Sixteen top Pro racers, two rounds, three best lap times, championships in Las Vegas. Find all of these below.
While May has been gone for ten days already, some of the racers were still stuck in it. After a very intense racing we have made it into the Finale of the AWS DeepRacer League. It’s both scary and exciting, and anything can happen. With 15 seconds difference between the best and the worst in the group, if we had similar times in the finale, we would have around five seconds difference between the best and the worst. Add a bit of randomness and limited attempts and who knows what can happen!
DeepRacer News
AWS Machine Learning NanoDegree Scholarship Program at Udacity
Udacity and AWS invite you to participate in the scholarship program where you can be given access to a Machine Learning NanoDegree teaching various aspects of working with machine learning and deploying it in production. Registrations end 23rd of June. See the details.
June Races
While we’re looking at the last week of May racing, June edition has already begun.
Visit the AWS DeepRacer Console to join the fun!.
The race format
Race info:
- Track named after Kuei-NCTU-CGI from Taiwan who finished third in 2021 DeepRacer Championship
- 60.74 meters long
- 107 centimetres wide
- Object Avoidance format
- Racers are trying to complete as many laps as possible, best 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
Stream recording
Round one
First to race was JPMC-HandsFree-London. The first lap started with a quick off-track, and then another. Box three has become the main problem with 4 crashes and the first lap of 52.191. Second lap went a little smoother with 47.478 seconds. Third lap went slower than that and fourth lap has not completed. 47.478 seconds is not abysmal but then remember the roughly 25 second laps in the main race.
AJM-Model-1 opened up with a significantly more stable performance. One reset means the first lap is 30.392 seconds. Second one went even better. Zero resets and 26.453. Third lap ended in the area of 29 seconds after a reset. In fourth 25.595, just narrowly improved and then a few resets in the next one for a 34 seconds. It seems that the car is very cautious in the turns section with its slowdowns but then the offtracks and collisions happened mainly on straights, interestingly.
Rober started quite aggressively and into a reset on the chicane and then a box collision on the straight, still 33.193 is pretty impressive. In second lap single reset and a lap of 31.532 seconds. In third he managed to beat the sector one best time of the tournament so far and in sector three for an impressive zero resets 26.664s. The resets slowed him down in lap three but in four it started fastest again in sector one, and two, and three, but third wasn’t fast enough to improve. Last lap went overall slower and with a reset got denied completion.
Breadcentric started his debut in finales with a solid sector one, fastest sector two so far and fastest third which produced a time of 24.422. Second lap solid as well and 24.028s lap. Third lap again, top times in sector one and three and 23.644s lap. Third lap looked promising: fastest sectors one and two but right on the last straight the car lost its stability and the time did not improve. Lap four was generally slower at over 25 seconds and lap five just under 25 seconds. Lap six has seen the first reset. This is a leading time for now.
rosscomp1 started pretty smooth with 29.071, some resets in the second, third has seen best time for sector two and despite a reset he improved to 27.322 seconds. Resets are unforgiving even despite the reset time being reduced from five to three seconds. Next three laps denied by resets. There’s a potential in the model but boxes say no.
Matt representing Argentina has started nicely with zero resets which is enough for 27.078. Second lap sees time improvement in sectors one and three for 25.503 and then even better in lap three for 24.747. Seems like Matt has stepped on it in lap four with fastest sector one followed by a reset and a 27 seconds lap. Another reset in fifth and four more in sixth. The time places Matt in second spot for now
Penguin-DNPds has raced in a banana car with a pretty good start but then started struggling with the boxes on the straight. Especially box four. After ten crashes into it I have lost count, really. Terrible pity, the model seemed to have something in it. Finally the car went past the box and managed to clock a time of 2:34.302
dartjason started quite smooth but with a few resets had to settle on 41.118 seconds. Second lap was also plagued with resets but enough for an improvement to 34.662. Sector 1 improved even more but in second there has been a reset. Still, 31.169 s and time left for improvement, though too many resets to improve.
Duckworth is one of the finale veterans, having competed in three out of three. Started smooth as expected but hit box number three to clock 31.382 seconds. Three resets in lap two and no improvement. Then a clean lap and 24.596 but then a few resets. Another clean lap but the time of roughly 25 seconds and more resets. That’s enough for the second.
Yi-Li-NYCI-CGI with a pretty time but not enough speed – zero resets, 27.043s, good start. Resets have lead to a time of over 32 seconds in the second. Third started slower in sector one but faster in sector two and three which was enough for 26.004. Fourth, fifth and sixth were pretty flawless except for speed, no time improvement. Still, a very enjoyable performance. That’s currently enough for the fifth
Yoway-NYCU-CGI also presented a pretty stable model with a first lap at 25.781, then a reset in second which means no time improvements. Third lap started slower so despite no resets time was over 26 seconds. Yoway is showing how important both stability and speed are, despite zero resets two laps were too slow and with increased speed came resets. And then cam the single lap with the fastest sector one which was enough for the time of 24.672 s.
