Let’s have a look what the last week brought in the AWS DeepRacer League. This update has been prepared with use of AWS DeepRacer Race Data Project.
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
AWS Deepracer Women’s League India 2021 Invites students in India to join the race for the ML knowledge, bragging rights and some pretty sweet prizes. More info on our blog.
Pro Division
Race info:
- Track named after the winner of 2020 AWS DeepRacer League
- 89.24 meters long
- 107 centimeters 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: 2021-03-21 18:00:00 UTC Comparing to 2021-03-14 23:00:00 UTC Racers: 53 (+5) Total submission attempts: 40760 (+17797) Best total time: 01:27.232 (-00:00.935) by JJ Mean total time: 03:21.908716981 (-00:26.018824686) Best lap: 00:28.262 (-00:00.794) by JJ Mean best lap: 01:02.921415094 (-00:07.844959906) Mean average lap: 01:07.302584906 (-00:08.672935928)
Five racers have joined this week which brings the total to 53 out of 75. I’m wondering if others are training at all – taken into account that three out of five have made a debut in the top 16 I wouldn’t be surprised if that was the case.
JJ holds the lead firmly and keeps improving but others haven’t been idle. A significant drop in mean total time suggests that we’ve had a few outliers (until now) jump right into action. Looking at the standings below I can see that Fumiaki, RobinCastro-DBS and Joffe have contributed to this quite significantly.
Time for the top 20:
Top 20 submissions Update time: 2021-03-21 18:00:00 UTC Comparing to 2021-03-14 23:00:00 UTC Racers: 20 Total submission attempts: 28970 (+9682) Best total time: 01:27.232 (-00:00.935) by JJ Mean total time: 01:46.129750 (-00:11.716750) Best lap: 00:28.262 (-00:00.794) by JJ Mean best lap: 00:33.823550 (-00:02.776350) Mean average lap: 00:35.376250 (-00:03.905550)
I’m really curious to know if JJ can get anything extra out of his model. If he had three top laps in his run that would mean a total around 1:24.7. Interestingly though, he just keeps improving both the best laps and the total. The rest of the group keeps closing in on him but last week the average lap got 20 seconds closer to the best and this week only 10 seconds. It will get pretty crowded near the top by the end of the month.
This is what the standings look like for the top 20 this week:
| Rank | Racer | Total Time | Avg Lap | Best Lap | Resets | Submissions |
|---|---|---|---|---|---|---|
| 1 (↔0) | JJ | 01:27.232 (-00:00.935) | 00:29.077 (-00:00.312) | 00:28.262 (-00:00.794) | 0 (-0) | 2859 (+1158) |
| 2 (↔0) | RogerRabbit | 01:30.203 (-00:03.075) | 00:30.067 (-00:01.025) | 00:29.277 (-00:01.249) | 0 (-0) | 2073 (+1094) |
| 3 (–) | flatearth | 01:30.383 (–:–.—) | 00:30.127 (–:–.—) | 00:28.964 (–:–.—) | 0 (–) | 233 (–) |
| 4 (↔0) | Duckworth | 01:35.791 (-00:02.983) | 00:31.930 (-00:00.994) | 00:31.727 (-00:00.457) | 0 (-0) | 1260 (+348) |
| 5 (–) | Karl-NAB | 01:36.495 (–:–.—) | 00:32.165 (–:–.—) | 00:31.596 (–:–.—) | 0 (–) | 279 (–) |
| 6 (⬆4) | DBro | 01:37.312 (-00:20.134) | 00:32.437 (-00:06.711) | 00:31.919 (-00:01.094) | 0 (-3) | 1657 (+1401) |
| 7 (⬇4) | Jochem | 01:37.827 (-00:00.000) | 00:32.609 (-00:00.000) | 00:32.259 (-00:00.000) | 0 (-0) | 873 (+110) |
| 8 (↔0) | Maikel | 01:41.198 (-00:07.515) | 00:33.732 (-00:02.505) | 00:33.041 (-00:00.883) | 0 (-1) | 746 (+146) |
| 9 (⬆39) | Fumiaki | 01:42.893 (-09:56.599) | 00:34.297 (-03:18.867) | 00:33.508 (-03:08.686) | 0 (-105) | 593 (+592) |
| 10 (⬇3) | Ernesto | 01:43.859 (-00:04.551) | 00:34.619 (-00:01.517) | 00:32.268 (-00:01.073) | 1 (-0) | 2338 (+975) |
| 11 (⬇6) | AJM-Model-1 | 01:44.016 (-00:00.000) | 00:34.672 (-00:00.000) | 00:33.823 (-00:00.000) | 0 (-0) | 1434 (+132) |
| 12 (⬇6) | PolishThunder | 01:44.028 (-00:00.000) | 00:34.676 (-00:00.000) | 00:32.252 (-00:00.000) | 1 (-0) | 3315 (+1188) |
| 13 (–) | Yxcas-DBS | 01:46.912 (–:–.—) | 00:35.637 (–:–.—) | 00:35.397 (–:–.—) | 0 (–) | 1030 (–) |
| 14 (⬆10) | RobinCastro-DBS | 01:47.637 (-00:52.862) | 00:35.879 (-00:17.620) | 00:34.972 (-00:13.976) | 0 (-3) | 1183 (+1167) |
| 15 (⬇6) | dartjason | 01:54.965 (-00:00.000) | 00:38.321 (-00:00.000) | 00:37.801 (-00:00.000) | 0 (-0) | 1998 (+1111) |
| 16 (⬇5) | atnmn | 01:59.058 (-00:00.442) | 00:39.686 (-00:00.147) | 00:36.499 (-00:00.993) | 1 (-0) | 1530 (+616) |
