Communication Efficient Cooperative Visual-Range-Inertial Localization

In this paper, the cooperative navigation issue is investigated for a group of robots. A distributed estimation architecture fusing range and visual measurements is proposed. Both the relative range and co-observed features are utilized to construct direct and indirect geometric constraints between robots respectively. To solve the difficult problem of calculating the cross covariance between local estimators, the covariance intersection (CI) algorithm is adopted to design a fully distributed and scalable cooperative estimator. To avoid unnecessary data transmission in feature matching process between different robots, a frustum view overlapping prediction algorithm is developed, which works effectively on reducing data transmission. Compared with existing algorithm, the proposed estimator achieves better accuracy with lower communication cost.