Improving robustness of terrain-relative navigation for AUVs in regions with flat terrain

Terrain-Relative Navigation (TRN) is an emerging technique for localizing a vehicle in underwater environments. TRN offers a means of augmenting an INS/DVL dead-reckoned solution with continuous position hxes based on correlations with pre-stored bathymetry maps instead of surfacing for periodic GPS hxes. TRN accuracy on the order of 3m has been demonstrated in recent held trials using MBARIs Dorado-class AUVs in Monterey Bay. However, these TRN algorithms have occasionally converged to incorrect solutions when the AUV operates for extended times over featureless terrain. Specihcally, the TRN hlter can become overconhdent in an incorrect position hx. This paper demonstrates that the cause of these false hxes in information-poor regions is an incorrect accounting of map uncertainty and sensor noise in standard TRN hlter implementations, and offers a modihcation to the algorithms that can eliminate the false-hxes. Specihcally, standard TRN algorithms assume that map noise and vehicle sensor noise can be lumped together when performing measurement updates. In regions where the ratio of terrain information to map error is low, this assumption leads to underestimation of position uncertainty as the hlter essentially converges on noise in the map. Adjusting the hlter variance to depend on the estimated terrain information in addition to map error and sensor error provides a more robust TRN solution. An improved algorithm is described which adjusts the filter variance using a technique employed by the robotics and statistics community for reducing the likelihood of overconfidence. The advantage of this adjusted variance technique is that it permits nominal convergence rates of the TRN filter over information rich terrain while mitigating the risk of false fixes in information poor terrain. The effectiveness of the modified TRN algorithm is demonstrated in simulations using held data from MBARI AUV runs over flat terrain in Monterey Bay.