AWS Deepracer League 2022 Update #2 March Finale Edition

And just like that, we’ve got new finalists unearthed. Find out how the racers did in the finale.

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

AWS Deepracer News

Invitation: AWS Summits in April and May

We’re racing in person! These are the upcoming summits of interest:

Be sure to visit AWS Summits Page and register for one (or more) to compete and qualify into the finals! Some of the Summits might be announced closer to the date so it’s worth checking the AWS page for the most up-to-date information.

April races are taking place

Visit the AWS Console to join the races in April. We’ve got two great tracks to tackle. Ace Speedway and Ace Super Speedway have been named after the AWS DeepRacer League 2021 runner up JPMC-Ace-Hyderabad. The Open Division track has got a great straight, some quick curves and tricky sharper turns and Pro Division track has taken it to the next level with the same straight but also many mind blowing hairpins and chicanes.

Ace Speedwar and Ace Super Speedway (Source: AWS DeepRacer Console)

AWS Summit Paris – Results

On the 12th of April 84 racers competed to select two participants of the EMEA Summit Online finale: gta and Paul. See the leaderboard

AWS Summit San Francisco – Results

21st of April around 122 racers competed to select the first two participants of the Americas Summit Online finale: Provectus and All Gas, No Brakes. See the leaderboard

AWS Summit San Francisco 20.04.2022 Standings

AWS Summit London – Results

On Wednesday the 27th of April 33 racers competed for the two spots in the regional summit qualifier finale for the EMEA region. The Summit was buzzing with attendants and despite a low number of racers it wasn’t low on emotions. We’ve had the defender of the London Summit title Matt Camp, the community’s Ernesto, JPMCLondon and Breadcentric, and many new racers. After a turbulent fight Matt has defended the title and was just a fraction of a second shy of a new track record, Breadcentric ended second and JPMCLondon was third.

AWS Summit finalists

Racers listed below will be taking part in AWS Summit Regional Online Finales to determine the lucky ones joining us in Las Vegas at AWS re:Invent to fight for the championship. Note: we are trying hard to keep this list up to date and correct but there may be errors here or some facts might change. If a racer has qualified into the finale but then qualified into the finals through a Virtual League then their spot will go to the next person in the Summit.

DateRacerTimeSummitRegion
31st March 2022DBro9.728AWS Summit Brussels (mention)EMEA
31st March 2022Chris13.054AWS Summit Brussels (mention)EMEA
12th April 2022gta13.149AWS Summit ParisEMEA
12th April 2022Paul14.968AWS Summit ParisEMEA
20th-21st April 2022Provectus12.229AWS Summit San FranciscoAmericas
20th-21st April 2022All Gas, No Brakes12.571AWS Summit San FranciscoAmericas
27th April 2022MattCamp9.969AWS Summit LondonEMEA
27th April 2022Breadcentric11.136AWS Summit LondonEMEA
AWS Summit finalists

Watch the stream

Race format

Rogue Raceway (Source: AWS DeepRacer Console)

Race Info:

  • 76.76 meters long
  • 107 centimetres wide
  • Object Avoidance format
  • Racers have two rounds, 180 seconds each to complete the fastest lap – all laps are continuous from standing start
  • Each collision results in a three seconds penalty
  • Each driving off the track results in a three seconds penalty
  • Top 3 qualify into the championships
  • 10 best racers who haven’t yet done so win an AWS DeepRacer EVO (it’s the first month of racing, so simply top 10 win an Evo)

First Round

The race was opened by RobinCastro-DBS presenting a pretty smooth racing but not enough knowledge in the model to deliver a clean lap. After three minutes and nine resets altogether the best of his laps was at 31.672 seconds with a single reset.

Deepak-dpk followed starting pretty fast out of the blocks. His model did much better giving a clean first lap at 28.994 seconds followed by another at 27.994 (yes, exactly one second less) before registering the first off track, possibly due to an overly eager speed setting. With more caution in fourth lap Deepak improved clocking 27.933 seconds before finishing in style, by smashing into a box as the timer reached zero seconds.

JPMC-DriftKing-Houston started by attempting to hit every single box on track in the first lap only narrowly missing the third one but making up by hitting the fourth one 13 times and adding an off-track right after. He missed the last box and completed the first lap at 1:39.852. Second lap wasn’t clean either but improved significantly completing with a time of 1:08.407.

