AWS DeepRacer Championships 2023: Celebrating 5 years of racing

It’s a wrap on the 2023 Season of DeepRacer. Let’s have a look at how the finals have played out.

67 racers came to Las Vegas from 28 countries showed up for the racing this year. Some came in person, some couldn’t, and the AWS DeepRacer Community has worked with AWS to make sure they could. To us AWS re:Invent is about learning, networking and fun, but it all is based on top of the community.

But in the end it is a competition, and the winner gets $20,000 cash prize, so let’s see how that has worked out. Below you will find a brief summary of each even of the championships.

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.

You might be interested in other posts from our blog

Round 1

On Monday Morning we have met at the Brooklyn Bowl for the first stage of the competition. This year we started with a virtual race. This allowed to have six parallel races running, one for each of the regions. Top five racers from each region qualified to round 2.

While we’ve seen virtual races happening all year, this has proved to be a challenge to some finalists, especially those who qualified through physical races at AWS Summits.

Below are results for each region:

Asia Pacific

RankRacerCountryTime
1 QCodeMachineAU21.731
2 QTonyJNZ21.798
3 QVersion-InfiniteAU22.197
4 QJPMC-SmartAccelerators-HYDIN22.734
5 QpatelaniketIN23.063
6RayG-DBSSG23.137
7Deepak-dpkIN23.402
8BSiID24.002
9RobinCastro-DBSSG24.663
10kashiJP26.47
11PITS-2JP39.802
12m45JP50.207

Europe

RankRacerCountryTime
1 QDBroGB21.743
2 QMattCampFR21.867
3 QDuckworthDE22.065
4 QEnsono-ErnestoGB22.131
5 QMarco-EvidenGB22.134
6JPMC-London-GriffcGB22.136
7IcePandaBoomFI22.2
8Eviden-Mark-RossGB22.399
9DaVinciDriverNL22.593
10LeadingAICA22.799
11JochemNL23.331
12Rober-rNiuboES23.867
13JPMC-LONDON-HandsFreeGB24.727
14ZoDCH26.264

Middle East & Africa

RankRacerCountryTime
1 QEcho1911CN24.935
2 QXMaroRadoXEG26.001
3 QFiatLuxLU27.466
4 QJPMC-RacingRhino-SGSG28.136
5 QmericexTR32.596
6sivasTR32.996
7YuliyaKZ37.129
8SilEyeSN41.933
9Gr1zzli3UG50.398
10WernerOscarGH54.136

Greater China Region

RankRacerCountryTime
1 QCYC-NYCUTW21.267
2 QdcongtinhTW21.477
3 QYi-LI-NYCU-CGITW21.666
4 QBennyLBSCN22.733
5 QnickylCN22.995
6Ele1CN23.005
7DecisionTransformerTW23.53
8HinsCN23.594
9LIZHIHK24.337
10MadCrossCN24.531
11HK-WilliamLeungHK42.67

North America

RankRacerCountryTime
1 Qrosscomp1US20.333
2 QJJCA20.942
3 QOff-trackUS21.732
4 QferriteUS22.14
5 QAJM-Model-1US22.334
6PolishThunderUS22.337
7JPMC-Rogue-COLUS22.411
8AIDeepRacerUS22.609
9DoomBuggyUS22.665
10JerecUS22.867
11JPMC-C0nf1ckr3US25.601
12raj-1205CA48.597

South America

RankRacerCountryTime
1 QJPMC-BA-MiguelGAR24.134
2 QJPMC-Speedy-Gonzales-LATAMCL24.867
3 QMattKrederAR25.395
4 QJPMC-BA-D3AR25.671
5 QC3R0CO35.929
6DelarMonstCO38.732
7londoso-monstCO51.062
8DaeMonstCO62.803

Wildcard race

In the evening there was a chance to grab the last two places to join the qualifying 30. All the non-finalists have had a tough challenge ahead, having to race with many of the Round-1-eliminated racers. And indeed, while a third of participants did not participate in round one and some of them had pretty competitive times such as F1DevelopersIO, it was LIZHI (杜键文, Jianwen Du) and Jerec (Daniel Morgan) who have proven they had more to show in Las Vegas.

