2021/05/28 by Keenan Burnett, David J. Yoon, Burnett, Keenan +5 · 6 citations
Computer Science · Engineering · Mathematics · #Anomaly Detection Techniques and Applications #Artificial intelligence #Autonomous Vehicle Technology and Safety #Computer science #Computer vision #Estimator #FOS: Computer and information sciences #Feature (linguistics) #Hidden Markov model #Mathematics #Mobile robot #Odometry #Probabilistic logic #Radar #Robot #Robotics (cs.RO) #Robotics and Sensor-Based Localization #Telecommunications #Trajectory #Visual odometry #Viterbi algorithm #cs.RO
paper · pdf · doi:10.48550/arxiv.2105.14152
published in arXiv (Cornell University) (Cornell University) · Accepted to Robotics Science and Systems 2021
openalex publication_date 2021/05/28 · arxiv created 2021/06/30 · arxiv updated 2021/07/02 · openalex created_date 2022/07/25 · openalex updated_date 2026/08/06
This paper presents a radar odometry method that combines probabilistic\ntrajectory estimation and deep learned features without needing groundtruth\npose information. The feature network is trained unsupervised, using only the\non-board radar data. With its theoretical foundation based on a data likelihood\nobjective, our method leverages a deep network for processing rich radar data,\nand a non-differentiable classic estimator for probabilistic inference. We\nprovide extensive experimental results on both the publicly available Oxford\nRadar RobotCar Dataset and an additional 100 km of driving collected in an\nurban setting. Our sliding-window implementation of radar odometry outperforms\nmost hand-crafted methods and approaches the current state of the art without\nrequiring a groundtruth trajectory for training. We also demonstrate the\neffectiveness of radar odometry under adverse weather conditions. Code for this\nproject can be found at: https://github.com/utiasASRL/heroradarodometry\n