Waymo Open Dataset Challenges make a game out of driverless car sensors

Chris Davies - Mar 19, 2020, 8:21 am CDT
0
Waymo Open Dataset Challenges make a game out of driverless car sensors

Waymo has thrown open its latest self-driving car data set, inviting autonomous vehicle researchers to take part in a set of competitions based on the plentiful data gathered by its driverless fleet. The Waymo Open Dataset Challenges comes as the Google spin-off adds 800 new segments to its multi-sensor dataset, and it could come as a good distraction if you’re stuck at home because of coronavirus self-isolation.

The dataset itself is a bundle of high-resolution sensor data, which Waymo’s vehicles have collected as they roam the roads in various locations. It includes a variety of different geographies and conditions, as well as information from a whole host of sensors. There’s everything in there from cameras mounted on the front and sides of the vehicle, to the mid- and short-range LIDAR, and synchronized data between the two.

Waymo has also labeled some of that data, such as for vehicles, pedestrians, cyclists, and road signs. “While this dataset is not reflective of the full capabilities of our sensor system and is only a fraction of the data on which Waymo’s self-driving system is trained,” the Alphabet company says, “we believe that for research purposes this large, diverse and high-quality dataset is extremely valuable.”

There are five different challenges that researchers can take part in. The first is around 3D detection: when given one or more LIDAR range images and the associated camera images, you have to produce a set of 3D upright boxes for the different objects in the scene. A second, 2D challenge does much the same, but for 2D boxes from sets of camera images alone.

The third and fourth challenge tackle tracking. They each include a temporal sequence of data – either LIDAR and camera, or just camera images – from which participants need to produce a set of 3D or 2D upright boxes and the correspondence between those boxes across frames. Finally, challenge five is around domain adaptation, adding additional segments from a new location – only a subset of which have labels – to the 3D detection challenge.

A team can have up to ten members, and each team can submit for evaluation up to three times within 30 days. Submissions are open from today, through to May 31, 2020. Waymo will be using automated validation to rank participants on a leaderboard – with manual checking of the top entries – and there’ll be three prize tiers for each of the five challenges. The best gets $15,000; second place gets $5,000; and third place gets $2,000. While Alphabet employees might show up on the leaderboard, they won’t be competing for the cash awards.

The data crunching continues, even as Waymo One is grounded

Waymo’s new challenge arrives only two days after at least part of the Waymo One fleet was grounded in Metro Phoenix. The trial ride-hailing program was offering autonomous rides to participants in a closed beta, but that was paused over COVID-19 social distancing and hygiene concerns.

“In the interest of the health and safety of our riders and the entire Waymo community, we’re pausing our Waymo One service with trained drivers in Metro Phoenix for now as we continue to watch COVID-19 developments,” Waymo said in a statement. ” We’ve also paused driving in California in line with local guidance. Our fully driverless operations in Phoenix will continue for now within our early rider program, along with our local delivery and truck testing.”


Must Read Bits & Bytes