Generalized Rapid Action Value Estimation in Memory-Constrained Environments
2026-02-26 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors studied a method called GRAVE used to help computers play many different games by looking ahead at possible moves. While GRAVE works well, it needs a lot of memory to keep track of game details, which makes it hard to use on devices with limited memory. The authors created three new versions—GRAVE2, GRAVER, and GRAVER2—that use clever tricks like searching two steps ahead and reusing nodes to save memory. Their results show these new versions use much less memory but still play games just as well as the original GRAVE method.
Monte-Carlo Tree Search (MCTS)General Game Playing (GGP)GRAVESearch algorithmsMemory optimizationTwo-level searchNode recyclingGame AIWin/visit statisticsAlgorithm efficiency
Authors
Aloïs Rautureau, Tristan Cazenave, Éric Piette
Abstract
Generalized Rapid Action Value Estimation (GRAVE) has been shown to be a strong variant within the Monte-Carlo Tree Search (MCTS) family of algorithms for General Game Playing (GGP). However, its reliance on storing additional win/visit statistics at each node makes its use impractical in memory-constrained environments, thereby limiting its applicability in practice. In this paper, we introduce the GRAVE2, GRAVER and GRAVER2 algorithms, which extend GRAVE through two-level search, node recycling, and a combination of both techniques, respectively. We show that these enhancements enable a drastic reduction in the number of stored nodes while matching the playing strength of GRAVE.