AWS Summit London 2022: Training your own AV with DeepRacer

The tiny car had a tenancy to crash if parameters were off.

Ben Wodecki

April 27, 2022

2 Min Read

The tiny car had a tenancy to crash if parameters were off.

Imagine being able to program your own autonomous vehicle, teaching it how to drive on a racetrack in a bid to beat a time trial.

That is what AWS DeepRacer is – though at just 1/18th scale.

DeepRacer is designed to teach those with little or no experience in AI to grow their skills. It provides a hands-on experience with machine learning through a cloud-based 3D racing simulator.

The offering was showcased at AWS’s Summit in London, with attendees being given a chance to see the tiny car operate in person.

The car is designed as four-wheel drive monster truck chassis and houses technical specs including 360-degree Lidar, 4GB of RAM and four-megapixel cameras. It runs on a lithium polymer battery and its 32GB storage is expandable.

Ananth Balasubramanyam, a senior solutions architect at AWS, led a talk on how the system – outlining that the car becomes autonomous due to reinforcement learning or RL.

RL is where a mathematical model is built – in this instance, to enable autonomy. The model is trained in a specific environment which in turn allows the system to learn from the actions it takes and reward it for being correct.

Balasubramanyam likened it to teaching a dog a trick using treats – “you don’t reward bad behavior,” he said.

Ananth Balasubramanyam explaining RL

The DeepRacer cars use a six-stage neural network, that takes pixels from an image, which then feeds the data through a feature extractor, which is sent to a policy network that determines a corresponding action.

The AWS architect went on to explain that model optimizers like Intel’s OpenVino tool could be used to adjust models to obtain specific end-goal targets.

One such target would be to avoid crashing into the wall – which is something that happened on occasion during demonstrations at the event.

DeepRacer doesn't always crash, however, as AWS hosts a highly competitive league format that pits developer teams against one another who try to achieve the fastest times.

Over 70,000 dev teams from 150 countries participated in the 2021 iteration of the league, hoping to scoop the coveted trophy and up to $20,000 in prize money.

The 2021 DeepRacer cup went to Sairam Naragoni from JPMorgan Chase with colleagues Yousuf Nizam and Tyler Wooten placing second and seventh, respectively.

About the Authors

Ben Wodecki

Assistant Editor

Get the newsletter
From automation advancements to policy announcements, stay ahead of the curve with the bi-weekly AI Business newsletter.