PROBE-GK: Predictive Robust Estimation using Generalized Kernels
This work addresses state estimation for computer vision and robotics in dynamic environments, offering a domain-specific improvement over existing methods.
The paper tackles the problem of inaccurate state estimation in dynamic environments where static uncertainty models fail, by developing a predictive robust estimator using fast nonparametric Bayesian inference to model sensor uncertainty. The result shows significant improvements in localization accuracy compared to fixed noise models on synthetic data, KITTI dataset, and an experimental platform.
Many algorithms in computer vision and robotics make strong assumptions about uncertainty, and rely on the validity of these assumptions to produce accurate and consistent state estimates. In practice, dynamic environments may degrade sensor performance in predictable ways that cannot be captured with static uncertainty parameters. In this paper, we employ fast nonparametric Bayesian inference techniques to more accurately model sensor uncertainty. By setting a prior on observation uncertainty, we derive a predictive robust estimator, and show how our model can be learned from sample images, both with and without knowledge of the motion used to generate the data. We validate our approach through Monte Carlo simulations, and report significant improvements in localization accuracy relative to a fixed noise model in several settings, including on synthetic data, the KITTI dataset, and our own experimental platform.