Blog

Swish AWS DeepRacer

October 28, 2022 By Jay Borthen

Since its inception circa 2018, the Amazon Web Services (AWS) Deep Racer platform has gained in popularity within the Machine Learning (ML) community. It’s touted as a “fun way to get started with Reinforcement Learning (RL)” and it accomplishes this by modularizing required hardware and abstracting away the intricacies of streaming data.

Reinforcement Learning

The DeepRacer vehicles are fully autonomous 1/18th scale model cars that learn how to navigate from virtually trained ML models using streaming video and LiDAR data.  The ML models are developed with the guidance of customizable reward functions written in Python.  During the virtual training, when the car performs an action that will ultimately make it faster or more agile, the algorithm is positively rewarded with a higher score.  Similarly, if the car performs an action – such as driving off the course or running into another vehicle – then the algorithm is negatively rewarded with a decreased score.

DeepRacer Leagues

There are many commercial organizations and government agencies as well as individuals who compete in the DeepRacer leagues.  The races are run virtually in the AWS RoboMaker environment or with real DeepRacer vehicles on real-life courses.

Swish Engineers unboxing DeepRacer car
Making initial configurations for LiDAR/Video sensors on DeepRacer

The Swish DeepRacer

Swish began developing its own DeepRacer models in the summer of 2022 and we quickly qualified for the Pro circuit in which organizations and individuals compete for recognition and prizes.  While the performance of the cars is largely dependent upon how much time each team is willing and able to spend training their models, Swish has been able to consistently perform well with only a moderate amount of model training time.  Cloning and training models is easy to do in the DeepRacer and SageMaker platforms and only requires the push of a button.  AWS also makes it incredibly easy to save your models in S3 and retrieve them when needed.

Within the next few months, Swish hopes to organize or participate in our first live competition.  If your organization is interested in learning more about DeepRacer or Swish, please feel free to reach out at COETeam@swishdata.com

Also, check out the annual DeepRacer competition between the US Army and US Navy which began in May of 2021 (https://www.aboutamazon.com/news/aws/army-and-navy-teams-compete-in-first-annual-aws-deepracer-competition).  A Naval Information Warfare Center (NIWC) Atlantic team called “DeepBlueSea” received the first-ever DeepRacer golden trophy after their win over six other Navy teams and seven other teams from the Army.

A screenshot of the virtual training environment is shown in the figure above and more information can be found at https://aws.amazon.com/deepracer/.