It is time to meet the ones who will be racing in the championships. To some the result may have come easily, others had to spend hours/days (well, nights) of thinking, learning and trying over and over again. All have however found a way to punch a ticket and will meet at the MGM Grand Arena on Monday morning for the first round of racing.
Make yourself comfortable and get a cuppa. The read is long, but there is a champion in this text and all have already achieved something amazing. See various backgrounds racers come from to get together and learn while competing and having fun.
Remember to also check all the details of DeepRacer-focussed events at re:Invent (read here), it’s time to get to know the racers.
AWS DeepRacer League 2021 Champion
Sairam Naragoni has been racing since 2020 and has been part of the Community from the very beginning. After his brilliant performance last year he’ll have many eyes set on his result in Vegas. And it will be interesting to see – as a JPMC employee Sairam has not only had good conditions to race virtually but also on a physical track. I had a brief conversation with Sairam.
Would you like to introduce yourself?
I am from Warangal, India working as a Software Engineer at JP Morgan Chase & Co. I did my undergrad in Computer Engineering and graduated in 2019. I have been an ambitious person for as far as I could remember and I love solving challenging problems with Code. I do Pencil Sketching in my free time, few of my ART works can be found here – https://www.instagram.com/d.rogue.artist/
I also started following Formula-1 after I started DeepRacer and totally in love with the sport – the pinnacle of racing and engineering.
What do you do at work?
Most of my experience lies in the Distributed Systems Architecture. I design and develop distributed microservices and event driven systems at JP Morgan Chase & Co.,
How did this year look for you DeepRacer-wise?
Yes, I was looking forward to the New Delhi in-person summit but AWS didn’t have the physical track at the Summit. Hence I traveled to the next closest city to take part in AWS Summit at Singapore where I came first. It was very exciting to meet RayG, Robin Castro, Pit Crew and a few of my fellow Deepracers from our Global Technology Center at Singapore. Had lots of fun at the event.
How has your racing evolved since the start?
There are 2 benchmarks that I have set up for myself over the course of the past 2.5 years:
– Testing the new strategy on reinvent 2018 track – from 14s to 7s (Virtual)
– JJ’s time on Virtual Circuit Leaderboard – How far behind am I ?
I started my journey on ML through DeepRacer in 2020. I gathered whatever I could from the Internet on AWS DeepRacer and basics of Reinforcement Learning while taking part in Internal JPMC leagues, refining my strategy trying to squeeze in every millisecond possible. The learning curve was pretty steep. I have trained around a 1000 models in a span of 4 months and ended up in the top-16 back in 2020. I was almost 3sec/lap behind JJ’s time.
In 2021, although I had enough knowledge on DeepRacer, I felt like I was missing the core knowledge of what powers it. So in the first half of 2021, I spent majority of my time, trying to understand Reinforcement Learning and Deep learning down from the basics – going through the research papers for QLearning, TRPO, PPO, SAC etc., I have read the book “Introduction to Reinforcement Learning” by Barto and Sutton which was very helpful. Again the learning curve was very steep. I struggled to qualify for the finals due to Object Avoidance format but those 4 months have helped me develop different OA & H2H strategies eventually leading me to earn the Championship Title. I was still a second per lap behind JJ’s time.
Having won the 2021 Championship, I have started my year on a high note – It was wonderful hearing everyone’s thoughts on my achievement. I was invited to multiple forums to talk about my journey, I have delivered workshops on Reinforcement Learning and DeepRacer, had the opportunity to interact with senior leaders and college grads joining the firm. I just hope there is at least one person who got inspired from my Journey.
This year I feel like I finally managed to be within 300-500ms/lap of JJ’s time on virtual circuit leaderboard but the format for 2022 league is something I have little experience on – Physical Racing. A New challenge and I am up for it. We will see how it will turn out 🙂
And how is your championship preparation going?
I’m not going to lie, , The 2022 Championship track is a bit scary for physical racing, with tight hairpins and chicanes. I find it very difficult to train a model on this track without overfitting certain parts. We have a 2018 reinvent track set up at our Global Technology center in Hyderabad. I am testing out new strategies on 2018 track but haven’t had a major break through yet. Physical racing is very unpredictable – I get sub 8s one day and the same model next day starts giving me 8.6s.
Where does that set your expectations defending your title?
Last Month, Max Verstappen was crowned as the World Champion in Formula-1 for the second time in a row – Just Saying 😛
Haha, #JustKidding, To be Honest, DeepRacer has been more of a Learning Journey to me than a Competition. The Wins and Titles are the cool perks along the way 🙂
Despite some hurdles along the way Sairam has managed to sort everything out and I’m super excited to meet him in person in Las Vegas. We’ll see how he does on track!
AWS re:Invent 2021 Wildcard Race
The AWS DeepRacer League 2021 championships happened online and we’ve had lots of emotions and learnings from that event. Not that many people know however that at re:Invent in 2021 we also had some racing for this season. We’ve had a track available at Caesar’s Forum and there was a day where the winner qualified for the 2022 championships. We raced using a time trial format on a 2019 re:Invent track, both new and professional racers welcome.
Enter Eric Morrison. You may not recognise the name from the community or racing in the season – that’s the beauty of physical races at the events where you get a shot and just might do well.
Eric is based in Western North Carolina. Professionally he’s a Senior .Net developer. He enjoys taking care of his animals and tinkering in his garage. He has shed some light on his adventure with DeepRacer this year:
DeepRacer has gotten me interested in ML with minimal feedback criteria. Companies often receive minimal points of feedback, so I believe models which can learn and adapt quickly on that limited information will be very desirable.
