AWS DeepRacer League 2019 Finals

After many races at AWS Summits and the Virtual League had completed, it was time to pick the winner of it all. The finalists have gathered in Las Vegas during AWS re:Invent 1-6 December 2019 to fight for the bragging rights (and a reward).

AWS DeepRacer League

AWS DeepRacer League 2019 included:

  • a series of 21 physical races at AWS Summits around the World from which all winners qualified for the finals; also three best performing runner-ups won their spots
  • a series of 6 virtual races that gave the six places for the winners and eighteen for the runner-ups

The finalists received an expense-covered trip to Las Vegas for the AWS re:Invent conference to fight for the title against the defending champion, Rick Fish, who also won his place.

The group was joined by top three racers in the last-minute AWS re:Invent qualifier and racers selected through events organised by some of AWS’ customers. This made a total of 64 finalists.

MGM Grand Arena

The finals’ rules

There were three phases of the finals to select the winner:

  1. Round 1: Qualifying 64 racing
  2. Round 2: Bracket of 16 racing
  3. Round 3: The Final 3 Showdown

The reward included:

  • A place in the AWS DeepRacer League 2020 finals
  • Bragging rights to the insanely heavy cup and a right to give a name to a track in 2020
  • A Coursera online Machine Learning degree scholarship with a maximum value of up to 25,000 USD or the equivalent in AWS credits

But before the finals had started, there was one more qualifier.

Last-minute AWS Summit

On Monday the conference attendants had their last chance to qualify for the finals. The race took place on the re:Invent 2018 track at the Quad at Aria where also the AWS DeepRacer Community Boot Camp took place. The interest was so high that the queue was filled within five minutes of the opening.

The lighting conditions were quite challenging which caused the contestants to go noticeably slower than in previous summits.

In the end the the last three finalists were:

2.Ensono – Phil J00:10.023

The nice detail from AWS was sharing information about the finals on the information screens at the conference:

The results of the last-minute summit

Round 1: Qualifying 64 racing

The finalists raced in four groups of sixteen with top four qualifying. Each racer had four attempts to get the best lap time and each attempt was a four minute window at a given track.

Starting groups for Round 1

Tuesday was the first day the finalists could try their models on the 2019 track. The track were really big – 10×8 meters is not something you will fit into your living room. Some even decided to train more models last minute to prepare for the afternoon.

Groups A and B were mainly filled with Summit races’ winners whereas C and D were mostly for those who qualified through the virtual league. All groups worked through the afternoon and the evening to determine the ones to return the following day.

You can watch the highlights of the Summit and Round 1 on DeepRacer TV:

Here are the results from each group:

Nalbam from the Community has assembled the standings accross the groups with top 20 results of the day:

Thank you Nalbam! Columns: overall place, time, group, place in group, racer name

Round 2: Bracket of 16 racing

Remaining sixteen racers were placed in a bracket and had a head-to-head match to determine a winner to go to the next round. The remaining two racers qualified for the round of three. The round followed the double elimination system, so defeated ones were entering last chance bracket where they were joined by others who fell out from undefeated bracket, and continued the fight for the third place.

Each match was a race between two competitors where each had two rounds of four minutes to clock the best time and in the end the fastest time determined the winner.

Two undefeated racers and one from the last chance bracket qualify into the great final race on stage before the Thursday Keynote.

Double elimination bracket. Source: AWS Blog

In the second round there one could see some exciting duels. Competitors did not waste the time between the races and trained even faster models. The undefeated series mostly involved times under ten seconds.

It’s worth mentioning that Sola managed to beat her own track reckord by a second: 8.089s will be really hard to beat.

Source: AWS DeepRacer 2019 Finals Top 16 results

You can watch Round two and the final on DeepRacer TV:

Round 3: The Final 3 Showdown

The rules changed for the final: the racers had a single 90 seconds try each. Each driver had a dedicated car. All this happened on stage, in front of the Keynote attendants.

Many of the community finalists came to see the finals.

New conditions brought old challenges – the lights did not provide as reliable environment as at the MGM Grand and the racers struggled to get reliable laps in their time slots. The times also reflect it: 10-12 seconds is much slower from what we’ve seen from Sola, Fumiaki and Roger in previous days. Interestingly, in the DeepRacer TV videos the cars can be seen running into the wall where the track manager’s shadow was cast. I cannot say for sure that this was the reason for a struggle, but it’s an option worth reviewing.

In the end Sola completed her domination of the league winning before Fumiaki in the second and Roger in the third.

You can watch the finals video (different from the DeepRacer TV) on YouTube:

Boot Camp

The AWS DeepRacer Community Boot Camp videos are being uploaded to YouTube. We will be posting the whole playlist to the blog, but in the meantime you can see them on our Community channel. Please subscribe to the channel so that you don’t miss out on more content that we’re preparing.

The season has ended, but not the fun

AWS DeepRacer Leage 2019 is over. We have learned a lot, met many brilliant people, had lots of fun, frustration, hopes and disappointments. We have formed a great community Where we got to share all of that. Join us as we prepare for the 2020 season. See you!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.