Congratulations to the Winners of the re:Invent Robocar Rally 2017!
To drive awareness of deep learning, machine learning, and the internet of things in autonomous driving, AWS hosted a hackathon—the Robocar Rally—at re:Invent in November 2017. We kicked off Robocar Rally in September with a series of blog posts and Twitch streams. At re:Invent, we had 100 attendees come for a hands-on two-day hackathon using deep learning and the open source Donkey Car platform with AWS machine learning services and AWS IoT. They formed teams, and built, customized, trained, and raced their own 1/16th scale cars. There’s a lot we could talk about, but we think this video shows the event better than we could write about it.
AWS also hosted some fun talks during the hackathon. Will Roscoe and Adam Conway from Donkey Car came to present the project and mentor attendees. Over lunch, Stefano Soatto, Director of Deep Learning at AWS, discussed autonomous vehicles in the cloud. We also had Ryan Baumann and Amy Ghate of Mapbox show how the technology used in the hackathon related to their open source mapping platform. And that night, Andrea Censi, Oberassistent at ETH Zürich Dynamic Systems and Control and CTO of Duckietown, discussed his robotics education and outreach efforts. A few special guests came for the race as well. Werner Vogels and Charlie Bell, senior executives at AWS, kicked off the race. Chris Anderson, CEO of 3D Robotics, and Tom Soderstrom, CTO of NASA JPL, presented their views on autonomous driving, and they judged the final race.
The rest of the blog post features a few photo highlights from the hackathon and a video of the final race.
Getting the cars ready to assemble
The process for assembling and configuring the autonomous vehicle can be found in this repo and in our blog series. It also includes a full materials list with links on where to purchase the individual components. The main components are the RC Car, Raspberry Pi, Pi Cam, and Adafruit Servo HAT. You can purchase all of this for less than $250. We set up the basics, and distributed one vehicle to each team. Here are our previous blog posts that detail how you can do this yourself.
1) Build an Autonomous Vehicle on AWS and Race It at the re:Invent Robocar Rally
2) Build an Autonomous Vehicle Part 2: Driving Your Vehicle
3) Building an Autonomous Vehicle Part 3: Connecting Your Autonomous Vehicle
4) Building an Autonomous Vehicle Part 4: Using Behavioral Cloning with Apache MXNet for Your Self-Driving Car
Will and Adam discuss the Donkey Car project
Will and Adam came out to mentor attendees throughout the hackathon, and talked about how they got started together.
Attendees put on the final touches, and could add a few additional sensors, such as an USB speaker, IR data collector, Hall effect sensors, and an accelerometer. We also required them to stream data from their cars to a dashboard.
We provided training data for people to jump-start their training, but we were pleasantly surprised to see attendees sharing their training data and learning from each other. Just as when you drive your car, their cars had to expect the unexpected. They incorporated both left- and right-hand turns of various degrees, and responses to unexpected obstacles, to give their neural network exposure to many different situations.
In fact, the winning team incorporated all the data they could, and they focused on improving their car’s performance in hard turns.
After 15 hours of hacking on Monday, Gregory Pierce and Mohan Muppidi won the final race in a tie breaker with their car, It Worked Before. They started Robocar Rally with no deep learning experience, and came away with some awesome Robocar Rally swag– including the cars they raced.
Although this year’s race is over, there’s plenty more you can do. Check out Donkey Car and DIY Robocars for local meetups in your area, and learn more about AWS machine learning services and AWS IoT.
Updated March 12, 2018