RadarLoc: Learning to Relocalize in FMCW Radar
This addresses the relocalization challenge in robotics and computer vision by applying deep learning to radar data, which is novel but incremental as it adapts existing deep learning paradigms to a new sensor modality.
The paper tackles the problem of relocalization using FMCW radar scans, proposing RadarLoc, an end-to-end neural network with self-attention that estimates 6-DoF global poses and outperforms existing radar-based and deep camera methods by a significant margin on the Oxford Radar RobotCar dataset.
Relocalization is a fundamental task in the field of robotics and computer vision. There is considerable work in the field of deep camera relocalization, which directly estimates poses from raw images. However, learning-based methods have not yet been applied to the radar sensory data. In this work, we investigate how to exploit deep learning to predict global poses from Emerging Frequency-Modulated Continuous Wave (FMCW) radar scans. Specifically, we propose a novel end-to-end neural network with self-attention, termed RadarLoc, which is able to estimate 6-DoF global poses directly. We also propose to improve the localization performance by utilizing geometric constraints between radar scans. We validate our approach on the recently released challenging outdoor dataset Oxford Radar RobotCar. Comprehensive experiments demonstrate that the proposed method outperforms radar-based localization and deep camera relocalization methods by a significant margin.