Increasing the Runtime Speed of Case-Based Plan Recognition.

摘要

We present PPC (Plan Projection and Clustering), an algorithm that creates a plan hierarchy for case-based plan recognition systems. PPC is motivated by a desire to improve the response time of robots working in collaboration with humans. It projects the plans of a case base into a Euclidean space and iteratively clusters plans within that space, producing an abstraction hierarchy. Case retrieval traverses down this hierarchy and requires fewer comparisons than a search of the corresponding flat case base. Our approach also has the advantage that it does not require substantial domain knowledge. We report PPC's empirical performance on synthetically generated plans, showing that it increases runtime speed without substantially reducing plan retrieval accuracy when the plans are generated using a non-random distribution.