an agent returns to a similar problem, it should avoid repeating the same mistakes. Case-based planning is the only planning paradigm which accounts for both failure tolerant planning and learning from failure. By re-using cases, MAYOR quickly retrieves and executes a plan. By monitoring the expectations associated with every plan, MAYOR determines when a plan underperforms, and recovers from the expectation failure using an approximate, causal model of the world to determine the cause of the failure. Lastly, MAYOR stores away factors which predicted the failure, and avoids the the failure in the future. This paper describes how MAYOR uses plans in the complex, dynamic world SimCity.