Robust adaptive terrain-relative navigation

Terrain-Relative Navigation (TRN) is a technique for localizing a vehicle in GPS-denied environments. TRN augments a dead-reckoned solution with continuous position fixes based on correlations with a pre-stored map. In underwater applications TRN accuracy on the order of 3m has been demonstrated, however convergence to incorrect solutions has been observed when operating for extended periods over featureless terrain. Specifically, the TRN filter can become overconfident in an incorrect position fix. Previous work by the authors introduced an adaptive technique for mitigating over-confidence in uninformative terrain. Specifically, the algorithm exponentially down-weights the probabilities with a factor α between zero and one, based on the estimated terrain information. This paper focuses on understanding the source of false fixes in uninformative terrain, and uses this insight to develop the method behind the adaptive technique. This paper shows that the cause of false fixes using standard TRN weighting in information-poor regions is the assumption that the terrain is uncorrelated. It also introduces a method to analyze the probability of false peaks using the standard TRN measurement weighting. This analysis is an extension of work in the statistics community on robust adjusted likelihood ratios, and is used to bound the probability of large false peaks. The resulting robust, adaptive technique is capable of realtime operation and its effectiveness is demonstrated in simulation and with field data from MBARI AUV runs over flat terrain in Monterey Bay.