Ford Shares Sensor Test Data for Autonomous Vehicles
To help spur further development of autonomous vehicles, Ford is releasing to outside researchers a year’s worth of information gathered from its test vehicles.
Captured by advanced sensors and mapping systems during testing in Michigan in 2017 and 2018, the records identify and categorize myriad objects and driving scenarios that vehicles could encounter. Such information is used to teach autonomous vehicles how to navigate real-world conditions.
Accounting for Variables
The test vehicles were manually driven at different days and times on Detroit-area routes that included a mix of highways, city centers, university campuses, suburban neighborhoods and other venues. Because the information was collected over the course of a year, it includes seasonal weather variations as well as changes in traffic, pedestrian density and construction conditions.
Data were culled from lidar, cameras, a 3D-point cloud and ground reflectivity maps. A plug-in app adds a visualization element for the data, which is offered in ROS format.
The integrated information provides a comprehensive view of what a self-driving vehicle “sees,” Ford notes. Researchers need such records to design robust algorithms that enable vehicles to recognize and safely adapt to as many variables as possible.
The data can be used to construct new testing simulations. The information also provides a baseline that can be used to compare other datasets and real-world tests, Ford points out.
Tony Lockwood, Ford’s autonomous vehicle manager for virtual driver systems, details the data-sharing initiative in a blog post on medium.com.
Ford is releasing the information online through its collaboration with the Amazon open data program. The first data set already has been uploaded.
Several other companies previously have announced plans to share their sensor datasets:
- Waymo’s initiative comprises 1,000 high-resolution driving scenes that label such objects as cyclists, pedestrians and traffic signs.
- Lyft is making 55,000 3D frames of video footage available to autonomy researchers.
- Aptiv has created an open-source dataset.
But Ford claims its program is the first to include data from multiple vehicles, which the carmaker says provides a more complete snapshot.
The 2017-2018 tests were separate from Ford’s partnership with its Argo AI affiliate to develop commercial applications for self-driving vehicles. Last week, Ford announced it was delaying those efforts, which aimed to launch robo-taxis and autonomous delivery services in 2021, until 2022 in part because of the impact of the coronavirus pandemic.
Ford has made an accomplishment that will never be bested, never even be tied.
The little car that could still can. And this time as a car that not only gets great fuel economy, but which has ride and handling that makes it more than an econo-box (and its styling is anything but boxy).
Chinese electric-car startup Nio Inc. is forming a manufacturing joint venture with Beijing E-Town International Investment and Development Co., which is investing 10 billion yuan ($1.5 billion) in the business.