We are nearing the end of March races. On Thursday we will know who makes it into the Pro Division and also who will qualify for the great finale live race to fight for the first three spots in the finals. We will also have a look at how the time improvements are shaping in among the top racers in each group.
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
If you haven’t done so yet, there’s still time to register for the AWS Deepracer Women’s League India 2021 which invites students in India to join the race for the ML knowledge, bragging rights and some pretty sweet prizes. More info on our blog.
I would like to share a few words on the finale race that is approaching. It will take place on the 8th of April at 5:30 pm PST (9th of April 1:30 am BST, 9th of April 10:30 pm AEDT). It will be presented live on the main AWS Twitch channel: https://www.twitch.tv/aws.
While it might be a bit of a challenging time for most of you, the AWS DeepRacer Community will make sure they are there to bring you the report from the race on the blog.
We’ve got some really exciting new learning resources available for analysing physical race performance using Amazon Rekognition Custom Labels and using game controllers to steer your DeepRacer. Be sure to check them out.
Pro Division
Race info:
- Track named after the winner of 2020 AWS DeepRacer League
- 89.24 meters long
- 107 centimetres wide
- Multiple difficult turns with the most tricky one exiting a chicane past the hairpin in the upper-right corner, branded Yellow Brick Road since it looks like the beginning of Yellow Brick Read from The Wizard of Oz
- Head-to-bot format with four bot cars cruising at 2 m/s without lane changes
- Racers are trying to complete three laps
- Each collision and each driving off the track results in a five seconds penalty
- Top 16 go on into a live race which will be streamed on Twitch on the 8th of April to fight for places in the League Finals and AWS DeepRacer EVOs
Let’s see the stats for the whole group:
All submissions Update time: 2021-03-28 18:00:00 UTC Comparing to 2021-03-21 18:00:00 UTC Racers: 62 (+9) Total submission attempts: 61059 (+20299) Best total time: 01:26.927 (-00:00.305) by JJ Mean total time: 03:29.082467742 (+00:07.173750761) Best lap: 00:28.676 (+00:00.414) by JJ Mean best lap: 01:05.275080645 (+00:02.353665551) Mean average lap: 01:09.693806452 (+00:02.391221546)
We have nine new racers which means that 62 out of 75 have started. I’d say it’s a pretty good turnout already. 20299 submissions in the last week means that we’re gaining momentum but the increase in average total time makes me suspect that some have just shown up to mark their presence. Maybe it’s prep for the next month? I have reasons to think that next month will be pretty interesting, watch this space.
One very interesting observation: JJ has improved his total time again, but his best lap has gone down. I’m curious what’s going on in there. Feels like he has found some serious improvements to his slower laps by progressing the stability more than peek performance. Or it might be just by chance since he’s slowly running out of improvement room by this point, and things are getting pretty crowded near the top:
Top 20 submissions Update time: 2021-03-28 18:00:00 UTC Comparing to 2021-03-21 18:00:00 UTC Racers: 20 Total submission attempts: 39105 (+10135) Best total time: 01:26.927 (-00:00.305) by JJ Mean total time: 01:39.244200 (-00:06.885550) Best lap: 00:28.676 (+00:00.414) by JJ Mean best lap: 00:32.212 (-00:01.611550) Mean average lap: 00:33.081 (-00:02.295250)
The mean time for top twenty has gone down by almost 7 seconds, we’ve got seven racers less than 10 seconds behind him but he’s still maintaining that 3 seconds advantage over the second flatearth. The top 20 mean total improvements are slowing down which makes me suspect that JJ should be safe with his first spot. This won’t give him anything extra apart from maybe ability to start last in the great finale race but still it’s nice to dominate.
