ORORA: Outlier-Robust Radar Odometry (2303.01876v1)

Abstract: Radar sensors are emerging as solutions for perceiving surroundings and estimating ego-motion in extreme weather conditions. Unfortunately, radar measurements are noisy and suffer from mutual interference, which degrades the performance of feature extraction and matching, triggering imprecise matching pairs, which are referred to as outliers. To tackle the effect of outliers on radar odometry, a novel outlier-robust method called \textit{ORORA} is proposed, which is an abbreviation of \textit{Outlier-RObust RAdar odometry}. To this end, a novel decoupling-based method is proposed, which consists of graduated non-convexity~(GNC)-based rotation estimation and anisotropic component-wise translation estimation~(A-COTE). Furthermore, our method leverages the anisotropic characteristics of radar measurements, each of whose uncertainty along the azimuthal direction is somewhat larger than that along the radial direction. As verified in the public dataset, it was demonstrated that our proposed method yields robust ego-motion estimation performance compared with other state-of-the-art methods. Our code is available at https://github.com/url-kaist/outlier-robust-radar-odometry.

• The paper presents a decoupled GNC-based approach for robust radar odometry that accurately separates rotation and translation to handle outliers.
• It models anisotropic measurement uncertainty, significantly improving performance over traditional methods in feature-sparse, noisy conditions.
• Experimental results on the MulRan dataset demonstrate real-time performance at over 11 Hz, enhancing reliable autonomous navigation.

ORORA: Outlier-Robust Radar Odometry

The paper introduces ORORA, a novel methodology aimed at enhancing the robustness of radar odometry in the presence of outliers. As radar sensors gain prominence in autonomous systems operating under harsh weather conditions, the persistent challenge of noise and interference in radar measurements necessitates advanced solutions. ORORA addresses these issues through a decoupled approach to pose estimation.

Key Contributions and Methodology

The paper's contribution is threefold:

1. Decoupling-Based Approach: The technique utilizes a decoupled method consisting of graduated non-convexity (GNC)-based rotation estimation and anisotropic component-wise translation estimation (A-COTE). Decoupling rotation and translation allows for more accurate handling of gross outliers.
2. Anisotropic Uncertainty Modeling: By accounting for the anisotropic characteristics of radar measurements—where uncertainty along the azimuthal direction surpasses that in the radial direction—the method improves upon conventional isotropic noise assumptions.
3. Performance Demonstration: Extensive experiments demonstrate significant improvements over state-of-the-art methods in radar odometry, even with the presence of imprecise feature correspondences.

Experimental Setup and Results

The evaluation was conducted using the MulRan dataset, featuring various sequences with differing environmental characteristics. The experiments assessed both relative translation and rotation errors, utilizing two feature extraction methods, Cen2018 and Cen2019. Across diverse scenarios, ORORA consistently outperformed traditional RANSAC-based methods and its variants, such as MC-RANSAC with and without Doppler compensation.

The results emphasize ORORA's robustness, particularly in feature-sparse environments where outlier ratios are higher. The method's lessened sensitivity to the quality of feature extraction highlights its practical applicability across varying conditions, maintaining precision in both translation and rotation estimations.

Algorithmic Efficiency

ORORA achieves real-time performance, supporting operation at over 11 Hz, above the typical 4 Hz of radar image acquisition. This efficiency is crucial for deployment in real-time autonomous systems where timely and accurate sensor fusion is essential for operational safety.

Implications and Future Work

Practically, ORORA offers a robust framework for ego-motion estimation in autonomous vehicles, potentially integrating seamlessly into SLAM systems. The paper suggests future incorporation of loop detection and closure modules to expand the method into comprehensive SLAM frameworks.

Theoretically, the emphasis on anisotropic uncertainty modeling could inspire further investigations into sensor measurement characteristics and their implications on localization and mapping algorithms.

ORORA's introduction of a GNC-based decoupling strategy into radar odometry is a pivotal step in addressing the longstanding challenge of outlier handling, fostering improvements in reliable autonomous navigation.