The Amazon Deepracer
The Amazon Deepracer challenge is Amazon Web Services’ latest effort on autonomous vehicles. In short, the Deepracer project is a 1/18 scale race car that is driven by reinforcement learning, and a cloud-based 3D racing simulator that comes alongside it. With this advanced machine learning technique, the car’s algorithm learns very complex behaviors without any labeled training data, and it can also make real-time short-term decisions while doing optimizations on long-term rewards.
The competition itself can be divided into two parts. In the first part, the contestants are competing in a league-like structure to finish in the top 10% of the participants. This part of the contest happens in a virtual environment provided by AWS so that the teams can try their algorithms on virtual cars. With this simulation environment that is already different than other AI competitions, the participants get the chance to race against other competitors in real-time as well. At the end of each league, the top 16 race teams will have the chance to compete on livestream against each other.
Image taken from the F1 ProAm race in May 2020
During the second part of the competition, the top 3 teams will have the chance to try their algorithms each month on a real vehicle that has sensors and speakers. As the main goal of the competition, they will aim for the fastest (and safest!) lap in the designated race track to show that their algorithms are applicable to real-life situations. The competitors can also check the DeepRacer Pit Stop to gain more knowledge and use state-of-the-art methods of reinforcement learning specifically for this challenge as well as share their own information with others. If you are curious, you can check it out by yourself to discover this fascinating pool of material!
