News and insights on the impact of data on smart cities, businesses and technologies. Curated by RList

2019 Autonomous Driving Open Datasets Released To Date

To date, at least 9 large-scale datasets have been released openly to stimulate and accelerate the pace of research into self-driving cars.
Waymo self-driving car
Photo of Waymo self-driving car by Grendelkhan

Sources of Dataset

Today, we can already see several autonomous vehicle programs under trial, plying the roads of a number of cites in a controlled environment and collecting critical data for driving in a real-world environment. There are at least 9 well-known large-scale public datasets that researchers can work on for autonomous driving research, from major autonomous driving firms and A.I. research labs.
table listing all self-driving/autonomous driving open datasets
List of publicly released autonomous driving open datasets ( Interactive table link )

Waymo Open Dataset, August 2019

Example video of Waymo data collection
Snapshot of Waymo open dataset
Snapshot of the Waymo Open Dataset (Source: Waymo)
According to Waymo, the Waymo Open Dataset contains data from 1,000 driving segments. Each segment captures 20 seconds of continuous driving, corresponding to 200,000 frames at 10 Hz per sensor. Each segment contains sensor data from five high-resolution Waymo lidars and five front-and-side-facing cameras, and includes lidar frames and images with vehicles, pedestrians, cyclists, and signage carefully labeled, capturing a total of 12 million 3D labels and 1.2 million 2D labels. The Waymo team believes the dataset is "one of the largest, richest, and most diverse self-driving datasets ever released for research".

Location/Environment: The dataset covers diverse driving environments, including dense urban and suburban environments across Phoenix, AZ, Kirkland, WA, Mountain View, CA and San Francisco, CA with a wide spectrum of driving conditions (day and night, dawn and dusk, sun and rain).

Lyft Level 5 Dataset, July 2019

sensor configuration on Lyft's ford fusion vehicles for data collection
Sensor configuration on Lyft's Ford Fusion vehicles for data collection (Source: Lyft)
According to Lyft, Lyft used a fleet of Ford Fusion vehicles for data collection using two different configurations of LiDARs and cameras. the Lyft Level 5 Dataset includes over 55,000 human-labeled 3D annotated frames, data from 7 cameras and up to 3 LiDARs, a drivable surface map and underlying HD spatial semantic map of the surveyed region (including lanes and crosswalks).

Location/Environment: San Francisco

Argoverse, June 2019

sensor configuration on Argo vehicles for data collection
Sensor configuration on Argo's vehicle for data collection (Source: Argo)
According to Argo, the Argoverse Dataset is acquired from cars equipped with 2 roof-mounted LiDAR sensors, 7 HD ring cameras and 2 front-view facing stereo cameras, and includes 
  • One dataset with 3D tracking annotations for 113 scenes 
  • One dataset with 327,793 interesting vehicle trajectories extracted from over 1000 driving hours 
  • Two high-definition (HD) maps with lane centerlines, traffic direction, ground height, and more 
  • One API to connect the map data with sensor information

Location/Environment: 204 linear kilometers in Miami and 86 linear kilometers in Pittsburgh - two US cities with distinct urban driving challenges and local driving habits, across different seasons, weather conditions, and times of day to provide a broad range of real-world driving scenarios.

Aptiv nuScenes, March 2019

According to Aptiv, the nuScenes Dataset is acquired from the entire sensor suite of an autonomous vehicle (6 cameras, 1 LIDAR, 5 RADAR, GPS, IMU). The entire dataset contains 1,000 scenes of 20 second lengths, which includes approximately 1.4M camera images, 390k LIDAR sweeps, 1.4M RADAR sweeps and 1.4M object bounding boxes in 40k keyframes.
Example video of the nuScene dataset
sample dataset image from nuScenes
Sample image from nuScenes (Source:

Location/Environment: Boston and Singapore, two cities that are known for dense traffic and highly challenging driving situations.

Berkeley BDD100K, June 2018

Datasets released in 2019 by Aptiv, Argo, Lyft and Waymo have started to incorporate multi-modal data from other sensors such as LiDAR, radar and stereo cameras. BDD100K was released in June 2018 and while it lacks the multi-modal data of its newer counterparts, it is the largest dataset based on monocular videos with 120 million image frames across multiple cities, weather conditions, times of day and scene types.
comparison between BDD100K and other autonomous driving datasets
Comparison between BDD100K and other autonomous driving open datasets released before it (source:

Baidu ApolloScape, March 2018

The ApolloScape is part of the Baidu Apollo Program. Its dataset contains RGB videos with high-resolution image sequences (146,997 frames) and per-pixel annotation, along with survey-grade dense 3D points with semantic segmentation. The data is collected in different cities under various traffic conditions using mid-sized SUVs equipped with high resolution cameras and a Riegl acquisition system.
Sample data from ApolloScape with color labels and depth images

Hesai & Scale PandaSet, coming soon

Combining Hesai’s best in class LiDAR sensors with Scale’s high-quality data annotation, the full PandaSet dataset will feature:
  • 60,000 camera images
  • 20,000 LiDAR sweeps
  • 125 scenes of 8s each
  • 28 annotation classes
  • 37 semantic segmentation labels
  • Full sensor suite: 1x mechanical LiDAR, 1x solid-state LiDAR, 6x cameras, On-board GPS/IMU
Sensor configuration on Hesai and Scale's vehicle for data collection
Sensor configuration on Hesai and Scale'svehicle for data collection (Source:

Location/Environment: Pandaset scenes are selected from 2 routes in Silicon Valley: (1) San Francisco; and (2) El Camino Real from Palo Alto to San Mateo, showcasing complex urban driving scenarios, including steep hills, construction, dense traffic and pedestrians, and a variety of times of day and lighting conditions in the morning, afternoon, dusk and evening.

Where are we today, in terms of autonomous driving?

SAE J3016 Six Levels of Driving Automation
SAE J3016 Six Levels of Driving Automation ( click to enlarge )

There are six levels of driving automation in SAE (Society of Automobile Engineers) International's J3016TM driving automation standard from Level 0 (No driving automation) to Level 5 (Full driving automation). Most vehicles on our roads today are at Level 0, that is, manually driven. Tesla Autopilot and Cadillac Super Cruise qualify as Level 2. The 2019 Audi A8L with Traffic Jam Pilot will be classified at Level 3 when rolled out. Level 4 vehicles are in geo-fenced test-bedding stages at the moment and that is the frontline research of autonomous driving in the world today.

A future of Level 5 fully autonomous driving will greatly reduce fatalities on the road, solve traffic issues such as congestion and parking, and improve the environment by reducing personal cars on the road and maximizing shared transport.

Meanwhile, you may also be interested in U.S. states that have legalized PDDs (personal delivery devices or autonomous robots that perform last-mile deliveries within a community) in the past three years, namely, Virginia, Idaho, Wisconsin, Florida, Ohio, Utah, Arizona, Washington, Texas, North Carolina, Missouri, Colorado and Tennessee.