Deep Inference for Covariance Estimation: Learning Gaussian Noise Models for State Estimation

Katherine Liu, Kyel Ok, William Vega-Brown, Nicholas Roy

2018 · 57 citations · 18 references

DOIFull text

Open access

Concepts

TL;DR

Existing predictive sensor modeling outperforms conventional fixed models but requires extensive domain knowledge that affects accuracy and computational cost. The study introduces DICE, a deep neural network that estimates measurement covariance directly from raw sensor data, aiming to model uncertainty as a function of the measurement itself. The method learns a representation of the measurement model by supervised regression on pairs of raw sensor measurements and ground‑truth errors, using a deep neural network to predict covariance without hand‑coded features or parametric forms. The approach is sensor‑agnostic and achieves improved covariance prediction on both simulated and real datasets.

Abstract

We present a novel method of measurement covariance estimation that models measurement uncertainty as a function of the measurement itself. Existing work in predictive sensor modeling outperforms conventional fixed models, but requires domain knowledge of the sensors that heavily influences the accuracy and the computational cost of the models. In this work, we introduce Deep Inference for Covariance Estimation (DICE), which utilizes a deep neural network to predict the covariance of a sensor measurement from raw sensor data. We show that given pairs of raw sensor measurement and ground-truth measurement error, we can learn a representation of the measurement model via supervised regression on the prediction performance of the model, eliminating the need for hand-coded features and parametric forms. Our approach is sensor-agnostic, and we demonstrate improved covariance prediction on both simulated and real data.

References

18