Ernesto started pretty fast just to get off track past the hairpin. Another reset in sector three and first lap completed at 32.979. Second lap was much smoother, zero resets and 25.670s. Third lap started with another improvement in sectors one and two and a reset on finishing straight so no improvement. Lap four was clean with great speeds and the finishing time of 23.730, just fraction of a second behind Breadcentric! Next laps were under 25 still but not fast enough.
ZhengYi-NYCU-CGI has started with a struggle at box two. That’s a pity to see after a smooth sector one but with 11 collisions the car has managed to get past it. Few more resets and the first lap was 1:25.086. Sector one again smooth and fastest so far in this run but we’re back at box two for a few crashes. It might have been that the car was not trained for avoiding obstacles but for some perfect line. Second lap of 53.649 and third lap ended, again, on box two.
Jochem started with a solid sector one, past box two, three and four and 23.998 in lap one, right for the podium! Lap two – a reset but still just over 26 seconds! And another reset in lap three and again 26 seconds! And two more in lap four but fifth has just gone down to 21.995. Another reset in sixth so that’s the one fast lap. Jochem has managed to do a seventh lap as well but slower. That’s a new leader!
Joffe has started really well with 23.200 just to get into second spot on a start, a reset in second, third and fourth lap. Still solid but resets deny it. Fifth lap has started solidly and sadly got an off track just before the finish line. Another reset in the sixth and with a single clean lap he jumped onto the podium.
JPMC-DriftKing-Houston was the fastest in the May race. He started with a quick sector one and resets in sector two for a lap of 28.728. That’s still fast but another reset in lap two and the time is 26.457. Lap three has been clean with 22.483 and second place! Fourth lap at 24 seconds and a reset in fifth. Still impressive. After a few resets the round one has completed
Let’s see the the standings:
| Place | Racer | Time |
| 1 | Jochem | 21.995 |
| 2 | JPMC-DriftKing-Houston | 22.483 |
| 3 | Joffe | 23.2 |
| 4 | Breadcentric | 23.664 |
| 5 | Ernesto | 23.73 |
| 6 | Duckworth | 24.596 |
| 7 | Yoway-NYCU-CGI | 24.672 |
| 8 | Matt | 24.747 |
| 9 | AJM-Model-1 | 25.595 |
| 10 | Yi-Li-NYCU-CGI | 26.004 |
| 11 | Rober | 26.664 |
| 12 | rosscomp1 | 27.322 |
| 13 | dartjason | 31.169 |
| 14 | JPMC-HandsFree-London | 47.478 |
| 15 | ZhengYi-NYCU-CGI | 53.649 |
| 16 | Penguin-DNPds | 02:34.30 |
Round two
Racers need to be here to submit a model for the second round. If they don’t, they will be skipped.
Penguin-DNPds has better in sectors one and two only to start struggling on box four again. There were 37 resets in first run, second had 29 resets in lap one, still enough to get a lap of 2:23.754 which is an improvement only to get denied by box four again in lap two.
ZhengYi-NYCU-CGI has struggled with resets in round one and managed to get two resets in first sector of lap one. There were resets in just about every spot of the track but again, 52.669 and we have an improved time. Second lap was over one minute and lap three ended on box two. That’s the learning experience, I’m sure will see ZhengYi in future months.
JPMC-HandsFree-London has started really fast only to get a four resets in sector two and one more in sector five. Time of 48.915, no improvement here. Second lap at 49 seconds and more resets in third, but low enough to improve with 39.713. Not enough to fight for the top three but JPMC are building their presence in top division and we’ll be hearing about them in the future months.
dartjason has really struggled with box two and hasn’t managed to complete a lap in this run. Jason has however improved the reset record for the finale with a count of 43.
rosscomp1 started fast and straight into box two, but only once, as opposed to box four which received some serious beating. 40 more resets and DNF.
Yi-Li-NYCU-CGI has started pretty nicely and with one reset clocked 29.820. Two resets in lap two and over 33 seconds time. Looks like the car is struggling with higher speeds, that will have to improve to be higher. Finally Yi-Li has completed a great clean lap at 24.835. Remember, that’s only 1.5 seconds from the current third.
