GPS-limited cooperative localization using scalable approximate decentralized data fusion

This work addresses the problem of communication-based autonomous vehicle cooperative inertial navigation and localization in GPS-challenged environments. Specifically, we examine decentralized state estimation strategies that allow vehicles to opportunistically augment their onboard navigation filters with moving maps obtained from exchanged relative sensing and dynamic target tracking information between vehicles. While decentralized state estimation theoretically allows agents to share information efficiently with one another in a scalable, asynchronous, and ad hoc manner, several key technical issues arise. Firstly, augmentation of each vehicles navigation filters for decentralized estimation requires maintaining correlations between ownship vehicle pose states, ownship nuisance states (e.g. rate gyro biases), and pose states for all other tracked vehicles; this leads to expensive and unscalable onboard filtering requirements. Secondly, the use of distinct strapdown navigation and target tracking models onboard each vehicle implies that decentralized state estimation must occur across heterogeneous state space models; this case is usually not handled by conventional decentralized estimation theory. We present two novel strategies for addressing these problems in the context of cooperative navigation: the approximate channel filter, and factorized covariance intersection. We provide simulated examples to illustrate these approaches in simplified and realistic cooperative navigation settings, including a cooperative UAV-UGV navigation application based on recorded UAV flight data.