Lidar-level localization with radar? The CFEAR approach to accurate, fast and robust large-scale radar odometry in diverse environments (2211.02445v3)

Abstract: This paper presents an accurate, highly efficient, and learning-free method for large-scale odometry estimation using spinning radar, empirically found to generalize well across very diverse environments -- outdoors, from urban to woodland, and indoors in warehouses and mines - without changing parameters. Our method integrates motion compensation within a sweep with one-to-many scan registration that minimizes distances between nearby oriented surface points and mitigates outliers with a robust loss function. Extending our previous approach CFEAR, we present an in-depth investigation on a wider range of data sets, quantifying the importance of filtering, resolution, registration cost and loss functions, keyframe history, and motion compensation. We present a new solving strategy and configuration that overcomes previous issues with sparsity and bias, and improves our state-of-the-art by 38%, thus, surprisingly, outperforming radar SLAM and approaching lidar SLAM. The most accurate configuration achieves 1.09% error at 5Hz on the Oxford benchmark, and the fastest achieves 1.79% error at 160Hz.

• The paper presents a learning-free CFEAR algorithm that achieves 1.09% translation error at 5 Hz for large-scale radar odometry.
• It integrates motion compensation with one-to-many scan registration to mitigate radar sparsity, bias, and outliers.
• The method operates in real-time up to 160 Hz, demonstrating robust performance across diverse indoor and outdoor environments.

Lidar-level Localization with Radar: The CFEAR Approach

This paper presents a novel, efficient, and learning-free methodology for large-scale odometry estimation using spinning radar technology, applicable across varied environments, and introduces enhancements to the previously established CFEAR algorithm. Specifically, it tackles the odometry and SLAM challenges in environments ranging from open outdoor areas to confined indoor locales using Frequency-Modulated Continuous Wave (FMCW) radar systems, emphasizing robustness and computational efficiency without the need for parameter tuning across different environments.

The key innovation of this method lies in integrating motion compensation with a one-to-many scan registration process. By minimizing distances between nearby oriented surface points with robust loss functions, the approach addresses radar-specific challenges like sparsity, bias, and outlier mitigation. These enhancements improve the precision of pose estimation and extend the technique's applicability to environments uncharacteristic for radar-based methods.

The paper provides a comprehensive analysis of various components influencing radar odometry accuracy and efficiency. The authors conducted an extensive ablation paper to evaluate the impact of elements such as radar filtering, data resolution, registration costs and loss functions, keyframe history, and surface point computation weighted by intensity. The results demonstrate a marked improvement, with a notably reduced translation error compared to prior state-of-the-art methods. Specifically, the most accurate configuration of their proposed approach achieves 1.09% translation error at 5 Hz on the Oxford Radar RobotCar dataset, which notably challenges traditional lidar-based SLAM solutions.

One of the standout aspects of this paper is the simultaneous optimization of radar odometry for both accuracy and processing speed. The authors propose four different configurations of the CFEAR approach, catering to varying requirements of computational efficiency and drift reduction. The proposed methodology runs in real-time, achieving a frame rate of up to 160 Hz with minimal translation error, which is a significant stride towards practical deployment in autonomous systems.

Furthermore, the paper explores the theoretical implications and practical applications of radar odometry in robotic perception. By demonstrating robustness across sensor configurations and environmental scales, the authors suggest the method's potential adaptability to future radar sensor advancements, including those with higher resolution and different mounting configurations.

In conclusion, this paper offers valuable insights into advancing radar-based odometry, presenting an adaptable, robust, and efficient method that stands out for its practicality and theoretical contributions to the field of robotic perception and navigation. The implications of this research are profound, pointing towards more resilient autonomous systems capable of navigating diverse and challenging environments. Future work in AI and robotics might build on these findings to further enhance real-world robotic applications, potentially integrating more advanced sensor technologies and exploring additional use cases beyond traditional large-scale robotic mapping and navigation.