AWS DeepRacer League Update #24 July Finale Special

We’ve already completed four out of eight racing months, time to start completing the remaining twelve championships racers. First chance for this is July Finale.

As we’re nearing the re:Invent and AWS DeepRacer championships we’re still waiting to see if all the finalists will be able to attend the event. Traveling into the USA from some areas is still restricted due to Covid regulations which might make meeting in person difficult, that said we’re still hoping that it will happen. While we wait for the good news in this area, let’s recap the July Finale race.

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.

DeepRacer news

August races

Are you ready to hit the Hot Rod Speedway? Open division is facing 42.25m of fun with great straights and exciting turns, Pro Division is experiencing 59.69m a total madness where nothing is straight and simple for too long, yet under 60 seconds total time might be achieved if you do it just right!

Hot Rod Super Speedway Pro divison track (Source: AWS DeepRacer Console)

Head over to the AWS DeepRacer Console to burn some virtual rubber!

The race format

Baja Highway track (Source: AWS DeepRacer Console)

Race info:

  • 64.67 meters long
  • 107 centimeters wide
  • Object avoidance format with four boxes on the track
  • Racers are trying to complete as many laps as possible, the fastest lap counts
  • Each collision and each driving off the track results in a three seconds penalty
  • Top 3 qualify into the championships during the AWS re:Invent in Las Vegas
  • Top 10 (who haven’t won it just yet) win an AWS DeepRacer EVOs – so seven this month, the rest overflow into lower spots
  • The track is divided into three sectors and in the stream they have colours assigned: green means the fastest time for a given racer so far, purple is the fastest time across all racers and yellow means slower than the fastest time for that sector for a given racer

Stream recording

You can watch the whole race recording with a commentary and interviews by Blaine Sundrud and Ryan Myrehn:

Round one

AIDeepRacer started with a lap time of 33.334 after two resets. While there is some serious speed in this model’s run, resets were getting in the way of performing better. With a little bit of luck a tiny improvement allowed to get down to 33.065.

CGS-LargeLattice has started with a lot of resets making the the first lap in 01:07.921. The car was struggling especially in the first sector but managed to reduce time to 55.384 seconds and this was the last lap in this round.

Matt got stuck on box one just like the previous cars but struggled with over 20 resets making the lap one with a time of 02:06.448. In second lap it got past that box nicely and made the best time so far in the first sector and after two resets completed a lap in 33.872 seconds making a good prediction for round two.

RobinCastro-DBS almost started his run with a great run past box one on the inside, sadly touched on the box and started the reset struggle similar to Matt’s but with 18 resets before getting past it. Robin made a few tries modifying the speed to get a better behaviour before getting a few more resets and completing the first lap in 02:00.699. Unfortunately it faced another series of resets on box one finishing the round in the last spot.

dartjason began with an unsurprising reset on box one which was placed on the optimal line for the fastest laps. The car has tried to get past on the inside and on the outside but after 39 resets could not complete a single lap.

SorinB returned this month for another attempt towards the top three spots. The car flew beautifully past boxes one and two to hit box three but after that completed a fastest third sector so far and completed the lap in 29.744 seconds to jump into the first spot. Second lap had more resets but in third he managed to improve with a time of 27.269 seconds. Fourth lap completed at around 40 seconds and fifth at 41 seconds making it a very promising result and with a chance for improvement in next round.

GT-DevelopersIO has been fourth once in the finale before missing the ticket to Vegas by a fraction of a second. Sector one was flawless and so was the third, with one reset in second the lap completed at 31.278 seconds, good enough for the second place for now. Two resets and over 33 seconds in second lap. In third lap two resets again and too slow to improve, same as the laps to follow. Still, it’s a solid second spot for now.

TonyJ started with only two resets on box one. His model seems to be cutting the corners nicely but it may have become a curse on box two. The reset has placed the car on the inside of a sharp turn there putting the car in an unfortunate spot to restart and causing 16 more resets putting the lap at 01:38.474. In second lap the car had very similar issues that has lead to a lap of 01:06.960. If only the stability could go up Tony could have some serious chances to punch the ticket to Vegas in round two.

