Exploiting the solution structure knowledge to speed up non-learning planner

The modern state of the art planners are highly effective and has strong handing capability, but most of them can´t learn anything from previous experiences. In the past there have been many researches on learning problem in planning and make some progress. However, the knowledge used in these methods is not easy to learn and use such that Learning can often make performance degrade, learning did not improve overall performance compared to best non-learning planners. In this paper, we present a novel knowledge, plan solution´s structure knowledge, which is simple and easy to learn and use, in our methodology, each time a problem solved successfully, planner will analysis the solution and extract its structure knowledge, then save the solution´s structure knowledge in the planning domain description document as prior knowledge. In the future, when meeting the same or similar problem again, the planner will firstly read prior knowledge in the domain, and reconstruct solution´s structure, then the solution extraction process will be carried out to determine the final solution. We incorporate this method to GraphPlan and obtain WgraphPlan system. Experimental result shows that WgraphPlan based on this method can reduce enormously backtrack times, the efficiency enhancement is highest reaches 25%.