In my championship preparation I have been mostly trying to slim down my reward function and put more emphasis on the action space.
I am honestly just hoping my model works well outside of simulations and exploring ML in even more depth.
I’m curious to see how Eric does – I’ve been lucky to meet (and compete with) him that day, he’s a bright, friendly, smart and curious person – all the qualities that make a good DeepRacer contestant. He has not raced (publicly) throughout the season but physical racing always introduces challenges and opportunities so the game is open.
March Pro League qualifiers
Fun fact: the racing data history has information on over 60,000 entries into race from various racers since 2019.
Finale races in the virtual league have been following the same format as before: a live streamed race with objects to avoid on a track, some penalties for hitting the object or leaving the track, the tracks are large and challenging.
To qualify for the championships, you first had to qualify from an Open Division to Pro Division (top 10% of the racers), then win one of the top 16 spots in the Pro race. Once you’ve done that, the only thing left was to get onto the podium in the finale race.
Timothy Jennens is JJ’s son, he has been racing since May 2020 (he’s been trying to qualify for hat famous race in which RayG won against Daniel Ricciardo and Tatiana Calderón to win tickets to any F1 weekend anywhere in the world).
He rarely ends a race outside the virtual top 10, but also he doesn’t race regularly. That said when he does, he shows up, gets what he’s racing for pretty quickly and then disappears again.
This re:Invent will be the first time I’ll meet Timothy in person. I haven’t seen him race on a physical track yet so it is hard for me to predict what results he will get.
James Jennens or flatearth’s dad, has been around for a little longer. From 71 recorded participations he has won 49 races. A very friendly and passionate racer that is always around, first racing for a prize, then for fun. He started in July 2019, has been to the AWS Summit in Toronto where he finished second that year, then only narrowly missed a qualification into the 2019 championships where he had to accept a defeat from then emerging new racing superpower the NYCU students.
Since than he hasn’t been leaving anything to chance. Participates in most races from the first to the last hour, has made 268850 submissions across those 71 races. He has been taking part in the AWS underground races (where users can submit their models remotely for someone to race them on a physical track). This year he also joined five Summits winning in Milan, New York, Chicago and Ottawa, only having to accept PolishThunder’s victory in Atlanta. He has presented solid models. While there’s usually a lot of uncertainty around the physical races, I’m expecting James to be pushing to the end. We’ll see if his experience on other tracks will be sufficient for the championship race/
rosscomp1 is also a known name on the virtual tracks. Ross Williams joined the league in 2019, then had a break and returned in 2021, clinging to top spots ever since. He’s always presented solid results and has been present in the 2021 championships.
Here are a few word from Ross himself:
I’m from DeLand, Florida, got my degree in Computer Science at Stetson University, then moved to Orlando Florida. I’ve been working in tech for 15 years and working with AWS for 10, I’m the Director of CX Engineering for aqfer, a marketing data platform as a service, we help marketing and advertising companies with their big data problems, offering solutions that increase speed and reduce cost. I run a team of talented full stack developers that build our portals, interactive documentation, external tools, and examples. In 2019 I was introduced to DeepRacer by a colleague while I worked for Solodev, an Enterprise CMS built on AWS. I was able to win a DeepRacer that year, but had to take a break due to moving, having a new baby, and the pandemic hitting all at once. I got back involved in 2021, and its become one of my favorite hobbies. My other hobbies are playing golf, fishing, Home brewing, and BBQ.
During the 2021 season I dedicated a lot of time trying to catch up from what had happened the previous year, and improving my own strategies, with my goal being to make it back to re:Invent. I had previously attended as a vendor in 2016. I rose up the leaderboards quickly and secured a spot in the championship later in the season, ultimately taking 3rd place! This year, I had two goals, qualifying quickly to earn a spot at re:Invent, and to become the fastest virtual league racer in the monthly Qualifier by beating JJ. I achieved both of those by being one of the first 3 racers to qualify, and taking the #1 spot in the monthly race for July. Now my goal is focusing on keeping a spot on the Podium for the 2022 championship race.
This year my wife and I also welcomed another son in August, I hope to have both of my sons involved in DeepRacer once they are old enough!
My racing has evolved mainly by trying many different strategies, comparing the logs, and keeping what works. I’ve used a combination of console, EC2 instances, and home made gaming PCs to train tens of thousands of hours for models over the past two years. Many hours have been spent adjusting reward functions, trying various actions spaces, tuning hyper parameters, and then testing how different combinations of those work together. The result is being consistently one of the faster racers in the virtual league.
This year, a new challenge has presented itself now that we’re going to Re:invent: Physical racing. My experience has entirely been in the virtual league up until now. So the past month I’ve been making sure that my models will translate well from sim2real, by building a small track in my garage and testing with it. I plan on having a few different models prepared by the time I get to Vegas, and even though I lack the in person racing experience that others may have, I plan on being competitive.
And getting back on the podium 🙂
DeepRacer has been a great experience in many aspects, as it’s required me to use a lot of different technologies and different areas of expertise, such as AI/ML, data science, simulators, working on hardware for training, containerization, knowledge of various AWS services, etc. and that’s just for the virtual league. Taking it a step further and trying out the physical application on the DeepRacer cars, and seeing how your progress translates into the real world, on a real track, watching the car make the correct moves, gives a true sense of achievement
This will be the first time Ross will participate in championships in person. He says he hasn’t done physical racing until very recently but is catching up quickly so we might be seeing some competitive entries from him.
