Learning-Based Planning

Automated Planning (AP) studies the generation of action sequences for problem solving. A problem in AP is defined by a state-transition function describing the dynamics of the world, the initial state of the world and the goals to be achieved. According to this definition, AP problems seem to be easily tackled by searching for a path in a graph, which is a well-studied problem. However, the graphs resulting from AP problems are so large that explicitly specifying them is not feasible. Thus, different approaches have been tried to address AP problems. Since the mid 90’s, new planning algorithms have enabled the solution of practical-size AP problems. Nevertheless, domain-independent planners still fail in solving complex AP problems, as solving planning tasks is a PSPACE-Complete problem (Bylander, 94). How do humans cope with this planning-inherent complexity? One answer is that our experience allows us to solve problems more quickly; we are endowed with learning skills that help us plan when problems are selected from a stable population. Inspire by this idea, the field of learning-based planning studies the development of AP systems able to modify their performance according to previous experiences. Since the first days, Artificial Intelligence (AI) has been concerned with the problem of Machine Learning (ML). As early as 1959, Arthur L. Samuel developed a prominent program that learned to improve its play in the game of checkers (Samuel, 1959). It is hardly surprising that ML has often been used to make changes in systems that perform tasks associated with AI, such as perception, robot control or AP. This article analyses the diverse ways ML can be used to improve AP processes. First, we review the major AP concepts and summarize the main research done in learning-based planning. Second, we describe current trends in applying ML to AP. Finally, we comment on the next avenues for combining AP and ML and conclude.