Utilizing high-dimensional features for real-time robotic applications: Reducing the curse of dimensionality for recursive Bayesian estimation
Jie Li · P. Ozog · J. Abernethy · R. Eustice · M. Johnson-Roberson
TL;DR
Deep features are compressed into a space linear in the state, with perturbation-based uncertainty, enabling real-time Kalman filtering for sonar-based underwater localization.
BRIEF
Learned image features are rich but high-dimensional, which makes them hard to use as measurements in filters like the Kalman filter that robots rely on for real-time state estimation. This work maps those features into a lower-dimensional space that relates linearly to the robot's state, and estimates the uncertainty of the output by perturbing the data. The approach is demonstrated on underwater localization with an imaging sonar, running in real time with accuracy similar to benchmark methods.
Abstract
Feature learning has become popular in robotics due to recent advances in machine learning. In this paper, we propose a novel method to utilize the high-dimensional features from these techniques as observations in Bayesian estimation problems in a real-time manner. We develop an approach that: 1) pre-processes the observations and maps them into a new space with both reduced dimensions and a linear relationship to the estimation states; and 2) estimates the uncertainty of resulting outputs using data perturbation. The result is that deep learning approaches can be combined with more traditional filtering approaches like the Kalman filter (KF) to achieve state-of-the-art real-time performance. We validate the method by presenting the first real-time application of underwater robot localization using an imaging sonar. The proposed technique shows similar localization accuracy to benchmark approaches while simultaneously achieving real-time performance.