CFEAR-Teach-and-Repeat: Fast and Accurate Radar-only Localization

2026-03-06

AI Summary

This work addresses the challenge of robust vehicle localization under adverse weather conditions, where optical sensors often fail, by proposing a teach-and-repeat framework that relies solely on a single rotating radar. The method achieves efficient and robust pose estimation by simultaneously aligning live radar scans to both a prior teaching map and a sliding window of recent keyframes. It leverages a Doppler-compensated, sparse representation of oriented surface points, integrated within a pose graph formulation and optimized via sliding-window bundle adjustment. To the best of our knowledge, this is the first purely radar-based approach to attain localization accuracy approaching that of lidar, with notably improved heading estimation and consistent performance across seasons and weather conditions. Evaluated on the Boreas dataset, it achieves average position and heading errors of 0.117 m and 0.096°, respectively—representing a 63% improvement over the state of the art—at a real-time rate of 29 Hz.

Technology Category

• Intelligent Robots: Localization, Mapping, and Navigation
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Computer Vision: Motion & Tracking

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Reliable localization in prior maps is essential for autonomous navigation, particularly under adverse weather, where optical sensors may fail. We present CFEAR-TR, a teach-and-repeat localization pipeline using a single spinning radar, which is designed for easily deployable, lightweight, and robust navigation in adverse conditions. Our method localizes by jointly aligning live scans to both stored scans from the teach mapping pass, and to a sliding window of recent live keyframes. This ensures accurate and robust pose estimation across different seasons and weather phenomena. Radar scans are represented using a sparse set of oriented surface points, computed from Doppler-compensated measurements. The map is stored in a pose graph that is traversed during localization. Experiments on the held-out test sequences from the Boreas dataset show that CFEAR-TR can localize with an accuracy as low as 0.117 m and 0.096{\deg}, corresponding to improvements of up to 63% over the previous state of the art, while running efficiently at 29 Hz. These results substantially narrow the gap to lidar-level localization, particularly in heading estimation. We make the C++ implementation of our work available to the community.

Problem

• radar-only localization
• autonomous navigation
• adverse weather
• teach-and-repeat
• prior map

Innovation

• radar-only localization
• teach-and-repeat
• Doppler-compensated surface points
• pose graph
• adverse weather navigation