Overall, we have seen 31 racers from 10 countries on 4 continents participate in the wildcard race, here are their results:

RankRacerCountryFastest Lap TimeAverage TimeLap completion rationAttempts
1 QLIZHICN13.78821.989100.0%1
2 QJerecUS14.11221.401100.0%1
3rogue0197US14.24815.99100.0%1
4F1DevelopersIOJP14.57422.89580.0%1
5MadCrossCN14.72620.504100.0%1
6EvidenMarkRossGB15.25721.456100.0%1
7jcronUS18.15527.407100.0%1
8confickerUS18.36732.29580.0%1
9Atp6379US19.01625.438100.0%1
10skarlekarUS21.10530.853100.0%1
11BerbbobsGB21.31326.837100.0%1
12HinsCN21.79830.647100.0%1
13ahaCN22.27227.391100.0%1
14BSiID22.83530.066100.0%1
15RayGSG22.84726.154100.0%1
16AIDeepRacerUS23.5332.862100.0%1
17DTTW23.7531.484100.0%1
18PolishThunderUS24.08936.753100.0%1
19paradesiUS24.23829.44540.0%1
20ShanUS26.69536.105100.0%1
21robincastroSG26.69530.046100.0%1
22kashiJP27.02332.943100.0%1
23JPMCLondonGriffcGB27.14944.063100.0%1
24PITS_2JP28.71147.425100.0%1
25leadingaiCA33.72538.025100.0%1
26londoso_monstCO34.45838.661100.0%1
27m45JP36.76742.52780.0%1
28andreologyUS37.04745.162100.0%1
29makitGB42.28352.606100.0%1
30maartenUS45.30355.79770.0%1
31IcePandaBoomFI57.52757.773100.0%1

Thanks to the DeepRacer Event Manager, here are some wildcard race stats:

Total laps: 125
Total resets: 255
Avg resets per lap: 2.0
Avg laps per race: 4.0
Avg lap time: 00:30.746
Fastest lap time: 00:13.788
Slowest lap time: 01:11.063
Number of unique racers: 31
Number of races: 31
Most races by user: 1
Avg no. of races by users: 1.0

Round 2

32 Racers competed over two days. Each of the racers had two runs in which they were attempting to get the best time. In this round however, the times that mattered weren’t fastest laps. The result was taking three consecutive laps into account, calculating the average time. For instance, if the racer completed six laps, the best of average over laps 1-3, 2-4, 3-5 and 4-6 is their final result.

Let me just mention here how smooth the whole racing was at the Expo this year. A lot of attention has been dedicated to ensure good flow of the competition and on both days racers were done with their runs before time which in turn allowed non-competitive racers to have a bit more practice on the track, and that was highly appreciated.

The new timing format turned out to be much kinder to the cars and the pit crew. Having off tracks was so costly that racers did not risk pushing the limits unless the model showed sufficient stability. And even with that results were very impressive. Here they are:

RankRacerCountryFastest Lap TimeFastest Average TimeLap completion rationAttempts
1 Qrosscomp1US14.21214.574100.0%2
2 QDBroGB13.48914.633100.0%2
3 QFiatLuxLU14.47314.791100.0%2
4 QILEETW14.38415.3100.0%2
5 QJPMC-HYD-SmartAcceleratorsIN14.87315.566100.0%2
6 QnickylCN15.28115.628100.0%2
7 QJerecUS15.43215.745100.0%2
8 QLIZHICN14.93815.911100.0%2
9Echo1911CN14.78815.952100.0%2
10CodeMachineAU16.24516.625100.0%2
11DuckworthDE16.71917.22100.0%2
12aniketvp24IN15.84617.311100.0%2
13jcoronelcortesCL16.80517.339100.0%2
14AJM-Model-1US15.16817.873100.0%2
15dcongtinhTW17.26618.594100.0%2
16JJCA14.42119.286100.0%2
17mgugliattoAR18.49419.585100.0%2
18kadermeeraSG17.90619.776100.0%2
19YuchengChenTW17.83520.633100.0%2
20JPMCBAD3AR18.35820.746100.0%2
21BennyLBSCN15.73520.793100.0%2
22OffTrackUS18.74522.282100.0%2
23ErnestoGB18.87222.299100.0%2
24TonyJNZ17.50623.306100.0%2
25VersionInfiniteAU20.36324.129100.0%2
26C3ROCO21.6724.876100.0%2
27marcoevidenGB27.33827.86100.0%2
28MattCampFR19.98728.097100.0%2
29MattKrederAR28.7729.432100.0%2
30XMaroRadoXEG31.28535.032100.0%2
31ferriteUS23.41140.642100.0%2
32mericexTRDNSDNSDNS0

Here are the stats:

Total laps: 347
Total resets: 293
Avg resets per lap: 0.8
Avg laps per race: 5.6
Avg lap time: 00:23.076
Fastest lap time: 00:13.489
Slowest lap time: 01:12.551
Number of unique racers: 36 (involves some practice runs not recorded above)
Number of races: 62
Most races by user: 2
Avg no. of races by users: 1.7

Round 3 & The Final

Once the top eight racers have been established the championships have entered the deciding phase involving a double elimination bracket. Racers have been joined in competing pairs in which each racer had two runs, first runs were ordered based on the seeding and the second ones based on the times in first runs. If the slower racer would not improve in the second run, the faster racer would be through without theirs.

