Tightly-Coupled Radar-Visual-Inertial Odometry

Morten Nissov; Mohit Singh; Kostas Alexis

cs.ROMar 24, 2026

Tightly-Coupled Radar-Visual-Inertial Odometry

Authors: Morten Nissov, Mohit Singh, Kostas Alexis

Organizations: All authors are with the Department of Engineering Cybernetics, O. S. Bragstads Plass 2D, Norwegian University of Science and Technology (NTNU), Trondheim, Norway.

Abstract

Visual-Inertial Odometry (VIO) is a staple for reliable state estimation on constrained and lightweight platforms due to its versatility and demonstrated performance. However, pertinent challenges regarding robust operation in dark, low-texture, obscured environments complicate the use of such methods. Alternatively, Frequency Modulated Continuous Wave (FMCW) radars, and by extension Radar-Inertial Odometry (RIO), offer robustness to these visual challenges, albeit at the cost of reduced information density and worse long-term accuracy. To address these limitations, this work combines the two in a tightly coupled manner, enabling the resulting method to operate robustly regardless of environmental conditions or trajectory dynamics. The proposed method fuses image features, radar Doppler measurements, and Inertial Measurement Unit (IMU) measurements within an Iterated Extended Kalman Filter (IEKF) in real-time, with radar range data augmenting the visual feature depth initialization. The method is evaluated through flight experiments conducted in both indoor and outdoor environments, as well as through challenges to both exteroceptive modalities (such as darkness, fog, or fast flight), thoroughly demonstrating its robustness. The implementation of the proposed method is available at: https://github.com/ntnu-arl/radvio.

Figures & tables

Fig. 1: The aerial platform used to evaluate the proposed method, shown flying in nominal, obscured (due to fog), and dark environments.

Fig. 2: Information flow of the proposed method from the imu , radar, and camera sensors to the odometry and local-map outputs.

Fig. 3: Correlation between the feature patch in the image and the voxels resulting from querying the feature depth.

Parameter		Value	Unit
	Starting Frequency	$60$	$\mathrm{GHz}$
Max.	Range	$20.013$	$\mathrm{m}$
Max.	Doppler	$3.995$	$\mathrm{m}\text{/}\mathrm{s}$
Res.	Range	$0.078$	$\mathrm{m}$
	Doppler	$0.133$	$\mathrm{m}\text{/}\mathrm{s}$
	Azimuth/Elevation	$29$	$\mathrm{\SIUnitSymbolDegree}$

TABLE I: Radar Chirp Configuration

Sequence	Length [ $\mathrm{m}$ ]	ape [ $\mathrm{m}$ ]				rpe ( $\Delta:$ 10\text{,}\mathrm{m}$$ ) [ $\mathrm{m}$ ]				Final Position Relative Drift [ $\mathrm{cm}\text{/}\mathrm{m}$ ]
Sequence	Length [ $\mathrm{m}$ ]	RIO	VIO	RadVIO	RRxIO [ 15 ]	RIO	VIO	RadVIO	RRxIO [ 15 ]	RIO	VIO	RadVIO	RRxIO [ 15 ]
forest1	$190$	$2.311$	0.722	$0.744$	$1.335$	$0.128$	$0.214$	0.068	$0.105$	$0.506$	$0.312$	$0.311$	0.187
forest2	$195$	$3.298$	$0.861$	0.631	$0.700$	$0.193$	$0.161$	0.089	$0.151$	$1.030$	$0.277$	0.224	$0.256$
forest3	$208$	$2.727$	$1.923$	1.378	$2.186$	$0.178$	$0.513$	$0.080$	0.079	$0.640$	$0.847$	0.566	$0.900$
forest4	$213$	$2.618$	$1.844$	1.568	$3.622$	$0.201$	$0.457$	$0.135$	0.118	$0.891$	0.415	$0.434$	$0.921$
indoor_dark1	$210$	0.371	$0.550$	$0.476$	$0.700$	$0.055$	$0.112$	0.045	$0.118$	$0.205$	$0.227$	$0.220$	0.112

TABLE II: Numerical Results of the Proposed Method across Forest , Indoors , and Field Environments

Fig. 4: Position estimates of the proposed method alongside ablations from forest4 . The image shows the good visual conditions of the Forest environment.