April Pro League qualifiers
Darren Broderick has shown up as an enterprise wildcard racer on behalf of Liberty Mutual in the 2019 finals. Smashed that and has been racing individually since July 2019. He has been one of the pillars of the Community ever since.
I don’t think his results in virtual league reflect his racing abilities. First, Darren always qualifies quickly and then races casually, second, his physical models have built a reputation. One of them is being used for testing and at demos by our friends at AWS. Interestingly however, a little birdie told me that it could not complete the 2022 championship track. Curious to see what Darren does about that. Outside of racing, Darren has been performing support in the AWS DeepRacer Community, running commentary at Summits in Europe and is a great person. Has clocked time in 53 public races to which he has made almost 60,000 submissions. Here is his self-introduction:
I’ve been with DeepRacer since June 2019 and was lucky enough to win a wildcard ticket that year for the Championships, since then I’ve worked on improving the virtual aspect but looking forward to racing physical again. When I’m not training models I’m usually training for Touch rugby or running in general at Park Runs. I’m based in Northern Ireland and work for Liberty Information Technology, an IT department of Liberty Mutual Insurance whom I’ll be representing at re:Invent.
I work a lot with Typescript & React JS on web development using AWS services to support our blue/green deployments for renters insurance.
This year has brought many interesting events. I was able to play Touch Rugby for Ireland in August 2022 at the European Championships. At home we re-homed a cocker spaniel called Daisy who’s been giving us plenty of challenges everyday.
I would like to say my racing style has become more consistent and stable rather than a “Go For Broke” approach but I know there’s still a little of that in me and we’ll likely see it at Vegas. I’ve also been trying to help out with object avoidance testing as well, virtually most racers have got a good handle on it, but physically it’s still been very challenging.
It has been great to be able to go to a few summits this year, both as a racer and as a community expert, I hope I’ve been able to help a lot of people who had questions on the day of those events and that they’ve continued to race.
I’m trying to bring a mix of models to Vegas, this years track is the biggest yet so it’s very hard to see what way my models will perform on the day, looking forward to getting a quick practice session or try a few different models in round 1.
In Vegas I’m hoping to connect with new and old racers, enjoy the whole conference experience since my first and last re:invent was in 2019, help out AWS staff with feedback for this year and possibly next year and make it through the round one at least.
Darren forgot to mention that Ireland has finished second in the touch rugby championships.
Daryl Jezierski works for Disney in Florida and flies planes as a hobby. He started his DeepRacer adventure in 2019 and has raced in both virtual championships. While this will be the first championship in person for Daryl, he has experience with physical races thanks to a mixture of underground races, summits and Disney’s internal league. He has submitted models almost 200,000 times across 65 public races.
I’ve asked Daryl to introduce himself and share his experience with DeepRacer:
I’m a Lead Site Reliability Engineer for The Walt Disney Company based in Orlando, Florida, USA. When I’m not participating in AWS DeepRacer, you can find me flying airplanes, going to Disney World, playing around with my 3D printer, and golfing.
I started participating in AWS DeepRacer in 2019. After a win at an internal company event, I traveled to the Toronto AWS Summit where I got 5th place. For the 2020 season, I qualified in April and made it to the top 16 in the championships. For the 2021 season, I qualified in April in the virtual circuit and got 1st place at the Atlanta AWS Summit.
Prior to DeepRacer I didn’t know Python at all. DeepRacer has been a fun way to up-skill and learn python. The skills that I have picked up with AWS DeepRacer have directly contributed to my career growth.
For Vegas, I hope to make it to the Tuesday races. Championship prep has included training a lot of models and hanging out at Epcot.
Karl Stiegler has been dominating the league in 2019. To most of us he seemed to be showing up casually, winning what there was to win and then his primary focus would be to keep challenging JJ’s position on the podium. Karl is not insanely active in the community Slack putting a lot of focus on his family life and management work at National Australia Bank, but he’s around and is always happy to share his insights and promote the races.
Here are a few words from Karl:
Here’s a bit about me…. I’m Karl Stiegler. Home is Melbourne, Australia. I love to travel, spend time with my wife and kids, hit the gym and lift weights, and occasionally (or sometimes more than occasionally) do DeepRacer.
I’m an Engineering Manager in DevOps and Automation at National Australia Bank, in the bank’s Corporate and Institutional Banking division. My team and I build and maintain the CI/CD, testing and monitoring automation and supporting infrastructure for a front-to-back trading platform for fixed income, interest rate derivatives and credit derivatives. Most people would probably think that sounds boring (my wife certainly does) but I find it incredibly fascinating.
Since I started dabbling in DeepRacer I’ve always specialized in the Virtual League races, mainly because every time I’ve tried my hand at the racing the physical car it’s been an unmitigated disaster. Though this time around I have spent a bit more time trying to get a better understanding of the physical car. Whether that pays off is anyone’s guess, and wouldn’t assume that I’m going to do well and the championships this year, but doing about average would be a step up for me!
In 2019 physical racing has indeed proven challenging to Karl but since then he’s organised and mentored an internal race at NAB and has had opportunities to improve on it so I’m curious to see what he’ll show on the track at the re:Invent.
May Pro League qualifiers
Tobey Gossen. He started racing thanks to a company DeepRacer event. Joined the league in October last year and immediately won the Open Division and then qualified into the 2021 championships through an online wildcards race finishing eighth overall. This year Tobey has been present in races every month, qualifying swiftly into the Pro Division, and in May he made it into the finale race which he has won.
