Computers

Amazon tempts developers to machine learning with toy race car

View 5 Images
The DeepRacer is being offered as a way for developers to try out reinforcement learning
Amazon
The DeepRacer is being offered as a way for developers to try out reinforcement learning
Amazon
The DeepRacer 1/8th scale autonomous race car has a 5 mexapixel camera out front and accelerometer/gyro sensors inside
Amazon
Developers will create reinforcement learning models using software, test them on a 3D virtual race track and then train the real-world DeepRacer to drive itself
Amazon
The AWS Championship racing leagues will kick off in 2019
Amazon
The DeepRacer includes a built-in compute section featuring an Intel Atom processor, 4 GB of RAM, 32 GB of internal storage, a 4 megapixel camera, six USB ports and one HDMI, accelerometer and gyro sensors, and 802.11ac Wi-Fi
Amazon
View gallery - 5 images

Amazon Web Services has revealed an autonomous racing toy at re:Invent 2018 in Las Vegas this week. The DeepRacer is a 1/18th scale all wheel drive monster truck that's aimed squarely at budding developers, who will create reinforced learning models in software, test them on a virtual race track and then train the toy car proper to compete in a real-world global racing championship.

The DeepRacer includes a built-in compute section featuring an Intel Atom processor, 4 GB of RAM, 32 GB of internal storage and comes loaded with Ubuntu OS, Intel OpenVINO computer vision toolkit and ROS Kinetic (robot operating system). It's been designed to get developers into reinforcement learning, a form of machine learning that uses trial and error to achieve goals and successful outcomes.

Devs will start their machine learning journey by setting up reinforcement learning models in Amazon software. They'll use this "simple, yet powerful, interface" to train, test and tweak a virtual car in a cloud-based 3D racing simulator. The fine-tuned virtual learning model can then be uploaded to a community of like-minded machine learning enthusiasts.

"The AWS DeepRacer 3D racing simulator and car provides an ideal environment where you can test the latest reinforcement learning algorithms and simulation-to-real domain transfer methods," said Amazon Web Services.

The DeepRacer includes a built-in compute section featuring an Intel Atom processor, 4 GB of RAM, 32 GB of internal storage, a 4 megapixel camera, six USB ports and one HDMI, accelerometer and gyro sensors, and 802.11ac Wi-Fi
Amazon

The real-world car features a 4 megapixel camera, six USB ports (including one microUSB) and one HDMI, accelerometer and gyro sensors, 802.11ac Wi-Fi and a 7.4 V/1,100 mAh Li-pol battery for the drive section and a 13,600 mAh battery for the compute section.

The idea is that the developer would earn their reinforcement learning chops in a safe virtual environment, and then train the actual DeepRacer car before taking part in a new global racing league being launched early next year, with prizes up for grabs.

The DeepRacer is available for pre-order now for US$249 (list price is $399), ahead of release on March 6, 2019.

Source: AWS

View gallery - 5 images
  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
1 comment
Daishi
This is interesting work and I hope to see some interesting work out of the project. Humanity has all of the technology available today to have $200 autonomous telepresence robots in households that can be controlled by audio. The technology required is pretty much a tablet on a stick bolted to an RC car. Knightscope security robots are 300 lbs and cost $7/hr but all you need a security robot to do is autonomously navigate and have a camera and that hardware essentially exists on Amazon's smaller $200 version here. What's missing is software. I was hoping we'd be much closer to solving inexpensive telepresence robots by now but we aren't because almost everyone involved has been chasing the dead end path of thinking robots have to have mechanical legs instead of wheels. Robots with legs have had almost 30 years of failures creating things that are expensive and useless. It's time to make adjustments to the approach.