Resilient Decentralized Radar-LiDAR-Inertial Odometry in Challenging Subterranean Environments

Moumita Mukherjee; Avijit Banerjee; Anton Koval; George Nikolakopoulos

cs.ROOct 23, 2026

Resilient Decentralized Radar-LiDAR-Inertial Odometry in Challenging Subterranean Environments

Authors: Moumita Mukherjee, Avijit Banerjee, Anton Koval, George Nikolakopoulos

Organizations: Robotics and AI Team, Department of Computer Science, Electrical Engineering and Space Engineering, Luleå University of Technology, Luleå, Sweden

Abstract

Accurate localization in GPS-denied and visually degraded environments affected by smoke, dust, or fog remains a critical challenge for autonomous systems. While LiDAR provides high-resolution spatial perception, its performance deteriorates under particulate occlusion. In contrast, frequency-modulated continuous-wave (FMCW) radar offers robustness under such conditions but suffers from sparse measurements and outliers. This article presents a resilient decentralized LiDAR Radar Inertial Odometry (LRIO) framework that fuses the complementary strengths of LiDAR, radar, and inertial sensing to achieve robust pose estimation in challenging subterranean environments. At the core of the framework is an Adaptive Residual Thresholding (ART) mechanism that autonomously detects temporal sensor degradation and selectively adjusts the fusion process to minimize unreliable measurements. Building on this, a Robust Error-Tolerant Optimal Information Filter (RETOIF) is employed within a decentralized fusion architecture to integrate LiDAR-inertial and multiple radar-inertial odometry estimates in a fault-aware manner, enhancing resilience to environmental disturbances and intermittent sensor failures. The proposed framework is evaluated using real-world sensor data collected in GPS-denied subterranean tunnels under smoke-free and smoke-degraded conditions. In the smoke-free experiment, the proposed RETOIF framework achieves a root mean square localization error (RMSE) of 2.1733, compared with 2.9929 for the conventional centralized LRIO approach, corresponding to a 27.4% reduction in RMSE. Experimental results further demonstrate the framework's ability to maintain pose estimation under smoke-induced sensor degradation, highlighting its potential for resilient autonomous navigation and mapping in challenging subterranean environments.

Figures & tables

Figure 1: The performance of a 3D Velodyne LiDAR and multiple millimetre-wave radars was evaluated in a GPS-denied Sub-T tunnel without smoke. LiDAR scans (black points) clearly represent the tunnel structure, while radar points (multi-coloured, each representing a different sensor) are sparse and lack discernible structure.

Figure 2: Performance evaluation of 3D Velodyne LiDAR and IWR6843AoP mmWave radars in GPS-denied tunnels under fog shows severe LiDAR degradation, while radar remains unaffected

Figure 3: Left: Experimental setup with six mmWave TI-IWR6843AoP radars ( $R_{1}$ – $R_{6}$ ), LiDAR, and IMU mounted on a Pioneer 3-AT2 platform. Radars $R_{1}$ – $R_{4}$ are planar, while $R_{5}$ – $R_{6}$ are tilted to capture ceiling observations. Right: Single-frame radar point cloud from the multi-radar setup, with colour-coded targets corresponding to individual radars.

Figure 4: The upper and lower bounds of the adaptive threshold are determined using the residual vectors, $\mathfrak{R}_{e_{k}}$ , at each time instance. A statistical analysis is conducted on $n$ residuals under the assumption that they follow a normal distribution with nonzero mean. These bounds define the region of error for the maximum likelihood functions and serve as an accurate adaptive threshold for integrating pose estimates from multiple estimators in the final layer. The range of these threshold bounds can be adjusted using the tuning parameter $\mathcal{K}$ , making the region dynamic and adaptive.