Here’s Tobey’s great write-up about him and his racing:
I am based out of Orlando, FL – vacation spot for the rest of the world it seems like. I am a curious technologist! If it has a config file, buttons, knobs, or switches – I can hear it calling my name. “Assembly Required” is my favorite bullet item on a box.
My hobby of the moment tends to be whatever interests me – or whatever is broken. I like to dig in, figure out how it works and get it back online/operational again.
Professionally, I do about the same thing – just in a large enterprise setting. My primary role is to engineer / architect solutions that span business units and cross multiple technology silos. I tend to form ad-hoc teams to design and then prototype the solution. Once the solution appears viable, I’ll lead the rollouts and provide guidance to the implementation folks. I love what I do and the people who I work with.
DeepRacer popped up on my radar back in 2018 with a one-shot onsite event hosted by AWS at the invite of one of our internal teams. I thought it was interesting. Not really understanding all the jargon, I ended up scoring in the top 5 for the company. The same event was hosted again in 2019 – this time I did even better – got 3rd! When COVID hit, all the fun things were postponed, and I did not revisit DeepRacer until the tail end of the season in 2021 – that was when I realized that this was a much bigger event than just some once-a-year exercise.
2022 season, I spruced up my personal AWS account, added some funds to cover the training and off to the races! It was May of 2022 that I placed high enough to be selected to appear in 2022 re:Invent.
My racing has evolved a lot from when I started. Like most folks, you start with the default / examples and gingerly poke at a few variables. But, as I learned more and burned CPU time, I started to change my strategy. Hyperparameters were tweaked, reward file is hardly written to disk before I am making another change. I can write an action space freehand and it sometimes passes as valid json after the 3rd or 4th try! The biggest change for me has been getting stability out of the models and trusting the machine learning to do the heavy lifting. I’ll do odd-ball experiments – like train Object Avoidance config for TimeTrial – just to see what breaks.
DeepRacer has been a huge positive to my career! It has elevated my visibility and allowed me to branch into other areas. I can hold meaningful conversations with actual professional machine learning engineers. My python skills have gotten better (not good by any means, but better). I value CPU core counts and base my CPU purchasing decisions on getting the most cores for the money. And kudos to my wife who lovingly and patiently will let me prattle on about my latest DeepRacer exploits.
My prep for the Championship has been sporadic. I keep thinking have just the right one, only to discover the model is completely trashed or not good enough. I have had to reset and re-think my approach. The transition from Virtual to Physical is more challenging than I thought. Things that worked in Virtual do not work in Physical.
I am looking forward to RaceDay. As a Championship Rookie – I have no delusions of taking first. I just want to pass the first round. If I make it to the second round – that would be huge win in my book. I will have the best model I can make by RaceDay. No matter how my 2022 re:Invent debut turns out, I know that there will be more config files, buttons, knobs and switches that will make me stronger contender for 2023!
Nathan Liang will be racing in the second championships after last year’s qualification. He’s participated in 30 public virtual races so far and his best results were the first place in the August finale of 2021 and then second in May finale 2022, both giving him the tickets.
Here’s what Nathan has shared with us:
I used to work at ApolloMed as a software engineer intern creating automation scripts with Python. I really enjoyed the real work experience (albeit I worked from home because of the pandemic), but now I am prioritizing school. It’s been several hectic weeks for me preparing for the championship and managing my study work. I just started college this fall here in Southern California, and I’m currently working on a few machine learning projects in addition to DeepRacer as side hobbies.
As for my prep, I’m trying out a few different strategies, namely tuning the action space speed and training time. I’m also trying out a few other things that I can’t say since they’re a secret . I think I am pretty prepared, but let’s not get too cocky LOL. I genuinely look forward to meeting everyone and chatting in Las Vegas!
I haven’t seen much of Nathan’s physical racing but his name has been mentioned in talks about who might deliver on the tracks. We shall see!
James Faeldon is a returner to the championships but this is the first time he’ll get to race on a physical track. He lives in Philippines and works in analytics.
He has shared what racing was like this year and what hopes he has before the championship races:
This year, I think I focused more on training strategies and optimization. I don’t tinker much anymore on reward functions. I believe I have a deeper understanding of what works but I do feel there’s a long way to go if I compare my performance to some of the top racers.
I was able to share my DeepRacer experience at work and a lot of my colleagues are fascinated about it. Hopefully, I could inspire a few to compete and join the community as well.
This is my first time doing any physical race. I didn’t do any special prep to be honest. Just that I’m hoping I could learn a lot from this experience to maybe someday master physical racing as well.
June Pro League qualifiers
I’ve met Matt Camp at the AWS Summit in London in 2019 where he won the first prize. What you might not know (or remember), the first championship race happened in 2018 at re:Invent and Matt was there too! I have asked him for a self-introduction, here it is:
I’m originally from New Zealand, lived in the UK a long time but now live in France. I’m a Platform Engineering Team Lead and I spend most of my free time renovating a large old house (which sadly doesn’t have a DeepRacer track…. yet)
In terms of my racing evolution, I was part of the original 2018 championship winning team, 2019 and 2022 London summit winner, 2022 June pro virtual league champion, F1 Pro-Am finalist, and I sometimes help run the DeepRacer Underground events. I’ve been deeply involved in the open-source local training efforts and am now more focussed on how to train models for physical racing.
