Optimizing Long-Running Action Histories in the Situation Calculus Through Search
- Cite this paper as:
- Ewin C., Pearce A.R., Vassos S. (2015) Optimizing Long-Running Action Histories in the Situation Calculus Through Search. In: Chen Q., Torroni P., Villata S., Hsu J., Omicini A. (eds) PRIMA 2015: Principles and Practice of Multi-Agent Systems. PRIMA 2015. Lecture Notes in Computer Science, vol 9387. Springer, Cham
Agents are frequently required to perform numerous, complicated interactions with the environment around them, necessitating complex internal representations that are difficult to reason with. We investigate a new direction for optimizing reasoning about long action sequences. The motivation is that a reasoning system can keep a window of executed actions and simplify them before handling them in the normal way, e.g., by updating the internal knowledge base. Our contributions are: (i) we extend previous work to include sensing and non-deterministic actions; (ii) we introduce a framework for performing heuristic search over the space of action sequence manipulations, which allows a form of disjunctive information; finally, (iii) we provide an offline precomputation strategy. Our approach facilitates determining equivalent sequences that are easier to reason with via a new form of search. We demonstrate the potential of this approach over two common domains.
Unable to display preview. Download preview PDF.