AWS DeepRacer - RL on Wheels


2018-12-20, 07:58 Posted by: Sami Lehtinen

In our previous AWS re:invent recap blog post we were covering a bunch of cool new releases at re:Invent 2018. While we saw releases in varying categories, the one probably receiving the most was SageMaker and Machine Learning in general. In his keynote, Andy Jassy revealed a new addition to SageMaker family by introducing Reinforcement Learning (RL for short), which is considered to be one of the three main paradigms of machine learning.  

Without going too much into the details of RL, the basics involve the algorithm to do exploration and exploitation of the environment with each action tied to an observation and to a reward function. Further reading on the math and science behind the paradigm can be found from e.g. here.  

The Launch

After the introduction of SageMaker RL, something I certainly wasn’t expecting happened. Andy launched AWS DeepRacer, a 1/18th scale RC-car powered and controlled by Reinforcement Learning. Wait… WHAT? WHY? Is this a joke? That was my honest initial reaction. And to bend my mind even more, he also launched the DeepRacer League, consisting of actual races around the globe at AWS Summit events.  

After thinking this for a while, it all started to make a lot of sense. Launching a cheap device like this (pre-order for $249, US only ATM), it truly democratizes the technology by enabling hands-on experience for every developers and data scientist. Just like he mentioned in his keynote.

I have a strong feeling that DeepRacer is going to be a good enabler and will serve as a gateway into the world of machine learning for a bunch of new developers. I’m certainly looking forward seeing the development in this area and can’t wait to get my hands on a DeepRacer in the near future. Some lucky ones actually received their very own DeepRacer at re:Invent. They just happened to attend to a certain machine learning related session and were given one. I wasn’t that lucky, but my Swedish collegue, Viktor Svensson, was one of the lucky ones and got a DeepRacer. How cool is that? Must say, I’m a bit jealous and I think I really must pay him a visit or otherwise convince him to send it to me.  

So, are you ready to start exploring the world of Reinforment Learning? Well… start your engines!


comments powered by Disqus