While ZoD is a new name, you may remember SorinB – that’s him. His model was pretty eager and solid, securing him a lead after his run with a clean lap at 27.933 seconds. Laps two three and four had some resets which prevented time improvements and the last lap, while clean, was narrowly slower than the first one.

AIDeepRacer started fast but suffered from some off-tracks completing the first lap at 39.608. In second and third lap there was only one off-track in each giving him the times of 32.533 and then 32.268. Interestingly, there were no collisions with boxes in the whole run.

Karl-NAB arrived with another solid model. First lap was clean and at 26.663 seconds, followed by a narrowly slower lap and one with some resets. Throughout the run Karl kept pushing his model resulting in some resets but managed to complete the round with 26.472 seconds, strengthening his lead.

PolishThunder opened up with a few resets and a time of 35.458 seconds. His model didn’t look particularly bad but struggled with object avoidance. This is not uncommon with his models as he frequently relies on risky models and many submissions in the monthly races but this doesn’t work well in the finale where you have a very much finite pool of attempts. Daryl did however manage to get a single reset lap at 31.938 seconds, enough for fifth place at the time of racing.

Duckworth also relied on his trademark strategy putting stability above everything, allowing him to complete the first lap at 27.243 seconds, the second one at 26.598 seconds, third: 25.802 seconds, fourth: around 36.5 seconds, fifth: around 25.85 seconds before registering his first reset in the sixth, completed at above 30 seconds. As I said, stability first, speed as a side effect.

TonyJ had a great start with a clean first lap and 27.974 seconds time, reduced to 27.400 seconds in second, 26.741 in third, slightly more in fourth, 26.542 in fifth before gaining a few resets in sixth, completed at around 39 seconds. This puts Tony in third spot.

Ensono-Ernesto started with 4 resets and 45.306 seconds time. He improved significantly with a single reset and a time of 33.939 seconds before getting stuck before the third box for numerous resets and being unable to complete the next lap.

DBro took a break from his AWS Summit tour to join us for the finale and started strong at 26.592 seconds, followed by 26.065 seconds, then a little bit above 26.2 seconds, and a slower lap with a reset in fourth, again, followed by a clean 25.801 seconds and a narrowly slower sixth lap, giving him a lead at the time.

Rober started clean but slow at 28.603 seconds, going even slower in his second lap and registering a reset in third and a slow lap in fourth. His model didn’t seem particularly bad but did struggle from a few instabilities, an overly aggressive box avoidance and possible not enough ability to handle the speed.

AJM-Model-1 started with a clean but also slow lap at 30.199 seconds. He did manage to go down to 29.666 seconds which wasn’t enough for the podium.

Rosscomp1 began pretty aggressively at 27.407 seconds, then reduced to 26.466 in second and 25.535 in his third lap, securing him the lead after this run. All his laps were clear but most of them did not deliver enough speed to further improve. Still, an excellent run.

Flatearth started with 26.395 seconds and then went down to 25.466seconds in second lap, 25.071 in third before the first two resets in fourth. His model continuing at a ridiculous speed and remained at 25.071 seconds, securing a solid first spot.

JJ had an off-track in the first lap opening up at 30.034 but quickly improving to 25.734 in second and 25.661 in third lap. Fourth lap was looking promising but ended with an off-track, fifth wasn’t clean either. After a few more laps with resets James secured a third spot in the first round.

Let’s see what the standings looked like after round one:

PlaceRacerRound 1
1flatearth25.071
2Rosscomp125.535
3JJ25.661
4DBro25.801
5Duckworth25.802
6Karl-NAB26.472
7TonyJ26.542
8ZoD27.933
9Deepak-dpk27.933
10Rober28.603
11AJM-Model-129.666
12RobinCastro-DBS31.672
13PolishThunder31.938
14AIDeepRacer32.268
15Ensono-Ernesto33.939
16JPMC-DriftKing-Houston01:08.41

Second Round

Racers take turns in reverse standing order in second round.

JPMC-DriftKing-Houston has not managed to resolve the issues that haunted him in the first round. After 37 resets he had to accept the defeat from box four and completed the finale in 16th place.

Ensono-Ernesto started much better with 32.543 seconds after one reset, followed by slightly slower laps over 36, 38 and 35 seconds. He didn’t manage to have a clean lap this month but I’m sure it’s not his last finale this year.

AIDeepRacer started with two resets and a time of 36.74 seconds, then around 40 seconds followed by a clean lap and a time of 30.128 seconds and one with a reset but under 32 seconds, so at a faster speed. Again, AIDeepRacer had more off-tracks than box collisions.