| 17 (⬇1) | ep-dr-racer1 | 02:03.517 (-00:10.388) | 00:41.172 (-00:03.463) | 00:38.975 (-00:02.030) | 4 (+2) | 1140 (+744) |
| 18 (⬆1) | rosscomp1 | 02:03.838 (-00:15.177) | 00:41.279 (-00:05.059) | 00:35.764 (-00:03.023) | 3 (-1) | 979 (+680) |
| 19 (⬆13) | Joffe | 02:07.443 (-01:42.394) | 00:42.481 (-00:34.131) | 00:35.768 (-00:34.747) | 4 (-16) | 40 (+35) |
| 20 (⬇8) | kimwooglae | 02:07.988 (-00:00.000) | 00:42.662 (-00:00.000) | 00:42.399 (-00:00.000) | 3 (-0) | 3410 (+1162) |
Great to see that Fumiaki has jumped into submissions. His 10 minute improvement looks impressive but shouldn’t really shock – before he just submitted once and either trained aside till he had a model good enough to present or just waited for the right time to pick up training. Robin Castro seems to have found some great improvement opportunity and has jumped up to 14th, right behind his colleague Yxcas. Their supporters at DBS will surely be pleased (Let me just remind that DBS have used AWS DeepRacer and organised an internal league for over 3000 employees as part of their AI and ML training initiative; you can see more about it on YouTube).
Both flatearth and Karl-NAB have had a good start this week. Flatearth has been the finalist in the last year’s edition whereas Karl Has been with us from the very start and has dominated the 2019 qualifiers but struggled in the finals, and in 2020 he has been racing in the great finale grand prix race to finish fourth. Just a quick reminder of that event which features also Duckworth, Robin and Jochem from the current top 16:
As expected, there’s quite a lot of activity around the 16th spot. The great final race live on Twitch is surely tempting!
Five of the top 20 racers have not improved this week. Four of the remain in the top 16 but for how long? Time will tell.
A small extra this week: have a look at how current top 16 racers have been improving their times in March:
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 centimeters 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: 2021-03-21 18:00:00 UTC Comparing to 2021-03-14 23:00:00 UTC Racers: 760 (+199) Total submission attempts: 7361 (+2274) Best total time: 01:19.407 (-00:01.819) by VaughanDiesel Mean total time: 05:32.517264474 (+00:07.955337557) Best lap: 00:25.720 (-00:00.543) by VaughanDiesel Mean best lap: 01:44.741538158 (+00:02.543507854) Mean average lap: 01:50.838767105 (+00:02.651786713)
200 new racers mean that the Pro Division will welcome 76 new racers, and we’re not done yet. I would like to draw your attention to the mean total time and mean best lap time increase. Last week I was surprised to see those values go down, this time most likely the top group racers have not been too eager to improve their results compared to the new starters’ first-try performances. It would be interesting to observe the improvement trends per a group of racers starting on a given day.
VaughanDiesel has done it! After a week of trying we have a new-old leader in the Open Division. Nearly 2 seconds shaved off the previous result are quite impressive but clearly required a lot of effort and took quite a long time to deliver.
Despite 20 new qualifying places, the time required has gone down to 2:38.897 from 2:39.591 today. This suggests that we are having a noticeable activity around the qualification spot. tomroy1 who was in the last qualifying spot last week has been inactive recently and fell down by twenty two spots. This racer will need to try harder to get back into the game.
In the top 20:
Top 20 submissions Update time: 2021-03-21 18:00:00 UTC Comparing to 2021-03-14 23:00:00 UTC Racers: 20 Total submission attempts: 1278 (+473) Best total time: 01:19.407 (-00:01.819) by VaughanDiesel Mean total time: 01:44.601600 (-00:11.404800) Best lap: 00:25.720 (-00:00.543) by VaughanDiesel Mean best lap: 00:33.353900 (-00:02.522750) Mean average lap: 00:34.866950 (-00:03.801400)
Despite nice improvements of over 11 seconds in the mean total lap time, there’s still a huge gap of 25 seconds from the leader. I’m far from considering the leader a winner at this point as we can have a few turns of events in the upcoming week but if you’re waiting for VaughanDiesel to fall out of the qualifying group, well, I wouldn’t hold my breath if I were you.
