Driving machine learning with Tyler Technologies and AWS DeepRacer

Machine learning (ML) is one of the fastest growing areas in technology. AWS DeepRacer helps anyone with an Amazon Web Services (AWS) account get started with ML through autonomous driving. Tyler Technologies, an AWS Partner Network (APN) partner, held a DeepRacer event at their annual internal conference, Tyler Mesh, in September in Indianapolis, Indiana, to help their organization learn about ML and ML on AWS.

Racing with AWS DeepRacer

AWS DeepRacer is a fully autonomous 1/18th scale race car, which provides a fun and interesting way to get started with reinforcement learning (RL). RL is an advanced ML technique that takes a different approach to training models than other ML methods. AWS DeepRacer allows you to get hands-on with RL to experiment and learn through autonomous driving. To start, the user drives in a virtual car on a virtual track, then deploys physical, trained models onto a real track for a race.

Race day: The event at Tyler Mesh

Prior to the conference, teams of four people worked together to create a model that would race on the track at the event. AWS held six training workshops over the course of two months for the 17 racing teams from Tyler Technologies. During the two-hour workshop, attendees learned about AWS DeepRacer, ML, and how to create and train their first racing model.

Participants from Tyler Technologies brought their pre-built and trained model on a USB flash drive to be loaded onto a physical race car. Each team was given a four-minute slot to race their model as many times as possible, with the best time counted in the standings. After their four-minute slot finished, the racing teams could tweak their models and return later to race again in an attempt to beat their previous best time. By the evening, 16 teams had raced a total of 32 times.

On the second and final racing day, each team was given up to three four-minute slots, and were required to use at least two different race models. After regular racing ended, the three leading teams competed in the final race. The team “Skid Marks” team was crowned the winner with a race time of 12.53 seconds, beating team “404 –  Team not Found” with a time of 13 seconds.

AWS DeepRacer: Under the hood

AWS DeepRacer (as shown in the diagram below) uses several AWS services – Amazon SageMaker and AWS RoboMaker – which enable the building, training, and evaluation of the models used on the track. Through AWS Organizations we create accounts for each team to build and test models. This isolates each AWS DeepRacer event from others and creates an easy way to remove access to all services after the event.

For many participants, this was their first exposure to artificial intelligence (AI) and ML. When the event completed, many of the participants said the experience fueled their desire to look for opportunities to implement machine learning.

“Tyler Technologies views machine learning (ML) as an emerging component of the data services available through our data and insights solutions. We envision ML playing a key role in our Socrata data platform in order to derive insights on problems facing governments today. For example, ML services fit well with time-series forecasting and evidence-based policy making when it comes to driving sound decision-making. With an extended client base across a variety of domain areas, including public safety, courts and justice, K-12 education, appraisal and tax, civics services, and ERP financial services, machine learning is another technology that clients can leverage as part of their day-to-day tool management toolkit,” said Mike Teeters, Corporate Development Manager (Products), Tyler Technologies.

Learn more about how to run your own AWS DeepRacer event.