After taking some time away from DeepRacer it’s been great to get back into it this year, both from a technical interest but also reconnecting with the community.
My championship prep is best described as “last minute panic”. My hopes aren’t high, but I never go into events expecting much and have been pleasantly surprised a few times so we will wait and see.
I’m curious to see how much speed panic can generate at the last minute!
To make it even more interesting, the second qualifier of June is Tony Jenness, Matt’s friend. I also have met him in 2019 at the London Summit where he finished fourth. He’s been racing systematically ever since.
Tony is currently based in Auckland, New Zealand where he bought a house. He’s a Lead Engineer at Kiwibank, currently doing lots of Cloud engineering and automation as they continue on their Cloud adoption journey.
He says his racing is based on a proven strategy which he slowly improves based off a lot of testing.
The primary outcome from Tony’s racing this year is, as he claims, sleep depravation due to monthly finals happening in the middle of the night.
The championship preparation has been tricky for Tony with limited track time therefore his ambition is to not finish last.
Jochem Lugtenburg has been racing from the beginning, even before the Community was started. His first tournament was the AWS Summit in Amsterdam in April 2019 where he learned many useful lessons on how to (and not to) race.
He has participated in championships in 2019, 2020, 2021, he has been leading the community, creating projects, helping with events, taking part in meetups, troubleshooting issues, learning, teaching, recording videos for the community video channel – a great person to have on a team.
This year he has taken a little break from virtual racing after qualifying into the championships but that definitely does not mean he has not been preparing. He has shared some information about himself:
My name is Jochem Lugtenburg, based around Rotterdam, the Netherlands.
I work as a Data Engineer at Relive.cc setting up data pipelines using Kinesis, AWS Glue, AWS Redshift and DBT . My hobbies are Deepracing, gaming and going to the gym every now and then.
I didn’t do a lot of racing this year, my main goal was qualifying in the virtual league. I did that using similar strategies as last year. My goal is always to focus on stability over speed as I believe that gives you an advantage in object avoidance.
I haven’t raced using a physical car for a very long time. I’m focused on trying to figure out what the optimal speed is for a physical model, but so far my results are not that great. I’m afraid my models might be overfit and I don’t expect a lot from them. It would be nice to complete at least a lap using one of the models I’ve trained.
Jochem always presents curiosity and willingness to learn so expect to see a lot of him around the tracks where he will likely be trying out more and more models.
July Pro League qualifiers
Andy Miluzzi has started racing in late 2020 and since then he usually remained in the top 20 of the final standings. Very solid results in 2021 were accompanied by a few fourth places in Pro Finale races which meant Andy wasn’t qualifying into the championships but he managed to win his spot through a third place in the wildcard race. This year Andy qualifies in July when he won his Finale race.
Here’s some info about Andy:
I’m based in Orlando FL and I’m a Senior Ride Control Systems Engineer for Walt Disney Imagineering. Besides DeepRacer, I’m passionate about mentoring FIRST Robotics teams and I’m also an amateur radio operator, KK4LWR.
My racing has largely been academic and a learning experience. I’m always trying new things and tweaking things. I’ve learned a lot about what works for DR.
My hopes for Vegas are to race on Tuesday. I’ve been training nonstop and built a track to race on. My wife isn’t the biggest fan, by my dachshund, Daisy, loves it! She does laps all day.
David Mutton-Hughes has been racing since October 2019 which makes him one of the most experienced racers at JPMC. He has worked in technology for JPMC for 29 years now! Aside from DeepRacer his hobbies and interests include classical music and walking in the countryside.
Despite the experience, David has been often narrowly missing his spots in the championships until this year. David is the author of a very impressive DeepRacer log analysis tool, AWS DeepRacer Guru. Developing it has helped him evolve his strategies but this year he put more effort into racing and using his tool and this brought effect in July.
David hopes to have a model that can go round the track in Las Vegas. I can understand why he’s toning down his ambitions, the track is really tricky. But a good piece of advice from me would be not to underestimate his experience.
If you have been racing for a while chances are pretty solid that Lars Lorenz Ludvigsen (or LarsLL) has greatly supported your autonomous race car career. He is the owner and maintainer of deepracer-for-cloud project which allows you to train models in a local environment or on cloud infrastructure of your choosing. He has been racing, supporting technically other racers, researching the physical car and its limitations, benchmarking helping detect issues with DeepRacer service. He has made DeepRacer available to many many racers around the world.
Lars has started racing in late 2019 and has been present in 2020 (ending second!) and 2021 championships online. This will be the first time Lars will race in Las Vegas. He has been active in the virtual races league where his mean rank in 70 races is 14th. To give you an idea on his improvements, in the most recent 23 races he has been 10th or better, with emphasis on the better bit. Lars has also introduced himself:
I am a 43 year old Norwegian that have grown roots in Bavaria, Germany. I am a Mathematician by training and live with my wife and 3 kids. Hobbies include many techy things, but the last few years a lot of time has gone in to DeepRacer.
I work at Accenture, where I lead client project teams for a large client. I also co-organize our internal DeepRacer activities.
This year AWS Summit in Berlin was good. First time I had the chance to race a car on the track for real, and it was even better as I managed to bring home the trophy.
My virtual racing has been rather stable throughout the year; as always took a little while before I managed to secure my ticket to Vegas but after a series of close-calls I got there in the end.