PolishThunder opened up with a promising run but got denied by the fourth box. Sadly, that was the second run with no laps in this round.

RobinCastro-DBS manged to complete the first lap at 41.597 seconds and a second one just above that. Unfortunately boxes proved to be too challenging for his model and despite a few completed laps, his end time for this run remained unchanged.

AJM-Model-1 opened with a time of 31.327 seconds in one of very few clean laps so far. Then he almost got one under 29 seconds but lost grip and fell of track just before the finish line. After that many resets followed and none of the laps gave him a better time.

Rober started with a lap of 32.724 and one reset followed by a very nice clean lap at 27.936 seconds. He did not have many resets but there wasn’t a lot of smooth riding and the car seemed to keep skidding.

Deepak-dpk started with a clean lap at 28.777 seconds followed by a few laps with resets and then another clean one that managed to get down to 28.534 seconds. Pretty solid but a little slow run from Deepak.

TonyJ opened with a slightly slow but solid 27.726 second and then improved with 27.262 and 26.679 seconds. He seemed to have plateaued repeating a few laps at a similar speed and completed with zero resets.

ZoD started with many resets and the first lap at 31.132 seconds. The car did not improve throughout the run and that was the final time.

Karl-NAB had a clean lap at 27.11 seconds followed by a single reset one around 30 seconds. Then again, clean one with 26.329 seconds time and another just over that, and then a reset and around 30 seconds, and again, a clean one and just over his best time. He managed to improve his overall time but remained in the same position.

Duckworth also had a clean first lap at 27.369 seconds followed by 26.133 and 26.001 seconds. Then a few minimally slower laps and that was the end of a very solid run, sadly without improvement.

DBro was the last one to fight for the podium and clocked 26.879, then 26.194 seconds and had a few resets, including one just before the finish line in a podium winning lap. More resets have denied Darren’s chances and sealed the March winners.

JJ did not have to improve but the last time I have seen James not push for an even better lap he deeply regretted not doing so as he lost a chance for a 2019 championship spot. Ever since he’s been trying to squeeze as much out of his models as he could. His opening time of 25.991 seconds made it clear I was right. In second lap he was on his way to get ahead of at least rosscomp1 only to lose grip just in front of finish line. I must say, the line placement is simply genius and has been punishing everyone who tried to push their models even the slightest bit too far. Third lap (25.734 s) got him into the second spot, fourth (25.134 s) set the scene for the fight for the first place but it was a tiny bit slower this time and was the last completed lap in this brilliant run.

Rosscomp1 wasn’t going to give up without a fight. His times were 36.370 seconds, 26.268 seconds, 26.200 seconds, reset (around 32 seconds), around 27 seconds, and around 26.3 s. Even though the model looked great, he had to settle for the third spot, still giving him the ticket to Vegas.

Flatearth had his victory round and started with 26.132 seconds, then 25.601 seconds and a few resets. I think he tried to make his time even better but the more he pushed the more resets he had. Still, he’s definitely going to Vegas. This will be super interesting as I have never seen him race on a physical device.

Let’s see the final standings:

PlaceRacerRound 1Round 2Result
1flatearth25.07125.60125.071
2JJ25.66125.13425.134
3Rosscomp125.53526.225.535
4DBro25.80126.19425.801
5Duckworth25.80226.00125.802
6Karl-NAB26.47226.32926.329
7TonyJ26.54226.67926.542
8Deepak-dpk27.93328.53427.933
9ZoD27.93331.13227.933
10Rober28.60327.93627.936
11AJM-Model-129.66631.32729.666
12AIDeepRacer32.26830.12830.128
13RobinCastro-DBS31.67241.59731.672
14PolishThunder31.938DNF31.938
15Ensono-Ernesto33.93932.54332.543
16JPMC-DriftKing-Houston01:08.41DNF01:08.41

Overall, this was a very interesting finale. Note the difference in results for the top five: 0.731 second difference means that even the slightest mistake can get in the way of qualifying.

Some of the racers are still warming up and it’s totally fine – after all you need a single fast lap in a single finale and the formats differ wildly so it’s inevitable that things can go not as planned. One race you might not complete a lap, another you’ll punch a ticket to Las Vegas.

