Open Sourcing 3½ Hours of Driving Data (With LIDAR!)

Collected on El Camino Real, in Mountain View, CA, by our Lincoln MKZ

Data available on GitHub

A self-driving car cannot be developed without a huge amount of testing data, and at Udacity we have been working hard to get as much as we can released in the open (read the backstory and learn about our sponsored Challenges here!). To date, we have open sourced over 4 hours and 200GB of driving data to assist the participants of Challenge #2 and Challenge #3. Now, as of last night, we are nearly doubling that number with another 3 1/2 hours. This data is free for anyone to use, anywhere in the world.

The path this data covers is going to start to become a familiar one; from the Udacity office in Mountain View to San Francisco along El Camino Real. All of the required steering and GPS data is present for Challenge #2 and Challenge #3, and the distance traversed covers where the test data for Challenge #3 will be pulled from, albeit on a different day.

To download the dataset, please head to our GitHub repo.

El Camino Real

This release is in the same format as the last one (so it’s a 30GB download instead of 200GB), with one cool extra: LIDAR! While LIDAR doesn’t yet have a purpose within the scope of our Challenges, we know that many people in the community can’t wait to get their hands on open source LIDAR data.

Note: Along with an image frame from our cameras, we also include latitude, longitude, gear, brake, throttle, steering angles and speed.

We attempted to stay in one lane the entire trip, but construction and broken-down vehicles didn’t quite let that happen. One portion of the trip actually saw us move onto the other side of the road with cones, leading to an interesting test case for your network for Challenge #2! If your CNN can navigate this properly, be sure send us screenshots and video!


We can’t wait to see what you do with the data! Please share examples with us in our self-driving car Slack community, participate in Challenge #2, or send a Tweet to @olivercameron. Enjoy!

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.