Before the championships there has been a lot to do. The expectations of physical racing meant there was a lot of new things to look into. I have spent quite some time this year trying to understand the differences between simulation and real-life. This has included pimping my DR with high-performance parts, understanding and altering the on-car ROS modules, and adapting the AWS RoboMaker / Gazebo environment to more realistically represent the physical world!
There has not been too many opportunities to train on a track as advanced as the championship track, so there has been a bit too little feedback in the training to know what works and what not.
I have a lot of trained models for the finals, we’ll see if any of them do any good. It can be top, it can be bottom. I don’t know. I would be disappointed if I do not survive the first day.
It will be great to meet Lars in person. He is on my shortlist of racers to make it into top 8. I will be very curious to see how his experiments and research pay off.
August Pro League qualifiers
Jason has been racing staying in the community since 2019. 2022 is the second time he’s successfully punched the ticket to Vegas. Here’s some information Jason has shared about him:
I live in the Washington DC area. I started with DeepRacer after seeing it at the Washington DC Summit around summer 2019. My original interest in DeepRacer centered around the “sim2real” aspects. How effectively can one train in a simulator, and then translate that to real world performance. I spent quite a lot of time researching that aspect initially, but had to shift gears once everything moved to virtual racing as a result of COVID.
In 2020 I made it to the finals after winning one of the “virtual summit” races, but couldn’t make it out of North America group. I fell just short of the finals in 2021, including a couple of close calls in the Object Avoidance qualifiers. For 2022, I finished 2nd in the DC Summit, but then went on to win the August Object Avoidance race which put me into the finals.
I have already met my goal for the year, which was to make it to the finals. Any advancements beyond this point is bonus. For me, the finals is more about an opportunity to meet several other DeepRacer Community members I communicate with regularly through slack, but have never met in person.
Over four years of racing Jason has made 70,000 submissions to 55 races of which he’s won two (the ones that got him into the championships).
Deepak Kadian is one of the longer-standing Community members. He’s ben racing since 2019 and has taken part in the 2020 championships. In 2021 I remember how he was inching closer and closer to eventually reach a Pro finale in October and finishing 12th in it. This year he’s been present in all Pro races again but also managed to jump onto the podium in August and so I have asked him for a self-introduction to share with us:
I am based out of Haryana, India. Inner Engineering is my life line. I have recently started doing Angamardana (a form of Hatha Yoga). I am an Earth buddy, volunteer for #SaveSoil. I read manga, watch movies, play video games etc.
I am working as a Senior Software Engineer with Info Edge India (99acres.com). I am working in Front end with React JS and the focus of my work is improving page performance and core web vitals, while improving system stability, fixing live issues etc.
This year I was hoping to qualify for re:Invent, but didn’t have a plan in-case that actually happened. Also I had my first live interview with Blaine, that was fun.
Earlier I was just doing many experiments in trying to figure out how to improve my time. But now I feel that I have to finally read / research more about everything involved in DeepRacer and racing in general to see further improvements. I don’t even know the various terms related to racing like the ones Blaine or Ryan uses in their commentary.
My excitement has reduced a little, as I won’t be present in Vegas. So I got a little lazy in preparing for it . Need to review what I have done well so far for sim to real models.
I just want to make it to the bracket of 32, anything beyond that will be a bonus. I am thinking about having some setup for physical races for practice for next year.
As Deepak has mentioned, he isn’t able to come to Las Vegas so his models will be raced by a proxy racer.
Sorin has joined the league in 2021. He quickly qualified into the Pro Division and started making his way up the leaderboard. Despite four attempts in the finale he was just short of qualifying for the championships. In 2022 he was presenting similar results in the races but this time he has managed to jump onto the podium in August.
I am not sure if he had a chance to race on a physical track in person but he did participate in the DeepRacer Underground where a physical race is live-streamed and racers submit their models. It will be interesting to see how he does in Las Vegas.
September Pro League qualifiers
Yi-Li-NYCU-CGI, ZhengYi-NYCU-CGI, Yoway-NYCU-CGI
李頤 (Yi-Li), 李政毅 (ZhengYi) and 施囿維 (Yoway) are students from the Taiwan’s NYCU CGI (Computer Game Intelligence) lab where they study machine learning and reinforcement learning of many games such as board games, video games, DeepRacer etc.
National Yang Ming Chiao Tung University’s CGI lab is very well known in the DeepRacer world, primarily for introducing racers who know what they are doing. Every year they join the racing and present excellent results. YiLi, ZhenhYi and Yoway have started racing in 2021 so this is their second season in the league.
ZhengYi says DeepRacer is a very good competition, through which they learn a lot of skills about ML/RL, such as designing rewards such that model can work stable in the real world, and analyze different hyperparameters, etc.
To prepare for the championships they had the track printed and have done many tests on their models.
Unsurprisingly, they have quite some appetite for good results this year.
Students League qualifiers
AWS Student League racers have been using a separate DeepRacer console where they could participate in competitions without AWS accounts or any need to provide payment details. The format of the races was significantly limited to ensure students don’t take advantage of additional training outside of the console. All student had an equal amount of 10 hours to train with. They could explore strategies outside of the student tournament but the league itself only allowed for those 10 hours of training.
It so happened that the top two qualifiers are students lead by their teacher and an AWS ML Hero Cyrus Wong. Cyrus always uses the opportunity to make his students curious or to engage in various projects. Sometimes it was using lambda functions for training, sometimes it was tracing a race line of a physical car on the track, other times arcade-like steering of DeepRacer.
Mr. Lee Chi Kin (Tony) is the winner of AWS Student League qualifications this year.