Check out the top 20 standings:
| Rank | Racer | Total Time | Avg Lap | Best Lap | Resets | Submissions |
|---|---|---|---|---|---|---|
| 1 (⬆1) | VaughanDiesel | 01:19.407 (-00:02.150) | 00:26.469 (-00:00.716) | 00:25.720 (-00:00.883) | 0 (-0) | 434 (+216) |
| 2 (–) | BeaconHillTheFastestVer2 | 01:20.788 (–:–.—) | 00:26.929 (–:–.—) | 00:26.593 (–:–.—) | 0 (–) | 10 (–) |
| 3 (⬇2) | GregRacing | 01:20.963 (-00:00.263) | 00:26.987 (-00:00.088) | 00:26.407 (+00:00.144) | 0 (-0) | 17 (+4) |
| 4 (⬆5) | s1thind | 01:25.966 (-00:36.504) | 00:28.655 (-00:12.168) | 00:28.316 (-00:03.550) | 0 (-5) | 52 (+25) |
| 5 (↔0) | Jerec | 01:30.958 (-00:13.680) | 00:30.319 (-00:04.560) | 00:29.660 (-00:04.675) | 0 (-0) | 145 (+126) |
| 6 (–) | xuqian | 01:34.356 (–:–.—) | 00:31.452 (–:–.—) | 00:30.056 (–:–.—) | 0 (–) | 67 (–) |
| 7 (⬇4) | AIDeepRacer | 01:34.883 (-00:07.245) | 00:31.627 (-00:02.415) | 00:30.665 (+00:00.528) | 0 (-2) | 75 (+11) |
| 8 (⬇4) | satussy777 | 01:43.095 (-00:00.000) | 00:34.365 (-00:00.000) | 00:30.529 (-00:00.000) | 5 (-0) | 85 (+38) |
| 9 (⬇3) | Redfalcon5 | 01:46.018 (-00:00.000) | 00:35.339 (-00:00.000) | 00:34.997 (-00:00.000) | 0 (-0) | 49 (+11) |
| 10 (⬇3) | jmp543 | 01:47.158 (-00:00.000) | 00:35.719 (-00:00.000) | 00:35.609 (-00:00.000) | 0 (-0) | 59 (+3) |
| 11 (↔0) | DanMeon | 01:48.651 (-00:15.258) | 00:36.217 (-00:05.086) | 00:35.149 (-00:05.425) | 0 (-0) | 14 (+3) |
| 12 (⬇2) | muzi | 01:51.642 (-00:10.871) | 00:37.214 (-00:03.623) | 00:34.444 (-00:00.584) | 2 (-1) | 152 (+119) |
| 13 (–) | greyracer | 01:53.440 (–:–.—) | 00:37.813 (–:–.—) | 00:34.543 (–:–.—) | 1 (–) | 33 (–) |
| 14 (⬆234) | TheMontrealTurtle | 01:53.528 (-02:43.978) | 00:37.842 (-00:54.660) | 00:35.141 (-00:53.384) | 1 (-0) | 13 (+12) |
| 15 (⬆18) | LawrenceWong-hk | 01:53.724 (-00:27.190) | 00:37.908 (-00:09.063) | 00:36.864 (-00:07.642) | 0 (-0) | 14 (+3) |
| 16 (–) | Mungi | 01:55.962 (–:–.—) | 00:38.654 (–:–.—) | 00:37.458 (–:–.—) | 0 (–) | 2 (–) |
| 17 (⬆71) | Sl1deways | 02:01.086 (-01:16.809) | 00:40.362 (-00:25.603) | 00:39.326 (-00:25.609) | 0 (-0) | 3 (+2) |
| 18 (⬇10) | TestVinodRacer | 02:02.146 (-00:00.000) | 00:40.715 (-00:00.000) | 00:38.675 (-00:00.000) | 1 (-0) | 43 (-0) |
| 19 (⬆75) | amanelso4 | 02:03.675 (-01:17.425) | 00:41.225 (-00:25.808) | 00:39.458 (-00:25.080) | 1 (-3) | 5 (+3) |
| 20 (⬇8) | m-hosokawa-nes | 02:04.586 (-00:00.000) | 00:41.528 (-00:00.000) | 00:37.468 (-00:00.000) | 2 (-0) | 6 (-0) |
We’ve got four new-starters in the leading group this week, pretty good. If it wasn’t for the today’s dash for the finish line by VaughanDiesel, one of them would be a new leader. All I can tell about BeaconHillTheFastestVer2 is that UX experts on the AWS DeepRacer Team might want to ensure people don’t confuse racer name field with a model selection on the race page. Soon we should be expecting some new features on the page including a racer profile, so perhaps it will now be easier to avoid the confusion.
The dynamics of improvements in the top 20 are pretty interesting, you can cut 10% off your time and it would be enough to only hold on to your spot. Interesting stuff. I wonder if we will have all times below the two minutes limit in the standings next week. Last week we’ve had 13, this week it’s only 4. I think it’s unavoidable.
Let’s finish off with a plot of time improvements of current top 20:
The first four racers have improved their times in the last 48 hours and only five from the leading group have not improved this week. A few of them might not hold on to the top 20 after this week.
Two areas to watch closer in the remaining ten days are the podium where we have three racers within 1.5 seconds of each other (and no one can guarantee that they will still be the top three next week) and the qualification line where, unless we get a new sudden influx of new racers, I’m predicting the required time to go down every day. Will tomroy1 improve enough to qualify?
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!