AJM-Model-1 started rather cautiously but that has not prevented a reset right after a hairpin. That has made a lap of 30.574 s. He’s done similarly in lap two but slightly slower but a clean lap three had the time of 26.398 s. In lap four there have been four resets and in lap five a single reset which sadly is not enough to improve.
Matt started pretty cleanly only to skid and fall off the track before the finish, first lap completed at 30.018, but has managed to get down to 28, then 27, then 24.191. These are great times but just a little bit too slow to pose a threat to the podium. Still, well done with this time improvement!
Yoway-NYCU-CGI started aggressively and almost got off the track in sector one. Managed to do it in second and third sector and the first lap was completed with 32.861 s. The chicane has proven challenging in lap two and no time improvement at around 35 seconds. Third lap started clean in sectors one and two but an off-track in sector three. Still 31.615 seconds – slower than round one. Yoway appears to be pushing the speed as they have nothing to lose, but no improvement.
Ernesto started with a reset but still 27.755 seconds in lap one. In Lap two two resets and no improvement. Lap three improved in sector two and it’s 23.532s which goes up to fourth place currently. Next two laps around 26 seconds with a reset and that’s the end of the run.
Breadcentric is the last to compete with a chance to go onto the podium. Started with a clean lap of 23.929 which would put him in the very same fifth spot that he’s currently occupying, not enough for the championships ticket. As he pushed more he started getting more and more resets. Lap two in the area of 27 seconds, lap three over 29 seconds, laps four and five over 28 seconds and we know pretty much everything that matters. Breadcentric will have to fight some more.
Joffe is already going to Vegas, now it’s just time to show off. First lap clean at 24.339 and then some resets in lap two, lap three at 23.605. That’s a fast model. Lap four again, at 22.999 followed by a reset on a starting line and a few more later. Joffe may have pushed the speed a bit too much in that lap. Slightly more conservative behaviour in the next lap and with one reset it was just under 25 seconds – that could compete with Jochem without the reset. Great finishing lap.
JPMC-DriftKing-Houston said in the interview he wanted to push a bit more in round two. First lap started nicely but not fast enough with 24.114 s, followed by a similar time, no improvement – the car got a bit confused in sector two. Third lap however brought 22.169 seconds. Not enough for the first but much better then round one. Two resets wasted lap four. All of his laps were pretty impressive, all in the area of the second spot but just not faster. DriftKing goes to Vegas!
Jochem has a victory round now! First lap was pretty safe at 24.420. Second lap sees improvement to 23.439. That’s slow for him but bear in mind this is a podium time. Lap three has seen a few resets and ended over 30 seconds. Lap four was around 23.7 and a reset in lap five. Lap six was similar to 23.439, only a bit slower. Not that it matters, really. Jochem goes to Vegas as well.
| Place | Racer | Time 1 | Time 2 | Final |
| 1 | Jochem | 21.995 | 23.439 | 21.995 |
| 2 | JPMC-DriftKing-Houston | 22.483 | 22.169 | 22.169 |
| 3 | Joffe | 23.2 | 22.999 | 22.999 |
| 4 | Ernesto | 23.73 | 23.532 | 23.532 |
| 5 | Breadcentric | 23.664 | 23.929 | 23.664 |
| 6 | Matt | 24.747 | 24.191 | 24.191 |
| 7 | Duckworth | 24.596 | DNS | 24.596 |
| 8 | Yoway-NYCU-CGI | 24.672 | 31.615 | 24.672 |
| 9 | Yi-Li-NYCU-CGI | 26.004 | 24.835 | 24.835 |
| 10 | AJM-Model-1 | 25.595 | 26.398 | 25.595 |
| 11 | Rober | 26.664 | DNS | 26.664 |
| 12 | rosscomp1 | 27.322 | DNF | 27.322 |
| 13 | dartjason | 31.169 | DNF | 31.169 |
| 14 | JPMC-HandsFree-London | 47.478 | 39.713 | 39.713 |
| 15 | ZhengYi-NYCU-CGI | 53.649 | 52.669 | 52.669 |
| 16 | Penguin-DNPds | 02:34.30 | 02:23.75 | 02:34.30 |
This is it: Jochem, JPMC-DriftKing-Houston and Joffe are going to Las Vegas. Congratulations, well deserved!
Also, since I believe at least six finalists have already been in the finale before, everyone who hasn’t yet will receive a DeepRacer Evo!
This was a very impressive race, just look at this: the difference between the first and tenth place is 3.5 seconds. Racers have presented solid models and only some have struggled with stability which really is unavoidable when you don’t know where the obstacles will be on track.
If you haven’t done so, join the AWS Machine Learning Community and race with us for the ultimate bragging rights with AWS DeepRacer. Follow the blog for regular updates from the races and other news about DeepRacer.