He started racing thanks to an encouragement from Cyrus after which he joined the workshop by Cheuk Lam Yuen (Oscar) called the AWS DeepRacer Hong Kong Club. Tony says his fellow students also did well in the Hong Kong tournament including the champion title. Here is some more info from Tony:
I love IT and I am studying Higher Diploma in Cloud and Data Centre Administration in IVE Hong Kong. My course let me have a consolidated background in AWS and cloud tech. Since my AWS skills is good, I can pick up AWS DeepRacer easily.
Currently, I am working on my internship in a cloud vendor partner – SOS Group (Hong Kong) Ltd. I support and maintain cloud infrastructure. Sometimes, I am in charge of setting up the Backup and Disaster Recovery solution in the cloud.
We are using AWS Academy Learner Lab and local training to develop and test the reward function. And I think I am very lucky to become the winner in the qualifications as my model is not really stable but it managed to show outstanding performance in the final.
To improve on my racing I have been reviewing different blog posts and open source projects. It will be useful to learn from other experiences too. The opponents in the open session are very strong and often have more resources. I just have a few credits to training my model so I don’t expect to win but hope I can complete for a few laps. That will be fine with me.
Ms. Yu Nga Man (Wina) has won the second qualification through the Student League. Just like Tony, she doesn’t have any former public racing experience with DeepRacer outside of the Student League.
I am the student of Higher Diploma in Cloud and Data Centre Administration in IVE Hong Kong. I am interested data analytic and cloud tech before playing AWS DeepRacer. Now, I am also interested in machine learning.
Currently, I am working on my internship in a cloud vendor partner – LeadingEdge Technology in Hong Kong. I main works on BI solution and big data project. I sometimes use Spark, Kafka, and BI tools such as Tableau. My internship job is data engineer focus but I hope to become a Machine learning engineer after I graduate from IVE.
My course at IVE used AWS DeepRacer as an assignment and all classmates were required to get a good ranking in AWS DeepRacer Student league to get mark. I played AWS DeepRacer for my mark at the beginning and a unsuspectingly became a monthly winner, and got into the final.
I use many AWS Academy Learner Lab to build and test my model since my higher diploma programme uses AWS Academy Leaner lab for every individual course and sometimes, there is some credits left after the end of the course. I have used all of them to try and develop my reward function under the same constrain environment in AWS DeepRacer student race.
To be honest, I have no plan of preparation and I will try my best for the race, but I don’t have confidence to get any prize in the championships. I think it will be a great experience for me to get involved in the final and be able to attend AWS Re-invent 2022. To me, it is a great opportunity as it is too expensive to attend AWS re:Invent or travel to Las Vegas myself. I want to attend keynote and go to the re:play party, I cannot wait!
Summit Regionals EMEA
That’s me. My name is Tomasz Ptak, I live in Bedford, UK with the loveliest wife there is and two amazing children, 8 and 11. Originally I come from Ożarów, Poland. I am a Senior Software Developer at Duco where we build data automation services for mission critical data, and financial reconciliation solutions are my thing. Oh, and I’m an AWS Machine Learning Hero.
My hobbies include bread, DeepRacer and community. My core values are belonging and understanding.
I had my first contact with DeepRacer at the AWS re:Invent in 2018. I took part in the workshops and got my first DeepRacer. Next I returned to it before the AWS Summit in London where I finished third and have been racing since. I won one race in 2019 (which was an achievement with Karl-NAB grabbing all first places till then), took part in the 2019 championships where I bragged about being the fastest person not to qualify into round two. I took part in both virtual finals where I proudly ended my participation in the first round. This year I haven’t been racing much virtually but have secured a place through the Summit Regional Qualifier for EMEA. In the name of consistency I expect to nail Round One so well that I will not move on to Round Two.
In parallel I have joined a Slack group created by a great lad Lyndon Leggate after the 2019 summit. The group was initially called DeepRacer-London, then quickly stopped being a London group and became the AWS DeepRacer Community. I was one of those supporting the community form the beginning, and then took over together with a group of passionate, enthusiastic and curious racers. The nice thing about DeepRacer is that you provide an environment and feed it reward and compute and it gets better. The nice thing about the community is that you provide it an environment and feed it curiosity and challenges and it gets better. I still find it fascinating how we reached 43 thousand members who have at least touched some Reinforcement Learning. Even more if you count the Students League which had their own space.
This year I have focused primarily on work so there was little time for racing, but I have thoroughly enjoyed helping with DeepRacer Underground and coming to summits in London and Toronto. In virtual racing while I have relied on my old strategies, I have added some relatively successful experiments with shaping rewards. For physical racing I have to focus on simplicity and short training as I think these are essential to prevent overfitting.
To Vegas I will bring with me three models one to fail during the training phase, one to fail during the first run and one to fail during the second run of round one. I do however think that the best part of re:Invent will be finally meeting the members of Community and the AWS DeepRacer Pit Crew.
Joel Larsson has been participating in virtual racing since April but has won his place in the championships through the Stockholm AWS Summit in May 2022. I have asked him to tell a little bit about himself:
I’m 33 years, based in Stockholm, Sweden. Living with my fiancée and our dog. I’m working as Product Manager of machine learning and analytics platforms at Telia (Nordic Telco).
Since I first heard about DeepRacer in April 2022 I’d say the learning curve has been quite steep, a lot due to the Community from which I have learnt a lot. I really put an effort into researching back then, which have proven to pay off. Since then I’ve mostly worked on improving my virtual racing, so it will be really interesting to race physically again.