The top 20 racers in the pro division go brrrrrrrrrrrrrrrrrrrrrrr, just look at this:
Rank | Racer | Total Time | Avg Lap | Best Lap | Resets | Submissions |
---|---|---|---|---|---|---|
1 (↔0) | JJ | 01:26.927 (-00:00.305) | 00:28.975 (-00:00.102) | 00:28.676 (+00:00.414) | 0 (-0) | 4024 (+1165) |
2 (⬆1) | flatearth | 01:29.800 (-00:00.583) | 00:29.933 (-00:00.194) | 00:29.325 (+00:00.361) | 0 (-0) | 1325 (+1092) |
3 (⬇1) | RogerRabbit | 01:30.203 (-00:00.000) | 00:30.067 (-00:00.000) | 00:29.277 (-00:00.000) | 0 (-0) | 3209 (+1136) |
4 (⬆2) | DBro | 01:31.268 (-00:06.044) | 00:30.422 (-00:02.015) | 00:29.935 (-00:01.984) | 0 (-0) | 2998 (+1341) |
5 (⬆4) | Fumiaki | 01:35.555 (-00:07.338) | 00:31.851 (-00:02.446) | 00:31.202 (-00:02.306) | 0 (-0) | 1597 (+1004) |
6 (⬇2) | Duckworth | 01:35.791 (-00:00.000) | 00:31.930 (-00:00.000) | 00:31.727 (-00:00.000) | 0 (-0) | 1260 (-0) |
7 (⬇2) | Karl-NAB | 01:36.495 (-00:00.000) | 00:32.165 (-00:00.000) | 00:31.596 (-00:00.000) | 0 (-0) | 279 (-0) |
8 (⬇1) | Jochem | 01:36.783 (-00:01.044) | 00:32.261 (-00:00.348) | 00:31.500 (-00:00.759) | 0 (-0) | 1973 (+1100) |
9 (⬆1) | Ernesto | 01:37.185 (-00:06.674) | 00:32.395 (-00:02.224) | 00:31.607 (-00:00.661) | 0 (-1) | 3293 (+955) |
10 (⬆5) | dartjason | 01:38.468 (-00:16.497) | 00:32.822 (-00:05.499) | 00:31.603 (-00:06.198) | 0 (-0) | 3074 (+1076) |
11 (⬆3) | RobinCastro-DBS | 01:38.503 (-00:09.134) | 00:32.834 (-00:03.045) | 00:32.302 (-00:02.670) | 0 (-0) | 2585 (+1402) |
12 (⬇4) | Maikel | 01:39.672 (-00:01.526) | 00:33.224 (-00:00.508) | 00:32.740 (-00:00.301) | 0 (-0) | 952 (+206) |
13 (–) | JPMC-DriftKing | 01:40.847 (–:–.—) | 00:33.615 (–:–.—) | 00:32.663 (–:–.—) | 0 (–) | 253 (–) |
14 (⬇2) | PolishThunder | 01:41.710 (-00:02.318) | 00:33.903 (-00:00.773) | 00:31.598 (-00:00.654) | 1 (-0) | 4653 (+1338) |
15 (⬇4) | AJM-Model-1 | 01:44.016 (-00:00.000) | 00:34.672 (-00:00.000) | 00:33.823 (-00:00.000) | 0 (-0) | 1442 (+8) |
16 (⬆3) | Joffe | 01:45.357 (-00:22.086) | 00:35.119 (-00:07.362) | 00:32.476 (-00:03.292) | 1 (-3) | 122 (+82) |
17 (–) | GT-DevelopersIO | 01:46.208 (–:–.—) | 00:35.402 (–:–.—) | 00:33.502 (–:–.—) | 1 (–) | 64 (–) |
18 (⬇5) | Yxcas-DBS | 01:46.787 (-00:00.125) | 00:35.595 (-00:00.042) | 00:34.938 (-00:00.459) | 0 (-0) | 2252 (+1222) |
19 (⬇1) | rosscomp1 | 01:49.679 (-00:14.159) | 00:36.559 (-00:04.720) | 00:36.228 (+00:00.464) | 0 (-3) | 1329 (+350) |
20 (⬇4) | atnmn | 01:53.630 (-00:05.428) | 00:37.876 (-00:01.810) | 00:37.522 (+00:01.023) | 0 (-1) | 2421 (+891) |
Only four racers haven’t improved in the last week and only two haven’t been trying. Looks like Karl-NAB and Duckworth have decided that their results are just good enough and diverted their focus to the great finale prep already. They both have 9 seconds advantage over the 16th place. RogerRabbit and AJM-Model-1 might have gotten stuck a little bit and while Roger is in a pretty comfortable spot, AJM might need to think of a plan B or the next three days might be a little bit nerve-wrecking. Just look at what’s going on behind: Joffe kicked back in with 22 seconds improvement and rosscomp1 managed to cut 14 seconds and this was just enough to hold on to top twenty.
Back to the good old question (this is not a betting advice): who will we see in the final race? Tricky question, we’re still missing 14 racers from the pool and until they show up or the race ends we will just not know if they are getting ready or not. Looking at the March Qualifier, we’re still waiting for condoriano (11th in that race), Yeti (35th), IcePandaBoom (39th), Granville777 (41st), alexj (46th), Shibalnu-DevelopersIO (53rd), gigi-DNPds (54th), Jianlong-DBS (59th), iana (62nd), msckrishna (63rd), JaredF (64th), Ohrid (66th), SSB (67th), firnnauriel (71st). These might not be very impressive positions last month but quite a few of them have quite a few races under their belts and should not be written off just yet.
