Communication-Efficient Relative Pose Estimation with Vision Foundation Models for Ephemeral Collaborative Perception

1 min read

CERPE estimates relative pose between robots that meet briefly, over limited bandwidth, often with little or no shared view. Robots share small fixed-size descriptors all the time and request raw images only when the descriptors suggest overlap; when views do not overlap, relative pose is carried forward through metrically scaled ego-motion from vision foundation models. In simulation and on real robots it improves 6-DoF relative pose over the chosen baselines.

Relevance

Field teams meet in short windows over bad links, which is the normal case at sea and in the air. It is a practical piece of collaborative SLAM for multi-robot perception, though poses propagated without overlap need some check on drift.