Robust GPS/INS-Aided Localization and Mapping Via GPS Bias Estimation

Summary

We consider the problem of pose estimation in the context of outdoor robotic mapping. In such cases absolute position information from GPS is often available. However, the peculiarities of GPS can lead to significant inconsistencies in mapping when a naive approach is used. We thus present a two-stage pose estimation system to address this problem. The first stage consists of a best-effort “blind” pose estimator based on a robust and extensible Rao-Blackwellized particle filtering framework. The estimate from this stage is then fed to a “seeing” HMM-style filter that attempts to infer the uncorrected bias of the first stage by matching stereo maps under an assumption of scene rigidity. Results are shown that demonstrate a vast improvement in pose estimates and map consistency using this method over the naive approach.