Then there’s a question of how much progress we can see in the last three days and it might be a lot. There are five racers within 10 seconds of the top 16 and we’ve seen time and again that if someone finds a way past whatever hurdle they are dealing with right now, those seconds get shaved off pretty quickly.
We’ve had some nice improvements this week. JPMC-DriftKing started very strong and I’m curious to see if he can hold his spot. I suspect he will improve the time, but then almost everyone is.
Open Division
Open races will always be slightly easier than the Pro ones. Let’s see what’s on in March:
Race info:
- Po-Chun Speedway is a simplified version of the Pro track
- 68.68 meters long
- 107 centimetres wide
- Simpler and shorter than the pro track but still pretty challenging with a variety of straights and turns, some of them pretty tight
- Time Trial format
- Competitors need to complete three laps and total time counts for their classification
- Each off-track results in five seconds penalty.
- Racers fight for qualification to the Pro group which is won by the top 10%
- They will also receive a Pro Welcoming Kit which is a bit of a mystery
Let’s take a look at the overall stats:
All submissions Update time: 2021-03-28 18:00:00 UTC Comparing to 2021-03-21 18:00:00 UTC Racers: 915 (+155) Total submission attempts: 9659 (+2298) Best total time: 01:14.027 (-00:05.380) by Mungi Mean total time: 05:35.620308197 (+00:03.103043723) Best lap: 00:22.356 (-00:03.364) by Yoway-NYCU-CGI Mean best lap: 01:45.710 (+00:00.968461842) Mean average lap: 01:51.873112568 (+00:01.034345463)
Pretty good influx of new racers. With the counter at 915 this is the ninth biggest race so far already and it can still realistically jump up by a place or two. This also means that there are 92 spots in the Pro Division qualification pool. The last qualifier, hduddu has a time of 2:36.130 which is almost three seconds less than last week. tomroy1 who was the last to qualify last week has not managed to improve and is currently at 97th. That may seem like not a lot to improve but when you’re stuck it can be really tricky to get out of this state, and frustrating too.
Fun fact: right before hduddu we have GummyBear, a 2020 League Finalist. You might need to squeeze a few seconds extra to make it, I’m afraid!.
Mean total time is only marginally slower than last week which makes me thing that we’ve had a few significant improvements on the record this week.
For the first time we’ve got a split between the best total time and best lap. Mongi has jumped up by 15 places and cut his time down by almost 42 seconds to get five seconds ahead of the new second. VaughanDiesel may have concluded the racing for the month: This wouldn’t surprise me – there is no prize for the winner in the Open Division (maybe there should be, AWS?). There are a few new submissions but nowhere close to what we’ve seen in the previous weeks.
Top 20 submissions Update time: 2021-03-28 18:00:00 UTC Comparing to 2021-03-21 18:00:00 UTC Racers: 20 Total submission attempts: 1714 (+436) Best total time: 01:14.027 (-00:05.380) by Mungi Mean total time: 01:33.494300 (-00:11.107300) Best lap: 00:22.356 (-00:03.364) by Yoway-NYCU-CGI Mean best lap: 00:29.890750 (-00:03.463150) Mean average lap: 00:31.164500 (-00:03.702450)
With all the improvements that Mungi presented this week the top twenty as a whole still delivered even more and we’ve got that crowded area near the top. Racers are pretty active. If you look at the standings below you will see where the split of best lap and total time comes from:
Rank | Racer | Total Time | Avg Lap | Best Lap | Resets | Submissions |
---|---|---|---|---|---|---|
1 (⬆15) | Mungi | 01:14.027 (-00:41.935) | 00:24.675 (-00:13.979) | 00:24.249 (-00:13.209) | 0 (-0) | 13 (+11) |
2 (–) | Yoway-NYCU-CGI | 01:15.805 (–:–.—) | 00:25.268 (–:–.—) | 00:22.356 (–:–.—) | 1 (–) | 1 (–) |
3 (⬆4) | AIDeepRacer | 01:16.075 (-00:18.808) | 00:25.358 (-00:06.269) | 00:24.138 (-00:06.527) | 0 (-0) | 137 (+62) |
