The rivalry between the U.S. Military Academy and the U.S. Naval Academy is one of the most enduring in college football, dating back to their first game in 1899. Each year, the football teams from each academy compete in a game that has become a well-loved tradition in college sports. The century-old rivalry, while friendly, is intense, with the victor earning bragging rights for the year. The game itself embodies the spirit of the inter-service rivalry of the U.S. armed forces. Even the mascots (the Navy Goat and Army Mule) participate in the rivalry by playing pranks on each other.
AWS DeepRacer Army vs. Navy competition overview
Continuing the tradition of inter-service rivalry, Army and Navy competed in the inaugural Amazon Web Services (AWS) DeepRacer competition, a two-part event designed to teach machine learning in a dynamic and competitive environment. Army and Navy machine-learning teams gathered virtually to build the fastest machine learning–enabled autonomous race car, making strategic investments to train personnel in autonomous technologies like self-driving vehicles.
In the DeepRacer competition, teams got up and running with a workshop and hands-on tutorials to learn the basics of machine learning. From there, they developed their own models using Amazon SageMaker to train, test, and fine tune their reinforcement learning models. After their models were trained and tested in simulation, finalists experienced the thrill of a race in the real world by deploying their models onto autonomous AWS DeepRacer cars that raced on a physical track.
Powered by AWS RoboMaker, AWS DeepRacer is designed to offer a fun and hands-on way to get started with machine learning through a cloud-based 3D-racing simulator and a fully-autonomous 1/18th scale race car driven by reinforcement learning. Reinforcement learning is an advanced machine-learning technique that takes a different approach to training models. This type of machine learning relies on the small-scale cars to learn from their environment through a reward system. For example, staying within a vehicle lane earns a reward, while veering out of a lane results in no reward.
36 racers participated in the inaugural event. The top seven semifinalists from Navy were selected to battle the top seven from Army in the final competition. The finals were held with the racers gathered remotely while a live AWS DeepRacer-pit crew prepared the autonomous race cars with each team's trained model, which was livestreamed to participants and viewers. Each car was loaded with the individual team trained reinforcement learning model and multiple timed races were conducted on the physical track. The best time from each team was recorded and continually updated on the leader board. Virtual crowds cheered from across the nation as nearly 300 fans streamed the competition live.
Sai Liu, a software U.S. Army Communications-Electronics Command Software Engineering Center, development lead and a member of Army's Command and No Controls DeepRacer team, said machine learning was something his entire team was curious to learn more about. While none of the systems he helps sustain use machine learning at the moment, what he learned could come in handy moving forward.
"I think it's good to learn about the new technologies," said Liu. "So in the future, if there's anything that can allow us to apply our knowledge, then we have a little background to start."
The results
After a winning time of 7.112 seconds, Navy Team DeepBlueSea, led by Rob Keisler of Naval Information Warfare Center (NIWC) Atlantic took home the AWS DeepRacer golden trophy.
"This event was a great learning experience and one of many ways we will ensure the Navy leverages data and AI to achieve decision superiority in warfighting and business," said Keisler.
Plans for AI and machine learning continue to expand at NIWC Atlantic to include exploring how reinforcement-learning techniques can be applied to a wide range of naval problems like autonomous navigation, command and control, scheduling and logistics, automation, and condition-based maintenance via the Naval Innovative Science and Engineering (NISE) program.
Army and Navy plan to continue this competition in line with their long-standing rivalry, and an event is planned to coincide with this year's Army-Navy football game in December.
Until then, people from all backgrounds and skill levels can participate in public DeepRacer events throughout the year.
Individuals can also learn more about using simulation to train reinforcement learning models in a re:Invent 2020 session titled "Reinforcement learning and robotics."
To learn more about AWS RoboMaker solutions, contact the AWS Robotics DOD team.
