Waymo is releasing more autonomous driving data to facilitate industry research

Waymo is making more data from its self-driving vehicles public, which it says is for the benefit of the research community. Building on a database it released in 2019, the Alphabet company calls this latest batch of data the largest interactive dataset ever released for studying behavior prediction and motion prediction in autonomous driving.

The motion dataset includes more than 100,000 segments of objects such as cars and people and their trajectories captured by Waymo's self-driving vehicle sensors, each about 20 seconds long. The company includes corresponding 3D maps and geographic details in each segment to provide context for researchers' predictive modeling. Waymo said it has released a total of 570 hours of unique data.

The data includes a variety of road types in different urban environments, including San Francisco, Phoenix, Mountain View, Los Angeles, Detroit and Seattle. Waymo says it contains unique data mined specifically for research purposes, including situations where cyclists and vehicles share the road, cars speed through busy intersections, or groups of pedestrians gather on sidewalks.

Waymo also claims that its “perception boxes,” the colored boxes it uses to distinguish between different objects on the road, are state-of-the-art, outperforming other open source datasets. This is important because the better a perception system gets at tracking objects and people in motion data, the more accurate the resulting behavior prediction models will be at predicting what they’re going to do.

To go along with the data dump, Waymo is also launching an open dataset challenge to encourage research teams to work on behavior and prediction. There are four challenges: Motion Prediction, Interaction Prediction, Real-Time 3D Detection, and Real-Time 2D Detection. The winner of each challenge will receive $15,000, the second-place team will receive $5,000, and the third-place team will receive $2,000.

Waymo isn’t the first company to release an open dataset. In March 2019, Aptiv became one of the first large-scale autonomous driving operators to publicly release a set of sensor data. Cruise, a majority-owned subsidiary of General Motors, has also released an AV visualization tool to the public. Ford-backed Argo has also released a set of high-definition map datasets called “Argoverse.”