GPS World, October 2009

Indoor Navigation DEFENSE GOVERNMENT 25 20 15 10 5 0 n 2 inu only dist w o anchor dist w 1 anchor dist w 2 anchors 0 5000 10000 Frame absolute RMS err m 25 20 15 10 5 0 Anchor s deployed at f 3000 INU dx 02m dy 02m dz 0001m MSSI d 1m n 4 0 5000 10000 Frame absolute RMS err m 25 20 15 10 5 0 σ σ σ σ n 6 0 5000 10000 Frame absolute RMS err m 25 20 15 10 5 0 n 8 0 5000 10000 Frame absolute RMS err m 25 20 15 10 5 0 n 10 0 5000 10000 Frame absolute RMS err m 25 20 15 10 5 0 n 12 0 5000 10000 Frame absolute RMS err m 25 20 15 10 5 0 n 14 0 5000 10000 Frame absolute RMS err m 25 20 15 10 5 0 n 16 0 5000 10000 Frame absolute RMS err m Unaided Operation Contact with Anchor Error Reset 1 and 2 Anchors RF aided Operation p FIGURE 12 Simulation validation of Reset Effect for groups of mobile nodes encountering anchor reference beacons after their position estimates have degraded for some time This opportunistic encounter resets the error level of the group by an amount influenced by the number of fixed and mobile nodes rate appropriate for a four node team Two thirds of the way through the simulation the small teams merge at which point their absolute errors decrease via the Reset Effect to the level expected as if they had been working together all along The error level and error growth rate resume at the levels expected for an eight node team Reset error levels are independent of the history of ranging activities among nodes long term error growth is insensitive to the intermittent dynamics splitting and joining of the group Implementation Considerations Real time performance was a primary motivating factor driving development of these multi sensor fusion algorithms The optimization based algorithms required approximately an order of magnitude more processing power than the EKFbased approaches The KF and EKF are common components in many real time systems suggesting that successful implementation is likely Furthermore the distributed implementation of both algorithms was shown to be up to two orders of magnitude more computationally effi cient depending on network size The CPU requirements for the distributed algorithms are divided among multiple nodes drastically reducing per node processing load and improving algorithm scalability as the network size grows Communications Assuming three iterations per estimate cycle plus one additional dissemination of the final current estimate for a network of 32 nodes that compute position estimates at 1 Hz the system requires a network capacity of 131 Kbps This data rate is easily sustained with modern commercial broadband wireless networks and lies within reach of short range tactical military networks Inter node ranging rate can drop well below 1 Hz with minimal performance Two 4 node teams merge teams operate independently nMCRs 50 INS Stds dx 02 m dy 02 m dz 0001 m MSSI std d 1 m 8 node team splits dynamical nodes n 8 n 4 Error Reset σ σ σ σ n 8 n 4 n 8 Ö Ö Ü ÖÜ Ü 2000 4000 6000 8000 10000 12000 14000 15 10 5 0 absolute RMS err m 2000 4000 6000 8000 10000 12000 14000 1 0 2000 4000 6000 8000 10000 12000 14000 1 0 relative RMS err m 2000 4000 6000 8000 10000 12000 14000 1 0 Frame p FIGURE 13 The Reset Effect collaborative error reduction is apparent in the case of teams splitting and subsequently merging Even after position errors have grown during independent operation the error is reset to the level expected had the group been together all along degradation on long term absolute position accuracy Future Directions Our future work will focus on improved modeling and simulations of INU sensor drift and range measurement errors design of a generalized data generation tool to create more realistic node group mobility models simulation of more realistic scenarios involving additional heterogeneous navigation sensors and more detailed examination of algorithm performance Acknowledgments The authors thank Fred Vannozzi Mike Bunting Bob Evans Joanne Evans Ansley Jessup and Scott Kruse for supporting this work through empirical sensor data collection and evaluation This work was funded in part by Offi ce of Naval Research contract N0014 06 C 0489 This article is based on material presented at the Joint Navigation Conference 2009 co sponsored by the Joint Service Data Exchange and the Institute of Navigation This version omits many equations and a diagram of the EKF an online posting at www gpsworld com fusion shows them SHUNGUANG WU is a member of technical staff at Sarnoff Corporation and received his Ph D in electrical engineering from Wright State University JIM KABA is a senior member of technical staff at Sarnoff Corporation and received his B S in electrical engineering from Rutgers University SIUN CHUON MAU is a senior research scientist at Telcordia Technologies Inc and received his Ph D in physics from Princeton University The work described in this article was done while he was at Sarnoff Corporation TAO ZHAO is now a senior research engineer at Intuitive Surgical Inc and received his Ph D in computer science from the University of Southern California The work described in this article was done while he was at Sarnoff Corporation www gpsworld com October 2009 GPS World 47