We’ve completed six out of eight months of racing in the 2023 AWS DeepRacer League. Let’s have a look at how it’s been working out so far and what’s left to win.
AWS DeepRacer
AWS DeepRacer is a 1/18th scale autonomous race car but also much more. It is a complete program that has helped thousands of employees in numerous organizations begin their educational journey into machine learning through fun and rivalry.
Visit AWS DeepRacer page to learn more about how it can help you and your organization begin and progress the journey towards machine learning.
Join the AWS Machine Learning Community to talk to people who have used DeepRacer in their learning experience.
The league
You may want to later look at the announcement of the league to see the detailed rules: in English, in Chinese, in Polish.
It may seem the league got a little quieter this year with no regular updates on progress. Nothing farther from the truth though. Up until August we’ve seen over 4400 racers from across six regions fight for prizes including a ticket to AWS re:Invent in Las Vegas where the League finals will be taking place (travel, hotel, conference entrance and pocket money included) or Amazon Vouchers/PayPal transfers. We’ve also seen over 600 racers participate in the physical racers at the regional AWS Summits with 12 of them also winning a chance to participate in the championships. You can see the detailed results at https://drboard.breadcentric.uk – including virtual races, summit races and cumulative points.
Changes
There were quite a few new elements to the league this year:
- six regions – racers compete in one global league but win prizes within their region of residence
- cumulative points – if it’s hard to win, keep trying – at the end of the season top point scorers without the tickets, well, will get the tickets 🙂
- submissions limit – at 50 submissions in each of the monthly races you’ll need to try harder to get your entry just right
- counter-clockwise racing – a nice addition to make old tracks a challenge again
To me splitting the racing into six regions is the next step in the AWS mission to popularize machine learning. I have seen how DeepRacer encourages learning and making it more accessible definitely helps: while it makes it a little harder for people like me to qualify (I haven’t qualified yet), it creates great opportunities for those who could not bring forward enough resources to challenge the top racers globally. And this is where DeepRacer comes to its finest – sparking curiosity and giving opportunities globally. We are having the first finalists from Turkey, Uganda, South Africa or Ghana.
The cumulative points idea has been there for the league in 2019 but has been iterated on and significantly improved, leveraging the regional split. With the recent addition of the prize information on https://drboard.breadcentric.uk you can now see in the cumulative list who still has the chances to qualify. It’s actually pretty interesting as while cumulative point winners will only be announced after October, some have already built up enough points to guarantee a place, and in other regions the battle continues.
Submissions limit might sound like a cosmetic change but if you look at races in 2022, the winners often had thousands of submissions to their name. Now their models must be more stable to present something competitive enough.
Where are we now?
Let’s look at some numbers.
Qualified to the championships at AWS re:Invent conference in Las Vegas: 51 racers
- 36 from virtual races,
- 12 from AWS Summits,
- 1 from pre-season race,
- 1 from the re:Invent Open race,
- 1 winner of AWS DeepRacer Championships 2022
$400 winners: 180 racers (5 in each region in each race)
Overall racers so far (March-August): 4418
Overall racers per region (March-August):
- Asia Pacific: 1695
- Europe: 648
- Greater China Region: 532
- Middle East & Africa: 92
- North America: 1303
- South America: 148
Overall racers per race:
- March: 2630
- April: 1740
- May: 2450
- June: 2760
- July: 2570
- August: 2770
- September (at the time of writing): 2460
Fastest and slowest times per race:
- March: 54.195s – 12:22.809
- April: 55.522s – 15:07.489
- May: 1:00.526 – 16:22.192
- June: 51.929s – 15:24.595
- July: 53.915s – 12:22.209
- August: 1:08.587 – 18:20.750
- September (at the time of writing): 51.457s – 11:31.666
To be honest, I did not know 18 minutes was possible. That said, most of those races are actually reset penalties. People train and try their models. Some work, some take over 18 minutes to complete three laps. This submission had 255 resets, each reset is 3 seconds so overall it’s 765 seconds which is close to 13 minutes, and that would mean the model took about 6 minutes to complete the track. Six minutes is in line with slowest clean runs in August, so it makes sense. Just remember: everyone was starting slow.
What’s left?
We have 7 days left in September. Many racers keep their submissions to last days, hoping to build that winning model and submit it late enough to make responding harder.
At the time of writing these people are in the qualifying positions:
- GiangCao (Greater China Region)
- Eviden-Mark-Ross (Europe)
- DoomBuggy (North America)
- KM-Harrier (Asia Pacific)
- Akiva (Middle East Africa)
- DaeMonst (South America)
Also in each region top 5 (who haven’t yet won that) will get $500 in vouchers and top 10% of racers (again, who haven’t yet won that) will be eligible also for $50 in vouchers.
Same set of prizes is waiting in October race, and seven more places through cumulative points. This will be interesting in a few places, seems that Europe, Greater China Region and North America may have change of leaders there.
And there’s more to win, and it will be exciting, and it will be announced next week. So stay tuned and check back for the next post. And in the meantime get your virtual wheels spinning!
We’re Coming! Tomasz, just trained up a new batch of students and we’ll see how they do! They are already in the race for September but October should see some real progress. I had them read this blog and I think they were inspired by your words. Thanks for all that you do for DR!