The AWS DeepRacer event, hosted last Friday by simulation technology provider Computer Modeling Group (CMG) in Calgary, showed how access to the right resources, a good old showdown and attractive prizes could be the answer to the global upskilling problem.
And what could be more fun than seeing machine learning (ML) in action in a thrilling race between miniature autonomous cars on a track, hyped by competitors ready to showcase their hard work?
For two weeks or so, developers of all skill sets trained and tested their ML models using AWS’ cloud-based 3D racing simulator and virtual race cars. They then loaded these models onto real, fully autonomous 1/18th scale race cars, provided by AWS, to compete for prizes.
The models are trained using an advanced ML technique called reinforcement learning (RL), whereby the computer learns how to behave in a specific environment through trial and error and the corresponding rewards and penalties.
The miniature vehicles have cameras that collect image data of the track, and the RL model controls the car’s steering and throttle. The car also has a tiny computer under the hood which translates the model into commands like ‘turn the wheels’.
But the only thing that competitors can control when the vehicles are going head-to-head on the track is the amount of power being supplied to the car. The ML model created by the competitors does the rest.
An event like this shows how concepts like AI and ML that sound broad and abstract can come to life and get better over time, said Coral Kennett, AWS education lead.
Successfully getting a car around a track is essentially an optimization problem that you can encounter in real-life business problems, explained chief technology officer of CMG, John Mortimer. “It is very much the same as when you look at the speed and performance challenges that we’ve got in simulation. It has to be able to do the thing it needs to do, but you want it to go fast and you want it to be reliable.”
Seeing the broader practicality of the technology in action, appreciating it, and getting excited to learn about it is key, adds Pramod Jain, chief executive officer of CMG.
Related: AWS makes machine learning fun
For DeepRacer participant Vivien Lui, the event puts the spotlight on young people, some of whom are interns, who end up doing really well in the competition.
She advises young AL/ML enthusiasts to be fast learners, to be open-minded to new technologies, and maintains that even if they do not have advanced knowledge, they can start by subscribing to a YouTube channel or looking up free courses from universities to learn the basics, or even go into building, training, and fixing AL/ML models.
“We really feel a strong responsibility to provide many pathways for upskilling with digital skills for employers, for students and for all kinds of different people who are looking to increase their technical knowledge,” said Kennett.
There is no cost to join a DeepRacer league, but you pay for training, testing (US$3.50 per hour) and storing (US$0.023 per GB) your ML models on the AWS DeepRacer service. Users also get 10 hours free and 5GB of free storage during their first month, which AWS says is “enough to train your first time-trial model, evaluate it, tune it, and then enter it into the AWS DeepRacer League.”
Developers can enter the racing leagues in multiple ways:
– Use a 3D simulator to compete in AWS virtual leagues with no cars needed
– Bring their own miniature cars to real leagues
– Buy their own AWS autonomous car for C$400 and participate in a real league.
Developers around the world also compete to earn one of 50 spots to participate in the DeepRacer global championship event, which takes place annually at AWS re:Invent.
To further address the issue of upskilling, Kennett hinted at the opening of a Calgary cloud computing region in late 2023/early 2024, which she says will help organizations with data residency requirements and have a big economic impact on the city.