Toward a fully decentralized architecture for multi-sensor data fusion

The paper describes a fully decentralized architecture for multi-sensor data-fusion. This architecture has no central processing facility and no centralized communications medium, it does not require any ad hoc structure, such as a hierarchy, to be imposed on it, and it does not require a controller to organize it. The structure of this architecture resembles a network of sensor nodes. Each sensor node has its own processing element, and its own communications facilities. It does not need to know what other sensors are in the network and it does not need to know a priori what information they can provide. Communications can take place between any two connected sensor nodes, and they can independently assimilate each others information. The starting point for this architecture is an algorithm which permits the full decentralization of the Extended Kalman Filter (EKF) equations. This algorithm is the Decentralized Kalman Filter (DKF). The DKF is an implementation of a multi-sensor EKF which has been divided up into modules, one associated with each sensor