NISB-Fusion: Multi-Agent Mapping and Map Merging With Neural Implicit Spatial Block

Recent advancements have demonstrated the potential of radiance representations for high-quality mapping and reconstruction. However, these methods face significant challenges in large-scale, multi-agent scenarios, particularly in terms of computational demands and transmission bandwidth requirements. To address these limitations, we propose a NeRF-based framework designed for multi-agent mapping in large-scale indoor environments. Our approach leverages Neural Implicit Spatial Blocks (NISB) to dynamically cover the scene during mapping. Descriptors extracted from NISB and their associated keyframes are used for top-down place recognition and hybrid submap merging, enabling effective integration. Experimental results show that our method outperforms existing implicit multi-agent mapping approach, delivering superior efficiency and accuracy.