4 (⬇2) | BeaconHillTheFastestVer2 | 01:17.297 (-00:03.491) | 00:25.765 (-00:01.164) | 00:25.263 (-00:01.330) | 0 (-0) | 14 (+4) |
5 (–) | TimJ-KMITLXRAI | 01:18.232 (–:–.—) | 00:26.077 (–:–.—) | 00:25.606 (–:–.—) | 0 (–) | 24 (–) |
6 (⬇5) | VaughanDiesel | 01:19.407 (-00:00.000) | 00:26.469 (-00:00.000) | 00:25.720 (-00:00.000) | 0 (-0) | 451 (+17) |
7 (⬇3) | s1thind | 01:19.551 (-00:06.415) | 00:26.517 (-00:02.138) | 00:26.198 (-00:02.118) | 0 (-0) | 108 (+56) |
8 (⬇3) | Jerec | 01:20.433 (-00:10.525) | 00:26.811 (-00:03.508) | 00:26.461 (-00:03.199) | 0 (-0) | 231 (+86) |
9 (⬇6) | GregRacing | 01:20.956 (-00:00.007) | 00:26.985 (-00:00.002) | 00:26.126 (-00:00.281) | 0 (-0) | 25 (+8) |
10 (⬇4) | xuqian | 01:34.356 (-00:00.000) | 00:31.452 (-00:00.000) | 00:30.056 (-00:00.000) | 0 (-0) | 123 (+56) |
11 (⬆39) | Jackandwho | 01:40.376 (-00:43.130) | 00:33.458 (-00:14.377) | 00:33.205 (-00:12.140) | 0 (-7) | 6 (+2) |
12 (⬇4) | satussy777 | 01:41.325 (-00:01.770) | 00:33.775 (-00:00.590) | 00:28.867 (-00:01.662) | 5 (-0) | 120 (+35) |
13 (⬇1) | muzi | 01:42.262 (-00:09.380) | 00:34.087 (-00:03.127) | 00:32.139 (-00:02.305) | 1 (-1) | 193 (+41) |
14 (⬇5) | Redfalcon5 | 01:46.018 (-00:00.000) | 00:35.339 (-00:00.000) | 00:34.997 (-00:00.000) | 0 (-0) | 49 (-0) |
15 (⬇5) | jmp543 | 01:47.158 (-00:00.000) | 00:35.719 (-00:00.000) | 00:35.609 (-00:00.000) | 0 (-0) | 61 (+2) |
16 (⬆6) | hypersky | 01:47.265 (-00:17.786) | 00:35.755 (-00:05.928) | 00:35.128 (-00:04.532) | 0 (-1) | 79 (+49) |
17 (⬇6) | DanMeon | 01:48.651 (-00:00.000) | 00:36.217 (-00:00.000) | 00:35.149 (-00:00.000) | 0 (-0) | 14 (-0) |
18 (⬇5) | greyracer | 01:53.440 (-00:00.000) | 00:37.813 (-00:00.000) | 00:34.543 (-00:00.000) | 1 (-0) | 33 (-0) |
19 (⬇5) | TheMontrealTurtle | 01:53.528 (-00:00.000) | 00:37.842 (-00:00.000) | 00:35.141 (-00:00.000) | 1 (-0) | 13 (-0) |
20 (⬇5) | LawrenceWong-hk | 01:53.724 (-00:00.000) | 00:37.908 (-00:00.000) | 00:36.864 (-00:00.000) | 0 (-0) | 19 (+5) |
Yoway-NYCU-CGI has submitted the model once. Yes, one single time, no retries, and it was enough to get into second with one reset – one reset less and there would be a new leader. If you’re new to the races, make sure you pay close attention to racers having prefixes NYCU-CGI or previously NCTU-CGI in their names. NYCU is the National Yang Ming Chiao Tung University created through a merger of National Yang Ming University (NYMU) and National Chiao Tung University (NCTU). NCTU bring their technical focus into the new school and that includes the Computer Games and Intelligence Lab. Their racers have been present since the 2019 season where Roger finished third in the finals. In 2020 PoChun-NCTU-CGI has won the final and Kuei-NCTU-CGI ended third. They aren’t messing around here. It will be interesting to see how they manage in the new format as while the finals suggest they have the head to head format worked out, last year’s qualifiers told a different story – all of their students qualified through other formats. There’s also another challenge which would be interesting: I believe I haven’t seen CGI lab students doing well against bots or obstacles when they don’t know their locations. This will be interesting in the upcoming months.
Compliments go out to Jackandwho who presents the most spectacular time improvement in the top 20: 43 seconds down, 35 of which come from getting rid of those seven resets. Even more actually as you should remember that after a reset you start from a halt. In the 2020 final that cost Duckworth a top spot on the podium.
In the last week we’ve had three leaders. Two of the top five racers have only joined the competition last week which shows how many surprises there may be for us in the upcoming three days. How it ends? We’ll soon learn with our end of month special update.
If you’d like to learn some Machine Learning skills while racing for glory, be sure to check out the AWS DeepRacer Page for hints on how to get started. Be sure to join the AWS Machine Learning Community on Slack for more tips and tricks on how to improve your model. See you next week!