Reusing previously found A∗ paths for fast goal-directed navigation in dynamic terrain

Abstract

Generalized Adaptive A∗ (GAA∗) is an incremental algorithm that replans using A∗ when solving goal-directed navigation problems in dynamic terrain. Immediately after each A∗ search, it runs an efficient procedure that updates the heuristic values of states that were just expanded by A∗ making them more informed. Those updates allow GAA∗ to speed up subsequent A∗ searches. Being based on A∗ it is simple to describe and communicate; however, it is outperformed by other incremental algorithms like the state-of-the-art D∗ Lite algorithm at goal-directed navigation. In this paper we show how GAA∗ can be modified to exploit more information from a previous search in addition to the updated heuristic function. Specifically, we show how GAA∗ can be modified to utilize the paths found by a previous A∗ search. Our algorithm-Multipath Generalized Adaptive A∗ (MPGAA∗)-has the same theoretical properties of GAA∗ and differs from it by only a few lines of pseudocode. Arguably, MPGAA∗ is simpler to understand than D∗ Lite. We evaluate MPGAA∗ over various realistic dynamic terrain settings, and observed that it generally outperforms the state-of-the-art algorithm D∗ Lite in scenarios resembling outdoor and indoor navigation.