On providing prior knowledge for learning relational search heuristics

In this paper, we propose the use of two relational learning systems, hamlet and evock, for acquiring useful search control heuristics in the context of automated task planning. In particular, we discuss the influence of different ways of providing prior background knowledge to such systems. We compare the results of providing initial information by means of a human-centered approach against two automated approaches. The first automated one consists of using the output of hamlet as input to the learning process of evock, and viceversa. The second automated approach consists of using another planner for providing guidance towards solutions of problems.