Figure 5: Generic Algorithm for the proposed decentralized Fusion (RETOIF)

Figure 6: RETOIF is designed for LiDAR Radar Inertial Odometry (LRIO). In this framework, RIO (1-R) denotes a single front-radar setup, RIO(2-R) uses front and rear radars, RIO (4-R) incorporates radars placed at the front, rear, left, and right, and RIO (6-R) uses a six-radar configuration.

Panel of Figure 7 — Figure 7: Representative experimental environment inside Sub-T tunnel with and without smoke. A map of the experimental area is shown in the upper left corner of each image and was generated from the 3D LiDAR measurements.

Parameter	Value
Laser Channels	16
Vertical Resolution	0.2°
Field of View (FOV)	360°
Range	100m
Rotation Rate	(1-100) Hz
Points per second	300,000 points/sec

Table 1: Configuration Parameters for 3D-Velodyne LiDAR

Parameter	Value
Frame-rate	8.64-10 GHz
Band width	60-64Hz
Maximum range	8-10 meters
Maximum doppler	8-10m/s
Range resolution	0.047
Doppler velocity resolution	3 cm/s

Table 2: Configuration Parameters for Millimeter-wave TI-IWR6843AoP Radar

Figure 8: The LRIO-based localization was compared with LIO. The radar body point clouds were transformed based on the pose estimates from the proposed decentralized framework. While the LiDAR can provide reliable measurements in the absence of smoke, its inherent sensor limitations result in less accurate odometry estimates compared to measurements obtained using a measuring tape.

Figure 9: The assessment of the first-layer likelihood functions (filters 1 and 2), incorporating the lower and upper threshold bounds, conducted in a smoke-free subterranean environment.

Figure 10: The assessment of the first-layer likelihood functions (filters 3 and 4), incorporating the lower and upper threshold bounds, conducted in a smoke-free subterranean environment.

Figure 11: The assessment of the first-layer likelihood function, with lower and upper bounds of residual thresholding, where Figure (b) represents estimated trajectories from all the filters compared with LIO

Figure 12: In this context, LIO is treated as the ground-truth reference, while acknowledging that each sensing modality has its own limitations. Notably, the Lidar radar inertial odometry (LRIO) closely follows the path recorded during the pre-experimental measurements.

Method	Mean Error	Std. Error	RMSE
Proposed (RETOIF)	1.6393	1.4270	2.1733
CF (LRIO)	2.1851	2.0453	2.9929
1-RIO	0.8599	0.4497	0.9704
2-RIO	0.9004	0.4996	1.0297
4-RIO	1.4840	1.7988	2.3319
6-RIO	2.9086	2.7226	3.9839

Table 3: Quantitative localization error statistics for the evaluated methods.

Panel of Figure 13 — Figure 13: Comparison of LiDAR-based localization and mapping performance in the experimental SubT tunnel under different visibility conditions: (a) no smoke, (b) dense smoke (Case-1), and (c) light smoke (Case-2). The results illustrate the progressive degradation of LiDAR-based localization and mapping in the presence of smoke.

Figure 14: Performance analysis of hypothesis testing based on the adaptive selection of the threshold region for the first-layer filters in the Case-2 smoke test scenario within Environment-2

Figure 15: Performance analysis of hypothesis testing based on the adaptive selection of the threshold region for the first-layer filters in the Case-2 smoke test scenario within Environment-2

Figure 16: Performance analysis of hypothesis testing based on the adaptive selection of the threshold region for the first-layer filters in the Case-2 smoke test scenario within Environment-2

Figure 17: Residual-based hypothesis evaluation for the two local filters under smoke conditions (Case-1), showing the estimated trajectories, residual bounds, and corresponding delta variations for LF-1 and LF-2.

Figure 18: Residual based hypothesis evaluation for LF-3 and LF-4 under smoke conditions (Case-1), showing trajectory estimates, residual bounds, and corresponding delta variations.

Figure 19: Residual and delta evaluation of LF-5, together with the trajectory performance of the individual first-layer filter estimates under smoke conditions (Case-1).

