AWS DeepRacer League 2021 Update #12 April Finale

It’s this time of April race when the racers fight for their spots in the AWS DeepRacer League 2021 Championship places in Las Vegas! Read on to know how they did!

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.

AWS DeepRacer News

May races

Kuei Raceway (Source: AWS DeepRacer Console)

AWS DeepRacer League races for May are already happening. Be sure to join in the AWS DeepRacer Console.

Race info

Lars Circuit (Source: AWS DeepRacer Console)
  • 59.233 meters long
  • 107 centimetres wide
  • Race is happening LIVE as the stream is showing it
  • Racers get to control the throttle on their models – they can boost or reduce the car’s speed decisions by a multiplier, steering is inferred using a Reinforcement-Learned model
  • Object Avoidance format with four boxes in fixed locations
  • Racers have 3 minutes to clock a lap as fast as possible in two rounds
  • In second round racers can choose a different model
  • Each collision and each driving off the track results in a five seconds penalty
  • Top 3 go to Vegas, 10 fastest ones win AWS DeepRacer EVOs

Stream recording

Round One

Penguin-DNPds started with a pretty nice drive by a box on the outside but 5 resets mean the first lap is 53.304 seconds. It’s a lap, but not fast enough, I’m afraid. His car has struggled with stability, in lap two after 10 resets a lap of over 1:20 meant end of hopes as it was not enough time left for another lap.

AIDeepRacer is a Rookie. Nathan, a junior high school student, qualified into the Pro Division in March, and straight into the finale in April. His first contact with Object Avoidance was really rough and he hasn’t managed to complete a lap but he deserves a great round of applause. I’m sure he will come back in the next finales.

Maikel started with a nice avoidance of the first box but still got six resets and a first lap of 59.021. His model had a potential to still improve. Second lap with four resets ended at 45.773 seconds and ahead of Penguin-DNPds. Lap three with four resets again and at 47 seconds not fast enough to improve. Last lap with a single reset ended just a few meters before the finish line when he ran out of time. He’ll need to repeat that in round two

Mungi started bold but faced a sudden stop at box one, for fifteen resets. Once past that he made a clean lap but a lap o 1:56.150 is not fast enough. In lap two, after 11 resets on box one, time has run out.

Rober from Spain started nicely and got past box one, two, three, an off-track next to box four and a decent 31.340 seconds on lap one put him ahead. Second lap very solid again with two resets and around 36 seconds, but still plenty time. Lap three again with a reset around box four and a time of 30.813, still lots of time to improve. Lap four – another reset and in the area of 31 seconds. After that it seems that Rober has pushed the throttle and the car went a little crazy.

RobinCastro-DBS was another one denied by box one, followed by a 360. Good style, but no cigar. It might be that object avoidance was not very effectively trained. after 20 resets he clocked 2:17.856 and sadly not enough time to improve

GT-DevelopersIO started with a spectacular 34.189 despite two resets. Second lap with one off-track and 28.536. Third lap with zero resets and insane 22.196. lap for has seen improvements in second and third sector but just a bit slower than the best lap so far. Next lap has seen a few resets. The cars seem to lock into a reset limbo around box four. With a few resets next laps were slower but 22.196 is a serious candidate for the spot in the finals.

Jochem from Netherlands started nicely around box one and two, and three but got a reset next to box four. First lap of 29.119. In second there has been a reset on boxes one and four and with 34 seconds started lap three with a fastest sector one and a reset due to speed after last box and a time over 30 seconds. Then a few more resets on box one. In last lap of this run two resets on the first box have denied chances of improving. 29.119 was current second fastest at the time.

AJM-Model-1 struggled with resets on just about all boxes and sadly did not finish a lap.

Duckworth (also known as Lars in the community) raced his circuit. In lap one after two resets he clocked 40.527 which was decent to start with. Sadly in lap two box one just stood there like Gandalf the Grey: You shall not pass! After eighteen resets Lars hasn’t managed to complete lap two.

PolishThunder just flew past box one as if it was his acrobatics plane. Lap one, resets zero, time 24.038. That’s a second time and in lap two improved to 22.489, first reset in lap three around 29 seconds and a second reset in lap four on box one. Box of Despair? Single resets denied any more improvements but 22.489 definitely made GT-DevelopersIO sweat.

Ernesto from the British community started with a gorgeous run past boxes one and two but for a difference hit box three. Lap one at 31.578, not bad at all. In lap two he started struggling on box four and above 41 seconds he moved into lap four. Box four is placed in a way that makes many cars falter. But then a clean lap and 24.879 put him in top three.

JPMC-DriftKing started with a nice pace and with one reset clocked a 28.829. Many more resets in lap two that was over 50 seconds. In lap three DriftKing drifted into off track. It is a fast model that he ran in round one but with many resets it just couldn’t show its full potential.

Dartjason decided to top the statistics in “most hits on box two” category. Sadly box two has defeated him, no laps finished.

Flatearth started with two resets and opened with a 33.084 lap which wasn’t very bad. Two resets on lap two and 32.917. Lap three had two resets again and over 34 seconds time. Lap four with one reset gave a time of 29.369. Then another single reset and 27.820 put him in fourth for now.