Fig. 5: Radar extrinsic estimation of RadVIO in forest1 with nominal and perturbed initial guesses. The perturbed initial guess is generated by applying a rotation of

80\text{\,}\mathrm{\SIUnitSymbolDegree}

about the radar sensor’s

y

axis to the nominal prior.

Fig. 6: Position estimates of the proposed method alongside ablations from indoor_dark1 . Even with onboard lighting, the image visibility, and therefore the availability of good features, varies across the environment.

Fig. 7: Position estimates of the proposed method alongside ablations from indoor_fog2 . One can clearly see how, in some regions, objects can be made out, whereas in others the image is rendered useless.

Fig. 8: Position estimates of the proposed method alongside ablations from field2 . The degraded radar performance is clear, resulting from limited clutter and high speeds (up to $11\text{\,}\mathrm{m}\text{/}\mathrm{s}$ ).

Fig. 9: The radar point cloud size, number of inliers from RadVIO measurement updates, and velocity estimates, all from field2 .

Robot Localization

Robot localization estimates a robot’s position and orientation from onboard sensors, maps, and observations of its surroundings.

See all 177 papers

Embedded Bare-Metal Radar-Inertial Odometry
Oct 5, 2026Nicolai Adil Øyen Aatif, Morten Nissov, Kostas AlexisUAV Navigation Radar-Inertial Odometry
Department of Engineering Cybernetics Norwegian University of Science and Technology O. S. Bragstads Plass 2D, 7034 Trondheim, Norway
Equivariant Filter for Radar-Inertial Odometry
Apr 24, 2026Giulio Delama, Jan Michalczyk, Morten Nissov +4Inertial Odometry Equivariant Filtering
Control of Networked Systems Group, University of Klagenfurt, Austria · Autonomous Robots Lab, Norwegian University of Science and Technology, Norway · Hexagon Robotics, Zurich, Switzerland
UNRIO: Uncertainty-Aware Velocity Learning for Radar-Inertial Odometry
Apr 15, 2026Jui-Te Huang, Tianshu Huang, Anthony Rowe +1Radar-Inertial Odometry Robot State Estimation
Carnegie Mellon University, Pittsburgh, PA 15213, USA

Oct 5, 2026J/K move · Enter open · S save

Embedded Bare-Metal Radar-Inertial Odometry

Nicolai Adil Øyen Aatif, Morten Nissov, Kostas Alexis

Department of Engineering Cybernetics Norwegian University of Science and Technology O. S. Bragstads Plass 2D, 7034 Trondheim, Norway

UAV Navigation Radar-Inertial Odometry Robot State Estimation

Open paper AskPDF arXiv

Compact extraplanetary rovers and micro aerial vehicles require robust state estimation frameworks designed to operate under strict computational constraints in unforgiving environments. Typical solutions involving vision- or LiDAR-based sensing are computationally expensive and vulnerable to environments with perceptual degradation, making them poorly suited for resource-constrained platforms and austere conditions. Alternatively, Frequency Modulated Continuous Wave (FMCW) radar offers both robustness and computational efficiency by directly providing velocity measurements coupled with inherent resilience to perceptual degradation. These factors enable robust estimation, thereby reducing reliance on human operators for supervision to ensure platform safety in challenging environments. In this manuscript, we propose an embedded radar-inertial estimator tailored for low-compute platforms. All sensor drivers, data processing, and aided inertial navigation are performed on a single-core microcontroller, demonstrating its computational efficiency. Flight experiments show translational APE of

0.51-0.82\,\mathrm{m}

and RPE of

<

3,% against motion capture, alongside closed-loop flight through an unmodified PX4 stack. The firmware and printed circuit board design are openly available on GitHub at https://github.com/ntnu-arl/embedded_rio and https://github.com/ntnu-arl/embedded_rio-pcb respectively.

Abstract

Robot Localization

Explore similar work

Embedded Bare-Metal Radar-Inertial Odometry

Equivariant Filter for Radar-Inertial Odometry

UNRIO: Uncertainty-Aware Velocity Learning for Radar-Inertial Odometry