AWS DeepRacer League Championship 2022 participants

Here’s who we already know will come to Las Vegas:

Date determinedRacer nameQualified through
1st Dec 2021JPMC-Rogue-HyderabadAWS DeepRacer League 2021 Champion
2nd Dec 2021Eric MorissonAWS DeepRacer re:Invent 2021 Live Stream Champion
11th April 2022flatearthAWS DeepRacer Virtual League March
11th April 2022JJAWS DeepRacer Virtual League March
11th April 2022Rosscomp1AWS DeepRacer Virtual League March
9th May 2022?AWS DeepRacer Virtual League April
9th May 2022?AWS DeepRacer Virtual League April
9th May 2022?AWS DeepRacer Virtual League April
13th June 2022?AWS DeepRacer Virtual League May
13th June 2022?AWS DeepRacer Virtual League May
13th June 2022?AWS DeepRacer Virtual League May
11th July 2022?AWS DeepRacer Virtual League June
11th July 2022?AWS DeepRacer Virtual League June
11th July 2022?AWS DeepRacer Virtual League June
8th August 2022?AWS DeepRacer Virtual League July
8th August 2022?AWS DeepRacer Virtual League July
8th August 2022?AWS DeepRacer Virtual League July
12th September 2022?AWS DeepRacer Virtual League August
12th September 2022?AWS DeepRacer Virtual League August
12th September 2022?AWS DeepRacer Virtual League August
10th October 2022?AWS DeepRacer Virtual League September
10th October 2022?AWS DeepRacer Virtual League September
10th October 2022?AWS DeepRacer Virtual League September
14th November 2022?AWS DeepRacer Virtual League October
14th November 2022?AWS DeepRacer Virtual League October
14th November 2022?AWS DeepRacer Virtual League October
?th October 2022?AWS DeepRacer Summit League APAC
?th October 2022?AWS DeepRacer Summit League APAC
?th October 2022?AWS DeepRacer Summit League APAC
?th October 2022?AWS DeepRacer Summit League APAC
?th October 2022?AWS DeepRacer Summit League APAC
?th October 2022?AWS DeepRacer Summit League EMEA
?th October 2022?AWS DeepRacer Summit League EMEA
?th October 2022?AWS DeepRacer Summit League EMEA
?th October 2022?AWS DeepRacer Summit League EMEA
?th October 2022?AWS DeepRacer Summit League EMEA
?th October 2022?AWS DeepRacer Summit League Americas
?th October 2022?AWS DeepRacer Summit League Americas
?th October 2022?AWS DeepRacer Summit League Americas
?th October 2022?AWS DeepRacer Summit League Americas
?th October 2022?AWS DeepRacer Summit League Americas
?th October 2022?AWS DeepRacer Student League
?th October 2022?AWS DeepRacer Student League
?th October 2022?AWS DeepRacer Student League
?th October 2022?AWS DeepRacer Enterprise Wildcard Finalist
?th October 2022?AWS DeepRacer Enterprise Wildcard Finalist
?th October 2022?AWS DeepRacer Enterprise Wildcard Finalist
?th October 2022?AWS DeepRacer Enterprise Wildcard Finalist
?th October 2022?AWS DeepRacer Enterprise Wildcard Finalist
?th October 2022?AWS DeepRacer Enterprise Wildcard Finalist
AWS DeepRacer League 2022 Championship participants

That’s all from the March finale. Remember to visit back for more updates on AWS DeepRacer and the league!

Announcing: AWS DeepRacer League 2022 on-line and in person

Join thousands of contestants on the learning path and race for the knowledge, prizes, glory and friends in the exciting Reinforcement Learning powered tournament.

I’m Tomasz Ptak, a senior software engineer at Duco, an AWS Machine Learning Hero, a DeepRacer racer (named Breadcentric), a hobbyist baker and a leader of the AWS Machine Learning Community on Slack where we learn and race, and help each other start and grow our adventures in the cloud. Let me introduce this year’s AWS DeepRacer League season.

Continue reading “Announcing: AWS DeepRacer League 2022 on-line and in person”

AWS DeepRacer League Update #28 September Finale Special

This is one of the two remaining finale races this season. We’ve got a few rookies joining the race and challenge some of the regulars looking to secure their spots in the championships.

We are awaiting details about what to expect during the championships (apart from it being all-virtual again). It’s a terrible pity that we won’t be able to meet. DeepRacer draws people together, sparks collective curiosity, it’s not surprising that we have had so many racers over the last years and even less surprising that we’ve had so many of them join the Community.

In the meantime let’s have a look at how the racers did in the September finale.

Continue reading “AWS DeepRacer League Update #28 September Finale Special”