Fig. 1. Structural diagram of interactions between canonical components used in information fusion. Two types of belief sources are shown: robotic sensor and human ui, as well as two types of belief sinks: robotic planner and human ui.

Fig. 2. A possible realization of the node component. Information flow to and from a node is also shown. Events corresponding to the four groups of messages occur asynchronously.

Fig. 3. Information flow in the Local Filter on arrival of a local observation. (Message 1 in Fig. 2.)

Fig. 4. Information flow in the Channel Filter on arrival of channel update message. (Message 2 in Fig. 2.)

Fig. 5. Information flow in the Local Filter on arrival of channel update message. (Message 2.1 in Fig. 2.)

Fig. 6. Information flow in the Channel Filter on local update. (Message 3 in Fig. 2.)

Fig. 7. Information flow in the Local Filter on sink update. (Message 4 in Fig. 2.)

Fig. 8. Structural diagram of the decision making task. Two types of action sources are shown: human ui and robotic planner. Services used in implementing the task of information fusion are omitted for clarity.

Fig. 9. Information flow inside planner in coordinated mode.

Fig. 10. Information flow inside planner in cooperative mode.

Fig. 11. Two-node information fusion example: (a) component diagram and (b) information flow between the component. Each node has a single local sensor. The two nodes are linked into a network through the Linkable interface.

Fig. 12. Motion and observation models for non-linear filtering: (a) Gaussian motion model and (b) probability of detection for an acoustic sensor [5]. Darker colors correspond to higher probability.

Fig. 13. Example of decentralized target tracking using nonlinear filtering techniques. The panels show platform positions (□) and current true target position (). Darker colors correspond to higher probability.

Fig. 14. Entropy evolution in the grid tracking example.

Fig. 15. The information fusion experiments: the “motion” map (a) and point feature tracking (b). Blue lines stand for DDF links. Tracking point features: (c) operator enters feature 2 and (d) platforms drive to and observe it.