Super Odometry: IMU-centric LiDAR-Visual-Inertial Estimator for Challenging Environments

Shibo Zhao, Hengrui Zhang, Peng Wang, Lucas Nogueira, Sebastian A. Scherer

Super Odometry is a robust sensor-fusion framework designed to maintain precise robot localization in the world’s most "perceptually-degraded" environments—including smoke-filled caves, pitch-black tunnels, and featureless corridors. Developed by researchers at Carnegie Mellon University and Northeastern University, this system served as a critical component for Team Explorer’s success during the DARPA Subterranean (SubT) Challenge.

The Challenge of Perceptual Degradation

Autonomous robots typically rely on LiDAR and cameras for state estimation. However, these sensors possess significant vulnerabilities in extreme conditions:

  • LiDAR often fails in "geometrically degenerate" environments, such as long, smooth tunnels, or when the signal is obscured by heavy dust, fog, or smoke.
  • Cameras struggle in low-light conditions or featureless areas, such as white-walled rooms or dark corridors.

Existing SLAM (Simultaneous Localization and Mapping) solutions generally fall into two categories: loosely coupled systems, which are flexible but less accurate, and tightly coupled systems, which offer high precision but are often computationally heavy and brittle under sensor failure. Super Odometry bridges this gap by creating a resilient, high-precision system capable of high-speed navigation in the most difficult terrains.

Challenging environments used for testing, including long corridors, dark rooms, and dust-filled caves, alongside the multi-sensor drone setup.

System Architecture: An IMU-Centric Approach

The core philosophy of Super Odometry is IMU-centricity. While most SLAM systems treat the Inertial Measurement Unit (IMU) as a secondary sensor for motion compensation, Super Odometry makes it the architectural backbone. This is based on the insight that the IMU is the only sensor truly independent of the external environment; it functions regardless of lighting or visibility.

The Super Odometry factor graph architecture, showing the IMU-centric pipeline with parallel Visual-Inertial and LiDAR-Inertial sub-modules.

The Pipeline

  1. Factor Graph Architecture: The system is divided into three asynchronous sub-modules: IMU Odometry, Visual-Inertial Odometry (VIO), and LiDAR-Inertial Odometry (LIO). The IMU provides a high-frequency (1000Hz) motion prediction, while VIO and LIO modules provide "pose priors" to constrain the IMU’s inherent bias drift.
  2. Coarse-to-Fine Recovery: Motion is recovered in stages. The IMU provides the initial guess, which is then refined by the visual and LiDAR modules in parallel. This asynchronous processing ensures that a slow LiDAR update does not bottleneck the high-frequency state output.
  3. Dynamic Octree: To handle 3D point cloud data efficiently, the authors replaced traditional static KD-trees with a "dynamic octree." Utilizing a hash table for voxel organization, this structure allows the system to update only specific parts of the map, resulting in a 10x speedup in data association.
  4. PCA-based Feature Extraction: The system employs Principal Component Analysis (PCA) to evaluate geometric feature quality. Noisy returns—such as those reflecting off dust particles—are assigned low weights or rejected entirely to prevent map corruption.

Experimental Results

The system was rigorously tested using datasets from Team Explorer’s drones and ground robots in real-world DARPA SubT scenarios, including abandoned nuclear facilities and dust-filled mines. Super Odometry significantly outperformed state-of-the-art methods like LOAM, LIO-SAM, and VINS-Mono:

  • Precision: In a "Long Corridor" test, Super Odometry achieved a Translation RMSE of 0.055m, whereas LIO-SAM drifted to 0.85m and LOAM reached 4.15m.
  • Robustness: In heavy dust environments where LiDAR-only methods typically fail due to noise, Super Odometry remained stable and free of misalignment.
  • Efficiency: The dynamic octree consumed only 10% of the processing time required by standard methods, enabling real-time operation on lightweight onboard computers like an Intel NUC.

Conclusion

Super Odometry demonstrates that centering sensor fusion on the IMU provides a level of stability and resilience essential for robots operating in unpredictable, hazardous environments. Future work aims to integrate additional modalities, such as thermal cameras and wheel odometry, to further enhance the system's reliability in extreme multi-sensor failure scenarios.

Summary updated

Abstract

We propose Super Odometry, a high-precision multi-modal sensor fusion framework, providing a simple but effective way to fuse multiple sensors such as LiDAR, camera, and IMU sensors and achieve robust state estimation in perceptually-degraded environments. Different from traditional sensor-fusion methods, Super Odometry employs an IMU-centric data processing pipeline, which combines the advantages of loosely coupled methods with tightly coupled methods and recovers motion in a coarse-to-fine manner. The proposed framework is composed of three parts: IMU odometry, Visual-inertial odometry, and LiDAR-inertial odometry. The Visual-inertial odometry and LiDAR-inertial odometry provide the pose prior to constrain the IMU bias and receive the motion prediction from IMU odometry. To ensure high performance in real-time, we apply a dynamic octree that only consumes 10% of the running time compared with a static KD-tree. The proposed system was deployed on drones and ground robots, as part of Team Explorer’s effort to the DARPA Subterranean Challenge where the team won 1 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">st</sup> and 2 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">nd</sup> place in the Tunnel and Urban Circuits <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> , respectively.
BibTeX
@inproceedings{zhao2021super,
  author = {Shibo Zhao and Hengrui Zhang and Peng Wang and Lucas Nogueira and Sebastian A. Scherer},
  title = {{Super Odometry: IMU-centric LiDAR-Visual-Inertial Estimator for Challenging Environments}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2021},
  url = {https://openreview.net/forum?id=SVT0koeyv2_},
}
Figure