Abstract
Reliable spatial perception is fundamental for autonomous systems operating in complex real-world environments. However, conventional sensing modalities such as cameras and LiDAR often degrade when exposed to adverse weather, low illumination, or texture-poor scenes. The emergence of 4D millimeter-wave radar provides complementary advantages including weather and lighting robustness and direct Doppler velocity measurements, yet its sparse and noisy point clouds make accurate state estimation challenging. To address these limitations, we propose 4D RaVI-SLAM, a robust 4D radar visual inertial SLAM framework tailored for reliable operation in degraded conditions. We design a Hybrid Error-State Kalman Filter (HESKF) that tightly integrates IMU preintegration with radar Doppler constraints to achieve stable ego-velocity estimation. Furthermore, we introduce a probabilistic Color Normal Distribution Transform(Prob-ColorNDT) registration method that enriches radar points with superpixel-based color features and associated uncertainties, enabling consistent alignment in geometrically weak or visually ambiguous environments. These components are jointly optimized within a pose-graph formulation with loop closures to ensure long-term consistency. Extensive experiments across diverse scenarios demonstrate that our system significantly improves accuracy, robustness, and consistency over existing radar-centric and visual-centric approaches. By combining the complementary strengths of radar, visual, and inertial sensing within a unified probabilistic framework, 4D RaVI-SLAM thereby provides a reliable perception backbone specifically for autonomous systems operating in perceptually degraded environments such as smoke, fog, low-light conditions, and texture-poor scenes where conventional cameras and LiDAR often fail.
| Original language | English |
|---|---|
| Journal | IEEE Transactions on Intelligent Transportation Systems |
| DOIs | |
| Publication status | Accepted/In press - 2026 |
| Externally published | Yes |
Keywords
- 4-D radar
- SLAM
- Sensor fusion
- autonomous driving
- robust perception
Fingerprint
Dive into the research topics of '4-D RaVI-SLAM: 4-D Radar Visual Inertial Simultaneous Localization and Mapping'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver