Single‐Shot Initial Mutual Localization for Micro Aerial Swarms

ABSTRACT Initial mutual localization of micro aerial swarms remains a challenging problem and is essential for establishing a common reference frame before coordinated flight. This task is difficult due to the limited scene overlap between nonadjacent drones, as well as sparse and low‐quality feature correspondences in environments with insufficient structural texture. To address these challenges, we propose a single‐shot dual‐view pose estimation method that performs initial mutual localization using a single‐view capture from each camera, without relying on multi‐view information. This method formulates feature matching and relative pose estimation within a unified optimization framework, which suppresses outliers and low‐quality matches while enabling the recovery of a larger set of high‐quality correspondences under the same visual conditions. As a result, the proposed approach significantly improves pose estimation accuracy and robustness. Extensive benchmark evaluations demonstrate that our method consistently outperforms the standard dual‐view pipeline based on feature matching and pose estimation across diverse scenarios, with particularly strong performance in challenging field environments. We further validate the proposed initial mutual localization method on a real micro aerial swarm platform, and release an open‐source implementation ( https://github.com/lyf‐FATAS/rpe ) for reproducibility and future research.