Using neural networks to improve estimation techniques

What is it about?

Nonlinear estimation scenarios often present challenges for current state-of-the-art filtering techniques, potentially resulting in issues such as elevated computational costs and divergence problems. This paper introduces a new and improved learning framework designed for aneural-network-based Gaussian nonlinear filter, which demonstrates promising improvementsin terms of consistency and accuracy when compared to prior approaches and other Gaussianfilters. In this approach, a neural network is used to approximate a nonlinear measurement update equation, while the estimated state uncertainty is characterized using two different techniques: the unscented transform and Monte Carlo sampling.

Featured Image

Photo by NASA on Unsplash

Photo by NASA on Unsplash

Why is it important?

This new iteration of neural-network-based filters introduces a different perspective on achieving consistent estimation in nonlinear problems.