Series Six of the Integer Intelligence Autonomous Driving Dataset Collection – Overview and Highlights

Integer Intelligence has released an eight‑series collection of autonomous driving datasets; Series Six highlights four notable datasets: StreetLearn, UTBM RoboCar, Multi‑Vehicle Stereo Event Camera, and comma2k19.

The eight series cover target detection, semantic segmentation, lane detection, optical flow, stereo, localization & mapping, driving behavior, and simulation datasets, each linked to detailed resources.

StreetLearn (DeepMind, 2019) provides a first‑person, partially observable visual environment built from Google Street View images of Pittsburgh and New York, supporting end‑to‑end deep reinforcement learning research with high‑resolution photos, diverse city scenes, and navigation tasks. Download: http://streetlearn.cc .

UTBM RoboCar (Université de Technologie de Belfort‑Montbéliard, 2019) offers multi‑sensor data collected with 11 heterogeneous sensors (cameras, LiDAR, radar, IMU, GPS‑RTK) using ROS, featuring dynamic urban and suburban environments, long‑term mapping, and behavior prediction data. Download: https://epan-utbm.github.io/utbm_robocar_dataset/ .

Multi Vehicle Stereo Event Camera (MVSEC) (University of Pennsylvania, 2018) contains synchronized stereo event‑camera recordings with event streams, grayscale images, IMU data, and high‑frequency pose and depth information from lidar and motion‑capture systems, captured under varied lighting and motion conditions. Download: https://daniilidis-group.github.io/mvsec .

comma2k19 (comma.ai, 2018) records over 33 hours of highway driving on California 280, split into 2019 one‑minute segments, using comma EON devices with road‑facing camera, GPS, temperature sensor, and 9‑axis IMU, providing reproducible, scalable data for localization and mapping. Download: https://github.com/commaai/comma2k19 .

The Annotated Laser Dataset (Carnegie Mellon University, 2011) includes laser scans, odometry, and three‑camera images with precise pose, object segmentation, and motion‑target annotations for SLAM and MOT research. Download: http://any.csie.ntu.edu.tw/data/ .

Ford Dataset (University of Michigan, 2010) comprises multi‑sensor data from Ford’s autonomous fleet collected in 2017‑18 across diverse urban, suburban, and highway scenarios, featuring 3D LiDAR, multiple high‑resolution cameras, GPS‑RTK, IMU, and ROS‑bag recordings for robust autonomous driving algorithm development. Download: avdata.ford.com .

Oxford RobotCar (University of Oxford, 2015) provides long‑term recordings of the same route under varying weather, lighting, and seasonal conditions, totaling ~20 million images, multi‑sensor data (LiDAR, cameras), and full‑resolution reflectivity and 3D point clouds, packaged in tar archives for convenient download. Download: https://robotcar-dataset.robots.ox.ac.uk/downloads/ .