AJM-Model-1 joined the box one club, getting past after eight resets. One more reset and a lap completed at 01:05.420, right in front of Tony and then went down to 47.395 seconds after a few more resets. One more lap completed at 61 seconds leaving room for improvement for the second round

JPMC-Rogue-Hyderabad also started with a few resets but with the fastest second sector so far completed the lap in 36.913 seconds. Another lap opened up with the fastest first sector and a slight improvement of 36.691 seconds. There was a lot of speed in this model but resets were getting in the way. Still, the third lap completed in 36.276 seconds. Last lap this round saw many resets and no improvement, putting Rogue in the fifth spot for now.

JPMC-Ace-Hyderabad got past box one in two resets but struggled with more resets further on completing the lap in 56.099 seconds. Second lap started cleanly and completed so clocking 26.170 seconds and marking the first lap with zero resets. Ace may have set the speed to higher after that lap bringing many resets but it is one lap that it takes to qualify and one lap he got.

Yi-LI-NYCU-CGI has completed the first lap in 28.171 with a reset and then made a second lap with zero resets and the time of 23.630. Lap three had not only zero resets but also fastest times so far in sectors two and three. A few more resets happened in the laps to follow but it’s still been so fast! This time put Yi-LI in the first spot and moved SorinB to third.

rosscomp1 started cleanly with an impressive lap of 25.487 only to get stack on box one for a few resets, then came another insane lap with 23.268 which still keeps him in second spot but ups the game significantly. Next lap followed with more resets and Ross has landed in a solid spot two for the next round.

Ernesto has been in every single finale so far. First clean lap of 25.317 has put him right on the podium. In lap two he improved further going down to 24.658 before the first reset in lap three and a few more in four. In lap five there has been a clean lap and a slight improvement of 24.329 and a solid third was a good prognosis for the second round.

ZhengYi-NYCU-CGI started pretty swiftly in with only three resets before completing a lap in 35.300 seconds. Four more resets in lap two and no improvement. Then in lap three ZhengYi improved in sectors one and two and completed in 22.939 taking the lead and then improving to 22.686 right after. A slightly slower lap right after that and the one after (by slower I mean still good enough for the first spot) completed the round one for ZhengYi and this completes the first round.

Let’s have a look at the standings after this round:

PlaceRacerTime
1ZhengYi-NYCU-CGI00:22.686
2Yi-LI-NYCU-CGI00:23.108
3rosscomp100:23.268
4Ernesto00:24.329
5JPMC-Ace-Hyderabad00:26.170
6SorinB00:27.269
7GT-DevelopersIO00:31.278
8AIDeepRacer00:33.065
9Matt00:33.872
10JPMC-Rogue-Hyderabad00:36.276
11AJM-Model-100:47.395
12TonyJ01:06.960
13CGS-LargeLattice01:07.921
14RobinCastro-DBS02:00.699
15dartjasonDNF
16DuckworthDNS
AWS DeepRacer League 2021 July Finale standings after round one

Round two

dartjason return to the track with a similar struggle on box one and finished the round on it without a lap time.

RobinCastro-DBS has returned to try and improve his time. Beginning was flaky around box one followed by getting stuck on box two. Robin completed round two with 42 resets and zero complete laps.

TonyJ flew past box one and got a few resets later on and completed in 32.783. Second lap was clean and flying with a time of flat 25.000 seconds. Great time but not good enough for the podium. Further laps got slowed down with resets and Tony landed in fifth spot for now.

CGS-LargeLattice returned to the track for the second run but this first sector has proven tricky for this model, so has the second one with the car running into box three multiple times. This was a bit awkward to watch the car turn slightly to hit a box on a straight section twelve times. Still, got past it and clocked a lap of 01:37 seconds and improved greatly afterwards with 31.807 seconds before hitting that box three a few times more. Great improvement!

AJM-Model-1 took to the track for a second round to struggle again on box one before getting past it with a neat 360, sadly no style points. First lap completed in 01:02.703 and then improved to 29.703 with both time and potential to improve. Next lap did not improve at around 33 seconds but then box one denied further improvement.

