SLAMuZero: Plan and Learn to Map for Joint SLAM and Navigation

• DOI: 10.1609/icaps.v34i1.31476
• Publication Date: 2024-05-30T13:15:20Z
• Authors: Bowen Fang; Xu Chen; Zhengkun Pan; Xuan Di
• Keywords: Baseline (sea); Tree (set theory)

MuZero has demonstrated remarkable performance in board and video games where Monte Carlo tree search (MCTS) method is utilized to learn adapt different game environments. This paper leverages the strength of enhance agents’ planning capability for joint active simultaneous localization mapping (SLAM) navigation tasks, which require an agent navigate unknown environment while simultaneously constructing a map localizing itself. We propose SLAMuZero, novel approach SLAM navigation, employs process that uses explicit encoder-decoder architecture mapping, followed by prediction function evaluate policy value based on generated map. SLAMuZero outperforms state-of-the-art baseline significantly reduces training time, underscoring efficiency our approach. Additionally, we develop new open source library implementing flexible modular toolkit researchers practitioners (https://github.com/bwfbowen/SLAMuZero).