Figure 20: Comparison of all RIO configurations with LIO in a smoke-filled tunnel for Case-1 and Case-2 scenarios

Panel of Figure 21 — Figure 21: Comparison of trajectory estimation performance under smoke-degraded conditions. (a) Performance of the proposed LRIO framework compared with LIO and centralized fusion in Case-1. (b) Performance of the proposed LRIO framework under dense-smoke (Case-1) and light-smoke (Case-2) conditions.

Robot Localization

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

See all 176 papers

When LiDAR Fades, Radar Takes Over: LiDAR Reliability-Aware Adaptive Localization Under Airborne Obscurants
Oct 9, 2026Moumita Mukherjee, Avijit Banerjee, Anton Koval +1Robot Localization LiDAR Odometry
Robotics and AI Group, Department of Computer, Electrical and Space Engineering, Luleå University of Technology, Luleå, Sweden
How IMU Drift Influences Multi-Radar Inertial Odometry for Ground Robots in Subterranean Terrains
Feb 27, 2026Moumita Mukherjee, Magnus Norén, Anton Koval +2Radar-Inertial Odometry Robotics
FIRE-LIVWO: Robust LiDAR-Inertial-Visual-Wheel Odometry via Failure-Immune mmWave Radar Enhancement
Sep 4, 2026Kun Hu, Menggang Li, Kaidi Wu +5Simultaneous Localization and Mapping Robot State Estimation
School of Mechatronic Engineering, China University of Mining and Technology · Jiangsu Collaborative Innovation Center of Intelligent Mining Equipment, China University of Mining and Technology · National Key Laboratory of Intelligent Mining Equipment Technology, China University of Mining and Technology

Oct 9, 2026J/K move · Enter open · S save

When LiDAR Fades, Radar Takes Over: LiDAR Reliability-Aware Adaptive Localization Under Airborne Obscurants

Moumita Mukherjee, Avijit Banerjee, Anton Koval, George Nikolakopoulo

Robotics and AI Group, Department of Computer, Electrical and Space Engineering, Luleå University of Technology, Luleå, Sweden

Robot Localization LiDAR Odometry Fault Detection Unsupervised Anomaly Detection

Open paper AskPDF arXiv

LiDAR-inertial odometry (LIO) is widely used for real-time localization and mapping, but its reliability can degrade substantially in subterranean environments affected by smoke, dust, and other airborne obscurants, whereas radar remains comparatively robust. This article presents a one-class temporal learning framework for LiDAR degradation detection and reliability-aware localization under such conditions. A broad set of degradation-sensitive cues, including spatial, cross-modal, geometric/motion, intensity, and measurement-recency features, is evaluated to characterize changes in LiDAR sensing quality. Their temporal evolution is modelled over sliding windows using an LSTM autoencoder trained exclusively on nominal operation, enabling deviations from healthy sensor behaviour to be detected without requiring labelled degradation data. The resulting LiDAR reliability estimate governs adaptive selection between LIO and radar inertial odometry (RIO), retaining LIO during reliable sensing and transferring localization authority to RIO when degradation is detected. The framework is evaluated using real-world ground robot data collected in GNSS-denied subterranean tunnels under nominal and dense-smoke conditions. Experimental results demonstrate degradation detection, adaptive modality switching, and continued localization output under severe LiDAR degradation. A feature family ablation study further shows that a reduced feature representation can improve degradation-aware localization performance in the evaluated dense-smoke scenarios, highlighting the importance of feature parsimony in one-class sensor degradation detection. The code will be released as open source to support reproducibility and further research.

Abstract

Robot Localization

Explore similar work

When LiDAR Fades, Radar Takes Over: LiDAR Reliability-Aware Adaptive Localization Under Airborne Obscurants

How IMU Drift Influences Multi-Radar Inertial Odometry for Ground Robots in Subterranean Terrains

FIRE-LIVWO: Robust LiDAR-Inertial-Visual-Wheel Odometry via Failure-Immune mmWave Radar Enhancement