Data Backup Digest

Modern GPS devices have undergone a great deal of change over the past few years. While the earliest models were bulky devices, today's GPS devices are much more compact. Some even rely on software-based GPS systems, either within their smartphone or the vehicle itself, to navigate from destination to destination.

However, even the best and most accurate GPS devices are only as good as the data that is stored within. The same principles apply to self-driving cars, too. Although they are capable of successfully navigating from point A to point B, they can only do so if they're using accurate maps.

The team with Argo AI released a statement reading, in part: ''"For our team at Argo, releasing this data collection is about giving academic communities access to the materials they need. We're excited to not only support cutting-edge developments in computer vision and machine learning but also to support the next generation of engineers and roboticists who are preparing for jobs at self-driving technology companies, Argo AI included."''

It's important to note that this data comes from a variety of sources. In fact, the majority of the maps come from the Ford Fusion vehicles that are used in the development of Argo AI. The maps available via Argo AI encompass approximately 127 linear miles of roadway in Miami, Florida, as well as 53 miles of roadway in Pittsburgh, Pennsylvania. Maps disseminate data related to lane geometries, drivable roadways, and terrain height, traffic flow, and much more.

Moreover, the team with Argo AI developed a customized Map API to streamline the process of accessing and utilizing this data. Coded in Python, the Argoverse Map API is a great tool for developers that includes a number of useful functions.

Datasets featured in Argo AI are also gleaned from several different sources. They primarily consist of 3D tracking models that are used to detect and track anything that enters into a roadway's drivable area. Argo AI currently includes approximately 11,000 different objects, but more are expected to be added in the future.

One of their most useful datasets, Argoverse Motion Forecasting, is used to train and validate motion forecasting models. This datasets contains nearly 330,000 different scenarios – each of which is five seconds long. This data was specifically curated for the project and compiled from more than 1,000 hours of data from their fleet of self-driving test vehicles.

Gathering the actual data, however, involved more than just self-driving test vehicles. Not only were these Ford Fusion Hybrids outfitted with Argo AI self-driving technology, but they were also equipped with two LiDAR sensors, two forward facing cameras, and seven cameras within a ring around the entire vehicle.

This combination of sensors and cameras ultimately store data in a comprehensive log, which includes detailed calibration data for both LiDAR sensors and all of the cameras involved. All of this data has been included in the datasets available via Argoverse.


For more information on Argoverse, or to download these maps and datasets for your own use, please visit their official site at www.argoverse.org.