comma.ai releases 7 hours of self-driving car data, calls for Tesla, Google and others to do the same

comma.ai CEO George Hotz recently praised Tesla, Google and Otto for being fairly opened about their self-driving car programs, but he is taking his own company a step further in openness with the release of a dataset of 7.25 hours of comma.ai’s prototype at work.

We’ve often discussed at Electrek how data will be extremely important in the race to create a fully self-driving car, and also in the race to get such a system approved by regulators, which is why comma.ai’s move here is particularly interesting.

Back in May, we talked about Tesla adding an impressive ~1 million miles of data every 10 hours due to its important fleet of about 100,000 cars equipped with Autopilot sensors. On the other hand, Google has just over 1 million miles of data since launching its program in 2009 due to its smaller fleet, but it’s arguably collecting more data per mile due to using more sensors than Tesla.

As for openness, Google releases a monthly report on its program with a few data points and Tesla offered to share its Autopilot data with the US Department of Transport. And of course, CEO Elon Musk often talks openly about the program in details. Other startups developing self-driving technologies are much more secretive about their efforts, like Cruise, recently acquired by GM, and Zoox, a company currently raising a lot of money from VC firms to reportedly develop a self-driving car.

Hotz called for all of them, the secretive and less secretive, to release datasets from their systems. He wrote in a Reddit post today:

“Today, we are releasing some stuff for you. Open stuff. Because we like openness. It’s better than closedness. Even been stuck in an elevator? And openness makes the Machine Learning community great.

7.25 hours of driving data. From this, you should be able to replicate our initial Bloomberg experiments. Why you say? Because we know we will win. And if you do amazing things with this data, you can join the winning team and do more amazing things with even more data. We love data.

[…]

PS: Zoox, Tesla, Google, and Cruise, we are waiting for your Open releases”

The Bloomberg experiments are comma.ai’s recent tests with Bloomberg’s Ashley Vance in Nevada:

In a paper (embedded below) released with the dataset, Hotz and Eder Santana, machine learning intern at comma.ai, wrote about the data you can find in the release:

“We mounted a Point Grey camera in the windshield of an Acura ILX 2016, capturing pictures of the road at 20 Hz. In the released dataset there is a total of 7.25 hours of driving data divided in 11 videos. The released video frames are 160 × 320 pixels regions from the middle of the captured screen. Beyond video, the dataset also has several sensors that were measured in different frequencies and interpolated to 100Hz. Example data coming from sensors are the car speed, steering angle, GPS, gyroscope, IMU, etc.”

Here’s the paper:

[scribd id=320095885 key=key-YTkc23CK9apUBzoNp1Qg mode=scroll]

If you like comma.ai, maybe you should consider using their chffr app, which turns your cellphone into a cloud-based dashcam and it feeds data to comma.ai in order to help them develop their system.