RogerRabbit started cleanly with spectacular 23.149, straight on the podium. Sadly in lap two two resets have held him behind with over 32 seconds. Lap three with two more resets was followed by another clean lap, just at 26 seconds, not fast enough. Then a beautiful lap about to take the lead got denied with a single reset, followed by another one with no resets this time and 20.915 seconds straight into the lead.

Let’s see the standings:

AWS DeepRacer League 2021 April Finale standings after round one

Round two

AIDeepRacer has managed to slide past box one and after many more resets managed to clock 1:28.270, first complete lap of the finale, and not a last one, I’m sure.

Dartjason has also improved significantly with only two resets and a time of 39.120. Lap two was also much more stable, I think he switched models. With 33.241 and one reset he moved in the area of 9th place for now. Towards the end he started clocking more resets, sadly but 33.241 was not time to be ashamed of.

AJM-Model-1, PolishThunder’s colleague started with two resets but then managed to progress until box four which denied his chances to complete a lap. DNF for AJM-Model-1.

RobinCastro-DBS managed to get a relatively fast run in section one and two but then in section three he got stuck on box four again. After six resets he finally managed to clock a time of 1:13.938. Not great, but significant improvement already. In lap two he ran into box four six times (again) and with 1:30 he finished his participation in April finale.

Mungi’s opening lap was flawless with a time of 26.115. Just wow! Second lap was slightly slower but in third zero resets again and 25.002 seconds. Fourth lap was at around 25.5 seconds. Fifth lap was at around 26 seconds and then first reset happened in lap six, just spectacular! This model is behaving beautifully but just not fast enough for Vegas. Still, really good result.

Penguin-DNPds is one of few racers who trained their model in the AWS DeepRacer console (join community to learn how to use a local training setup on EC2). He punched box one five times and after a few more resets clocked 2:21.801. In lap two he clocked 50.196 s which let him improve slightly compared to Round one. Further resets have taken hopes for another lap again. Really nice result and see you in the future races.

Maikel started really fast, on the edge of grip but improved with time of 35.701. Sadly resets have denied further improvements.

Rober with two resets opened his run with 36.866. Second lap had one reset and with 32.758 still wasn’t enough to improve overall. But then in lap three Rober clocked a clean 25.487 which put him in sixth then. After a bit of struggle on box four Rober has run out of time for faults and a reset ended his last attempt.

Jochem was going pretty stable but just not fast enough to be a threat to the podium. He managed to go down to 28.294 which still was too slow. He managed to go down to 24.977, enough for fifth but an off track in the next lap completed his round.

JPMC-DriftKing started with a a significantly faster speed than Jochem and managed to clock 23.970, enough for the fourth. Resets in laps two and three pretty much voided those laps. Only clean laps go to Vegas. Sadly he got stuck on box four. Interestingly in round two there have been significantly fewer problems with box one.

Flatearth returned with a quick reset on boxes one and four and opened up with 32.990. In lap two he improved in section one, then a reset but still managed to get 26.672 which is an improvement. Lap three had a fastest time in sector two (purple colour) but a reset denied him improvement. In lap four he did a spectacular clean run with 21.673 which put him in second. Incredible time. Further improvements lost to resets and flatearth landed in second to wait for the four remaining runs.

Ernesto started sixth. In the first lap after a few resets he made a 47.180 lap and then improved to 29.510, still not as good as round one. Another reset or four in lap three left him with a 45 seconds lap. Slight improvement to 29.204 wasn’t enough still and Ernesto completed his run as the first person that hasn’t improved the time in round two.

Polish Thunder fell right out of top three before his run. He had to beat GT-DevelopersIO to still have a chance or flatearth to be right in. First lap started pretty solid with zero resets and 23.934. Then resets in laps two and three caused issues. His chances were getting pretty slim but then he got the fasted sector one of the whole race and personal bests in two and three got him 21.828s, just ahead of GT-DevelopersIO

And speaking of the devil, it’s GT-DevelopersIO’s chance to bite back and get back into a place with tickets to Las Vegas. First lap was looking promising but a reset towards the end killed his chances. Lap two was clean but with 23.070 was too slow. Lap three clean again, 22.986 was not enough. Then a few resets have knocked him out of the rhythm. He still had plenty time, but not enough to waste on resets. And sadly these were popping up in every run putting PolishThunder on a plane to re:Invent and the finals.

With all places clarified it was time for a victory round from RogerRabbit, our winner. He managed to clock a 21.600 lap after a few worse ones. It was enough to win anyway even though not as fast as round one. But then he fixed it with a lap of 20.171. Next lap was slower around 20.6 seconds. That was quite a show. All the respect to RogerRabbit.

Let’s have a look at the final results:

PlaceRacerTime 1Time 2Final
Final standings of AWS DeepRacer League 2021 April Finale

Congratulations to RogerRabbit, flatearth and PolishThunder for winning their places in the Championship Cup, well done to all racers. See you in the next AWS DeepRacer League update!

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.