A Comparison of Knowledge-Based GBFS Enhancements and Knowledge-Free Exploration

Abstract

GBFS-based satisficing planners often augment their search with knowledge-based enhancements such as preferred operators and multiple heuristics. These techniques seek to improve planner performance by making the search more informed. In our work, we will focus on how these enhancements impact coverage and we will use a simple technique called epsilon-greedy node selection to demonstrate that planner coverage can also be improved by introducing knowledge-free random exploration into the search. We then revisit the existing knowledge-based enhancements so as to determine if the knowledge these enhancements employ is offering necessary guidance, or if the impact of this knowledge is to add exploration which can be achieved more simply using randomness. This investigation provides further evidence of the importance of preferred operators and shows that the knowledge added when using an additional heuristic is crucial in certain domains, while not being as effective as random exploration in others. Finally, we demonstrate that random exploration can also improve the coverage of LAMA, a planner which already employs multiple enhancements. This suggests that knowledge-based enhancements need to be compared to appropriate knowledge-free random baselines so as to ensure the importance of the knowledge being used.

Understanding and Improving Local Exploration for GBFS

“…Future work should explore whether the same problem occurs in classical heuristic search domains, such as sliding tile puzzles (Korf and Taylor 1996). Valenzano et al (2014) show that replacing preferred operators with random actions can achieve about half the improvement of preferred operators. Similarly, replacing the secondary heuristic in multiple heuristics with a purely random heuristic achieves about half the improvement of multiple heuristics.…”

Learning Classical Planning Strategies with Policy Gradient

“…-greedy search -greedy search was first considered in classical planning context by (Valenzano et al 2014). Like greedy best-first search, one application of this routine performs a single node expansion.…”

Biased Exploration for Satisficing Heuristic Search

“…In local minima, a good node, i.e., one that makes progress toward the goal node, has a higher heuristic value than other nodes, and GBFS needs to search through many nodes before reaching a good node. To escape local minima, a number of exploration methods have been developed (Imai and Kishimoto 2011;Valenzano et al 2014;. Usually, these algorithms perform GBFS most of the time but sometimes switch to exploration and select a node based on different criteria.…”

“…Usually, these algorithms perform GBFS most of the time but sometimes switch to exploration and select a node based on different criteria. Some of the algorithms such as -GBFS (Valenzano et al 2014) and Type-GBFS , which empirically per-form well, ignore the heuristic values during their exploration phase. However, although heuristics sometimes make mistakes, good heuristics should not be too inaccurate; intuitively, there should be some bound on the inaccuracy of a heuristic and exploration should consider this bound instead of completely ignoring heuristic values.…”