[Paper Review] The Oxford Radar RobotCar Dataset: A Radar Extension to the Oxford RobotCar Dataset

Dan Barnes, Matthew Gadd+3 more|arXiv (Cornell University)|Sep 3, 2019

Robotics and Sensor-Based Localization18 references18 citations

TL;DR

This paper introduces the Oxford Radar RobotCar Dataset, a large-scale, multi-sensor dataset collected over 32 traversals of a central Oxford route in January 2019, featuring 4.7 TB of millimetre-wave FMCW radar data, 2.4 million 3D LIDAR scans, and complementary camera, GPS/INS, and 2D LIDAR data. The dataset enables robust research in radar-based scene understanding for autonomous vehicles under adverse weather and lighting conditions, with released ground truth radar odometry to accelerate development in radar-centric perception and localization.

ABSTRACT

In this paper we present The Oxford Radar RobotCar Dataset, a new dataset for researching scene understanding using Millimetre-Wave FMCW scanning radar data. The target application is autonomous vehicles where this modality is robust to environmental conditions such as fog, rain, snow, or lens flare, which typically challenge other sensor modalities such as vision and LIDAR. The data were gathered in January 2019 over thirty-two traversals of a central Oxford route spanning a total of 280km of urban driving. It encompasses a variety of weather, traffic, and lighting conditions. This 4.7TB dataset consists of over 240,000 scans from a Navtech CTS350-X radar and 2.4 million scans from two Velodyne HDL-32E 3D LIDARs; along with six cameras, two 2D LIDARs, and a GPS/INS receiver. In addition we release ground truth optimised radar odometry to provide an additional impetus to research in this domain. The full dataset is available for download at: ori.ox.ac.uk/datasets/radar-robotcar-dataset

Motivation & Objective

• To address the limited availability of large-scale, long-term radar datasets for autonomous vehicle research, particularly under adverse environmental conditions.
• To extend the existing Oxford RobotCar Dataset with millimetre-wave FMCW radar data to support robust, all-weather perception and localization.
• To provide ground truth radar odometry to facilitate research in radar-based localization and scene understanding.
• To enable cross-modal research by integrating radar with existing vision, LIDAR, and inertial sensor data in a unified, calibrated dataset.
• To accelerate innovation in radar-centric autonomous systems by releasing a comprehensive, publicly accessible dataset with full tooling support.

Proposed method

• The dataset was collected using the Oxford RobotCar platform, an autonomous Nissan LEAF equipped with a Navtech CTS350-X FMCW radar operating at 4 Hz with 163m range and 4.38cm range resolution.
• Over 32 traversals of a 10km central Oxford route were conducted in January 2019, capturing data across diverse weather, lighting, and traffic conditions.
• The dataset includes 2.4 million 3D LIDAR scans from two Velodyne HDL-32E sensors, six cameras, two 2D LIDARs, and GPS/INS, all time-synchronized and calibrated.
• Raw radar data are provided in polar format (azimuth, range, power), with tools to convert to Cartesian point clouds using configurable resolution and size.
• MATLAB and Python tools were developed to load, visualize, and convert radar and LIDAR data, including batch downloaders and deep learning data loaders.
• Ground truth radar odometry was computed and released to support research in radar-only localization and mapping.

Experimental results

Research questions

• RQ1 Can millimetre-wave FMCW radar provide reliable, long-range perception in adverse weather conditions such as fog, rain, and direct sunlight?
• RQ2 How does radar-based scene understanding compare to vision and LIDAR in complex urban environments with dynamic traffic and lighting variations?
• RQ3 To what extent can radar data be used for accurate localization and mapping in the absence of visual features or under sensor degradation?
• RQ4 What are the performance characteristics of radar odometry in real-world urban driving scenarios with long-term environmental variability?
• RQ5 How can radar data be effectively fused with other sensor modalities (e.g., LIDAR, cameras) for robust autonomous vehicle perception?

Key findings

• The dataset comprises 280 km of urban driving collected over 32 traversals in January 2019, capturing a wide range of weather, lighting, and traffic conditions.
• The radar data cover a 360° field of view with a maximum range of 163 m and 4.38 cm range resolution, enabling detection of distant and low-contrast objects.
• The dataset contains 4.7 TB of raw radar scans and 2.4 million 3D LIDAR scans, with full sensor calibration and time synchronization.
• Ground truth radar odometry was computed and released to support research in radar-based localization and mapping.
• The dataset is publicly available at ori.ox.ac.uk/datasets/radar-robotcar-dataset with full MATLAB and Python tooling for data loading, visualization, and conversion.
• The integration of radar with existing vision and LIDAR data enables new research directions in multi-modal, all-weather autonomous perception.