This paper describes the perception pipeline linking two robots in the PERSEPHONE autonomous mining mission. A light Explorer robot maps an unknown mine and builds a 3D scene graph of inspection targets, finding mineral deposits from natural-language prompts with zero-shot vision-language segmentation; a second Inspector robot with a heavier payload uses the map and graph to plan close-range viewpoints. The geometric steps (per-view boxes, cross-view merging, plane fitting, polygon extraction) are tested in a subterranean test facility and an active magnesite mine.
Relevance
A scene graph handed from one robot to another with different sensors is a concrete shared representation across a heterogeneous team, under degraded underground conditions. See 3D scene graphs and multi-robot multimodal perception.
Links
- arXiv
- Subfield: Multi-robot systems, Semantic mapping