Output details

PbP2: Portfolio-based Planner

<22> PbP2 is the result of Vallati's PhD, that was focused on investigating different ways for efficiently exploiting automatically extracted learning for improving the performance of planning systems.

PbP was the first completely automatic portfolio-based planner able to effectively exploit knowledge for efficiently solving problems from a given domain. PbP2, an extended and enhanced version of PbP, is a multi-planner approach that automatically configures a portfolio of planning techniques for each given domain. The configuration process for a given domain uses a set of training instances to: (i) compute and analyze some alternative sets of macro-actions for each planner in the portfolio identifying a (possibly empty) useful set, (ii) select a cluster of planners, each one with a useful set of macro-actions, that is expected to perform best, and (iii) derive some additional information for configuring the execution scheduling of the selected planners at planning time.

PbP2 represents the state-of-the-art of learning-based planners: PbP was the overall winner of the learning track of the sixth International Planning Competition (IPC-6, 2008), and PbP2 was the winner of the learning track of the seventh International planning competition (IPC-7, 2011). The performances of PbP (and PbP2) inspired the exploitation and the development of portfolio approach in planning ("Learning Portfolios of Automatically Tuned Planners"-ICAPS 2012, "ArvandHerd: parallel planning with a portfolio"-IPC,2011 ); currently several portfolio-based planners, designed by the Artificial Intelligence Research Group of the University of Freiburg or Muller's research group at the University of Alberta, are available within the planning community.

The first paper that describes the PbP approach, "An Automatically Configurable Portfolio-based Planner with Macro-actions: PbP" (ICAPS,

2009) has 30 citations in google scholar.