1 year, 1000 km: The Oxford RobotCar dataset

Maddern, Will; Pascoe, Geoffrey; Linegar, Chris; Newman, Paul · 2016 · OpenAlex-citations

DOI: 10.1177/0278364916679498

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper introduces the Oxford RobotCar Dataset, a large-scale resource designed to address the challenges of long-term autonomous driving, specifically focusing on localization and mapping in dynamic urban environments. While existing datasets like KITTI and Cityscapes support algorithm development for tasks such as motion estimation and object detection, they often fail to capture the significant environmental variations encountered over extended periods. The authors argue that long-term autonomy requires handling changes in scene appearance and structure due to weather, illumination, seasonal effects, and construction. To facilitate research in this area, the authors present a dataset collected over one year, capturing the same route under diverse conditions. Data was collected between May 2014 and December 2015 using the Oxford RobotCar platform, an autonomous-capable Nissan LEAF driven manually along a 10km route in central Oxford, UK. The vehicle was traversed approximately twice a week, resulting in over 100 traversals and 1,000km of recorded driving. The platform was equipped with a comprehensive sensor suite, including a trinocular stereo camera (Point Grey Bumblebee XB3), three monocular fisheye cameras (Point Grey Grasshopper2), two 2D LIDAR sensors (SICK LMS-151), one 3D LIDAR sensor (SICK LD-MRS), and a GPS/INS navigation system (NovAtel SPAN-CPT). The collection captured nearly 20 million images and over 23TB of uncompressed data, covering conditions ranging from heavy rain and snow to night driving and direct sunlight. The dataset provides raw sensor data alongside full intrinsic and extrinsic calibrations, ensuring sub-millisecond synchronization. The authors emphasize logging raw data only, such as unrectified Bayer images and raw LIDAR packets, to maximize utility for future research methods. They provide MATLAB development tools for accessing the data, including functions for image demosaicing, undistortion, and 3D pointcloud generation. The dataset includes labels for environmental conditions and quality metrics for GPS reception, allowing researchers to filter data based on specific challenges like poor satellite signals or structural changes caused by roadworks. The significance of this work lies in its ability to challenge current approaches to long-term localization and mapping. By providing a dataset where the same location is observed under drastically different conditions and structural changes, it enables the investigation of lifelong learning and robust perception algorithms. The authors aim to accelerate research toward reliable autonomy in real-world cities and plan to offer a benchmarking service similar to the KITTI suite to allow public comparison of long-term localization methods.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• driverless ads

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: dataset resource, tool software