This paper presents AMC-Nav, a novel architecture for multi-room navigation in embodied AI that mimics hippocampal memory consolidation mechanisms. It features a dual-store memory architecture, a sleep-inspired consolidation mechanism, and an adaptive retrieval system, leading to significant improvements in success rate and path efficiency over state-of-the-art baselines.
Key findings
AMC-Nav improves success rate by 14.6% and path efficiency by 11.2% over baselines.
Adaptive consolidation enables superior generalization to novel room configurations.
The architecture shows robust performance under partial observability.
Limitations & open questions
The paper does not discuss the scalability of AMC-Nav to larger or more complex environments.
The long-term effects of continuous memory consolidation cycles on system performance are not explored.