Do we always need a filter?

TL;DR

This paper challenges the traditional reliance on filters for state estimation, demonstrating that simple observation-only inference can outperform filters in accuracy and speed, especially with advanced sensors and multiple data sources.

Contribution

It provides a quantitative analysis of when and why observation-only inference surpasses filters, highlighting its advantages in multi-sensor environments and complex tracking scenarios.

Findings

O2 inference can outperform filters in accuracy and computational efficiency.

Multi-sensor O2 inference effectively handles clutter and multi-target tracking without background info.

Filters are not always necessary for effective state estimation, especially with advanced sensors.

Abstract

Since the groundbreaking work of the Kalman filter in the 1960s, considerable effort has been devoted to various discrete time filters for dynamic state estimation, especially including dozens of different types of suboptimal implementations of the Bayes filters. This has been accompanied by the rapid development of simulation/approximation theories and technologies. While admitting the success of filters in many cases, this study investigates the failure cases when they are in fact ineffective for state estimation. Several classic models have shown that the straightforward observation-only (O2) inference that does not need system modeling can perform better (in terms of both accuracy and computing speed) for estimation than filters. Special attention has been paid to quantitatively analyze when and why a filter will not outperform the O2 inference from the information fusion perspective. Thanks to the rapid development of advanced sensors, the O2 inference is not only engineering friendly and computationally fast but can also be very accurate and reliable by fusing the information received from multiple sensors. The statistical attributes of the multi-sensor O2 inference are analyzed and demonstrated through simulations. In the situation with limited sensors, the O2 approach can work jointly with existing clutter filtering and data association algorithms for multi-target tracking in clutter environments. Given an adequate number of sensors, the O2 approach can employ the multi-sensor data fusion to deal with clutter and can handle the very general multi-target tracking scenario with no background information.