Hierarchical Width-Based Planning and Learning

Authors

  • Miquel Junyent Universitat Pompeu Fabra
  • Vicenç Gómez Universitat Pompeu Fabra
  • Anders Jonsson Universitat Pompeu Fabra

DOI:

https://doi.org/10.1609/icaps.v31i1.15999

Keywords:

Reinforcement Learning Using Planning (model-based, Bayesian, Deep, Etc.), Learning To Improve The Effectiveness Of Planning & Scheduling Systems, Applications That Involve A Combination Of Learning With Planning Or Scheduling, Theoretical Aspects Of Planning And Learning

Abstract

Width-based search methods have demonstrated state-of-the-art performance in a wide range of testbeds, from classical planning problems to image-based simulators such as Atari games. These methods scale independently of the size of the state-space, but exponentially in the problem width. In practice, running the algorithm with a width larger than 1 is computationally intractable, prohibiting IW from solving higher width problems. In this paper, we present a hierarchical algorithm that plans at two levels of abstraction. A high-level planner uses abstract features that are incrementally discovered from low-level pruning decisions. We illustrate this algorithm in classical planning PDDL domains as well as in pixel-based simulator domains. In classical planning, we show how IW(1) at two levels of abstraction can solve problems of width 2. For pixel-based domains, we show how in combination with a learned policy and a learned value function, the proposed hierarchical IW can outperform current flat IW-based planners in Atari games with sparse rewards.

Downloads

Published

2021-05-17

How to Cite

Junyent, M., Gómez, V., & Jonsson, A. (2021). Hierarchical Width-Based Planning and Learning. Proceedings of the International Conference on Automated Planning and Scheduling, 31(1), 519-527. https://doi.org/10.1609/icaps.v31i1.15999