On the Characterization and Limits of 4D Radar for Aided Inertial Navigation

2026-05-03

AI Summary

This work addresses the challenge of unreliable state estimation in inertial navigation aided by 4D FMCW radar, where complex measurements and high noise levels degrade performance. By deriving a state-dependent noise model from first principles of radar signal processing, the authors formulate a factor graph optimization framework that leverages first-order error propagation to accurately approximate the covariance structure. This approach represents the first integration of first-principles-based noise modeling with factor graph optimization, significantly enhancing navigation accuracy and robustness—particularly in non-standard scenarios beyond the radar’s nominal operating range. The theoretical claims are validated through real-world flight experiments, and the associated data and code have been made publicly available.

Abstract

Frequency Modulated Continuous Wave (FMCW) radar is a promising sensor for aided inertial navigation, due to its robustness in environments that challenge traditional alternatives, such as LiDAR and vision. However, its widespread adoption is hindered by complex, noisy measurements, which make reliable estimation difficult. This manuscript addresses these challenges by analyzing the fundamental measurement relations of FMCW radar sensing and developing a reliable estimator. Noise models are derived by applying first principles to the underlying signal processing of a typical radar sensor. These models guide the design of a factor graph-based estimator, utilizing a first-order approximation for the measurement noise propagation. The approach is first examined through simulation, evaluating the significance of different noise sources, the validity of the first-order approximation, and the state-dependent nature of the covariance expressions. Extensive experiments demonstrate the superior robustness and accuracy of the proposed method across diverse field environments and flight profiles, including beyond the radar's standard operating range. Furthermore, the experiments confirm the insights from the simulation regarding the behavior and performance of different estimator configurations relative to their operating conditions. The evaluation data and estimator implementation are made available at https://github.com/ntnu-arl/rig.

Problem

• 4D Radar
• Aided Inertial Navigation
• FMCW Radar
• Measurement Noise
• State Estimation

Innovation

• FMCW radar
• noise modeling
• factor graph
• inertial navigation
• first-order approximation