JPMC-Rogue-Hyderabad had a reset but still managed to improve in lap one with 28.239 seconds. In second lap and third laps resets slowed things down significantly. Lap four started promising but then lead to more resets giving just enough time for a flying lap and this got denied by another reset. This completed round two for Rogue.

Matt has hit two resets in the first lap in sector one and completed in 35.319 seconds. in second lap he had three more resets and a time of 37 seconds. Even more resets in lap four and over 50 seconds. Again a few resets in the last lap have denied his chances to improve.

AIDeepRacer has managed to shave off a bit with a first lap of 32.816. No improvement in second and resets keep slowing the car down. Lap four brought hope with a single reset in sector two and a time of 28.801 but more resets voided next attempts. Still, there was a time improvement and there will be more chances next month.

GT-DevelopersIO started in a promising single-reset-still-improving time of 30.632 seconds. Then a clean lap at 26.390 seconds brought great improvement. A slower (yet still clean) lap followed before lap four which was just a little bit slower than that. I think GT then pushed the speed dial a bit to far causing a bit of instability in the laps to follow.

SorinB returned to reclaim his good spot but the first lap at 33.857 seconds has not made it. Sorin has started pushing the speed slightly and managed to get down to 31.992 but started losing grip on the road. Still, he managed to get down to 29.074 seconds. Then he managed to get down to 27.525 seconds. still slower than round one but a good time nonetheless. Another reset voided the last attempt to improve in this finale.

JPMC-Ace-Hyderabad started aggressively but not too stable with a warm-up lap of 38.315 seconds. Sadly looks like the next lap consumed most of the time and got followed by more resets which ends his run this month.

Ernesto started with almost getting past box one four times. Started slow with a time of 47.370 seconds and then improved with 38.723 right after. In lap three he got past box one cleanly but had a reset in sector three to clock 28.081 seconds. In fourth lap he did a fraction of a second slower and then had another reset in fifth to complete the round without an improvement and punching the tickets for a three on the podium after round one.

rosscomp1 started with a few resets and a time of 36.237 seconds. More resets in lap two and a time of 33.044 seconds, then 32.872 seconds in lap three. Ross has put his bet on speed in round two and that has not paid off. Luckily it did not matter for him as he will hold on to his third spot.

Yi-LI started with a lot of resets in lap one which lead to time of 50.075. Even more resets in lap two have caused a time of over one minute and an off-track in lap three but still and improvement came at 26.730 seconds. He has managed to finish this round with a clean lap and a time of 23.196 seconds, just behind his round one time.

ZhengYi has returned for a victory round and he definitely decided to take hist time with the best lap at 30.740 seconds and 16 resets. No need to race super fast anymore. Congratulations for the win!

Let’s see the final standings:

PlaceRacerTime 1Time 2Final
1ZhengYi-NYCU-CGI00:22.68600:30.74000:22.686
2Yi-LI-NYCU-CGI00:23.10800:23.19600:23.108
3rosscomp100:23.26800:32.87200:23.268
4Ernesto00:24.32900:28.08100:24.329
5TonyJ01:06.96000:25.00000:25.000
6JPMC-Ace-Hyderabad00:26.17000:38.31500:26.170
7GT-DevelopersIO00:31.27800:26.39000:26.390
8SorinB00:27.26900:27.52500:27.525
9JPMC-Rogue-Hyderabad00:36.27600:28.23900:28.239
10AIDeepRacer00:33.06500:28.80100:28.801
11AJM-Model-100:47.39500:29.70300:29.703
12CGS-LargeLattice01:07.92100:31.80700:31.807
13Matt00:33.87200:35.31900:33.872
14RobinCastro-DBS02:00.699DNF02:00.699
15dartjasonDNFDNFDNF
16DuckworthDNSDNSDNS
AWS DeepRacer League 2021 July Finale standings after round two

Congratulations to three new championship participants, marked above in bold. Let’s hope we get to meet each other at the AWS re:Invent 2021 in Las Vegas.

That’s all from the finale race for July. See you soon with the August updates.

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.