The top eight has had a great combination of racers:

  • rosscomp1 – Ross has been racing for quite a while and always on top level. After dominating the virtual racing and winning in rounds one in North America and round two, expectations were set very high
  • DBro – just like Ross, he’s always presented solid results both virtually and physically. While second in round two, Darren’s best lap was much faster
  • FiatLux – Joussef was the Accenture Wildcard racer who could not come to Las Vegas and was raced through a proxy racer, his colleague Doug Wozniak. Apparently Joussef was not expecting to make it to round two and trained his model after the Monday round one results
  • ILEE (Yi-LI-NYCU-CGI, 李頤) – last year’s title defender. While the NYCU students did not dominate the physical racing this year, qualifying form the fourth spot and with the third fastest lap is a clear statement of abilities
  • JPMC-HYD-SmartAccelerators – representing JPMC, Tarun was the only racer to make it through both rounds of the championship for his company. He also qualified through the Mumbai summit which proved he had a lot to show
  • NickyL – Yaojin Liang, Nicky, has managed to go much farther compared to 2022. Having to stay home last year and only being able to participate as a proxy has clearly increased his appetite for the title
  • Jerec – Daniel has stumbled in round one but corrected course in the wildcard race and topped it in round two
  • LIZHI – 杜键文 (Jianwen Du) completed the final eight just like Daniel, showing that the wildcard race is a gateway for amazing racers to have one last try if virtual holds them back. LIZHI also could not come in person and raced through a proxy racer, Yiwei Huang.

The competition was fierce and many races required all four runs to determine the winner. In the end we reached the three last racers:

  • DBro
  • FiatLux
  • JPMC-HYD-SmartAccelerators

Final 8 stats:

Total laps: 215
Total resets: 164
Avg resets per lap: 0.8
Avg laps per race: 6.0
Avg lap time: 00:21.018
Fastest lap time: 00:13.854
Slowest lap time: 01:24.141
Number of unique racers: 8
Number of races: 36
Most races by user: 6
Avg no. of races by users: 4.5

The Final

The final was carried out in the same format as in 2022. Each racer was given one lap to complete with a running start. No retries, no retakes, no second laps. If there’s a reset, chances for a fast lap are ruined.

Thanks to the wonders of technology, let’s enjoy the final as it unfolded:

AWS DeepRacer Championships 2023 final

What an amazing final. FiatLux, racing for the first time, as a proxy, beating the track record in the final single lap of the tournament, has become the 2023 AWS DeepRacer League Champion, grabing the winner trophy and $20,000 prize. Congratulations Youssef, and see you in Las Vegas in 2024!

AWS DeepRacer Championship Final result

Let’s just finish this write-up with the winners ceremony:

And here are the final stats:

Total laps: 3
Total resets: 2
Avg resets per lap: 0.7
Avg laps per race: 1.0
Avg lap time: 00:17.613
Fastest lap time: 00:13.393
Slowest lap time: 00:21.145
Number of unique racers: 3
Number of races: 3
Most races by user: 1
Avg no. of races by users: 1.0

The Community

Sure, in the end someone will take the first prize, the trophy and all the bragging rights. And sure, I might sound a little bit as if what I am trying to say is because I’m too bad at racing to actually compete. But to me the best thing about AWS DeepRacer is the community. We get to celebrate each other’s accomplishments, share learnings, support the new curious minds. We try really hard to support those who could not be there in person and enjoy every second of being together and share the excitement of an amazing show put on hand in hand by the AWS DeepRacer Pit Crew and the racers.

And this is not all. I bring to you the on last show by the top four racers:

Summary

It’s amazing what we have achieved since 2018. We’ve just had our 45,000th community member, the racing keeps getting better, more and more new approaches to competing and learning keep popping up.

Now it’s time to give the proven and cherished EC2s and home computers a well deserver rest. As we approach the end of year, new question to answer is: who will join Youssef (and Daryl Jezierski, more on that in another post) in the fight for the 2024 AWS DeepRacer League title?

Will 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.