DeepRacer has given me so much recognition throughout the year. It’s been such a great platform for me personally to show hands-on skills in ML and it’s also been a great way for us at Telia to brand our company. It has really taken me way beyond what I could imagine.
My championships preparation is going quite OK, but still need to do a final push to see if I can optimize some more. Before Vegas I’m attending a competition in Stockholm between Ericsson, ABB and Telia where I plan to get some more physical experience before the race.
I never thought I would actually get this far, so my hopes for results aren’t that high this time either. But somehow I managed to create my best models just before the earlier races, so it ain’t over until it’s over!
I have met Adam Pye at the London Meetup in 2019 (remember those events where people were getting together in person to do things? I miss them too). He has been an enterprise wildcard racer for JPMC that year at the re:Invent and was one of the involved in plans which have materialised into strong representation of JPMC racers in the AWS DeepRacer League. Here’s what Adam has to say about himself:
I am based in Stoke-on-Trent in the West Midlands, Hobbies – DeepRacer (obviously!), I also play Badminton, D&D and volunteer in a local theatre. Work – I lead the Public Cloud platform SRE team at JP Morgan Chase and also help to run our internal ML training program using DeepRacer to teach developers about AI/ML, AWS and Python.
I have recently spent a week teaching autistic high school students about DeepRacer – they loved it, I hope to see some future champions there 🙂
For the championship I’ve trained a bunch of models which I’ve been testing on our re:Invent 2018 track we have in the office in London. Hopefully one of them will make it round the actual 2022 track! For Vegas i just hope to make it through round 1 currently!
This year Adam has qualified through the AWS Summit in London. He finished third on the day but because MattCamp had secured his championship place through virtual league, we have met in a regional qualifier for the EMEA region. I am very interested to see how Adam does. In 2019 he joined the finalists pool pretty late in the year and thus showing up was pretty much all he could do. This year with a little bit more time he might show some decent racing.
Summit Regionals Asia Pacific
I know very little kashi. He has raced since April, first in the Open Division and then in Pro but without qualifications into finales. That said, kashi has won a race at a Japanese Summit completing a lap in 9.859s! This way he made it into Asia Pacific finale and qualified from the first place. You can watch his entry at the time mark of 3:14:45.
Benny Li is based in Guangzhou in China and is an IT manager at HSBC Technology Center. He likes programming and while he doesn’t necessarily touch code in his daily job, he truly enjoys seeking opportunities to improve his work through automation.
This year he has participated in many DeepRacer events in China where he met many new friends while having fun. His adventure with DeepRacer has begun a little earlier – around August 2021 during a corporate event. In September he finished 316th in the Open Division and has continued racing through the pre-season, ending 21st in February and competing in the Pro Division throughout this year.
DeepRacer has inspired Benny to learn more about machine learning but also about racing, understanding the optimal track path and more.
Interestingly, the regional qualifier has caught Benny a little by surprise. In under 48 hours he has compiled all his learnings into a training strategy that delivered a successful model. Well done!
Benny had big hopes to come to race in Las Vegas but it turned out to be impossible this year. Therefore he will participate through a proxy racer.
Yaojin Liang is also based in Guangzhou in China and works at the HSBC Technology Center. He loves football.
He’s been racing virtually since late 2021. In March 2022 he has won the Open Division and has since been racing in the Pro Division. Nicky has participated in Chinese DeepRacer League and also at an HSBC Global DeepRacer event where he finished fourth in China and will participate in the final in March 2023.
Sadly he won’t be able to come to Las Vegas so he will use the services of a proxy racer. He says his model is already prepared.
Summit Regionals Americas
Jacky Chen’s racing records date back to May 2021 when he quickly qualified to the Pro Division and then raced pretty well in among the Pros. This year he’s eve been in two Pro finales but his gradual improvement was interrupted by his mind-blowing performance in the Regional qualifier where he has become the only person to break a barrier of 8 seconds per lap in live virtual racing on the 2022 Summit track, going as low as 7.525 seconds.
Here is a self-introduction from Jacky:
I am Canadian, now live in Los Angeles (City of Pasadena to be exact). I like technology (in IT industry for 30 years) and also like to travel, next stop will be UK & Europe Photo – travel with my kids for their college visit.
I am Master Cloud Architect at Oracle, helping enterprise customers in the USA to use Oracle Linux & virtualization technologies.
I am volunteer coach training high school and college students to learn AI & Leadership and I use DeepRacer in the process. I like the AWS DeepRacer Global League (it is always fun to race). I am very happy to win in the Toronto AWS Summit and later broke the world record of online racing in the 2022 Summit Track.
At the beginning, my students actually did better. These high school and college students spent lots of time practicing and training the models. I was busy at work but was able to spend extra time training the models and got improvement.
Reinforcement Learning (RL) is a great way to learn AI/ML. We (as the coaches) actually understand better about how humans learn, by teaching students on RL. Learning by doing / by teaching / by gaming in the DeepRacer platform is wonderful.
For Vegas we are preparing a couple models for different racing strategies. Still testing. This championship track is very challenging. I’ll just try my best and enjoy the time of racing and meeting with so many good racers face to face.
I had a pleasure of meeting Jacky in person at the Toronto Summit this year. I have heard opinions in the Community that he may be one of the finalists in the championships so I recommend tracking his results.
Roger Logan has been racing since 2019 but got more involved a little later.
He’s a Software Engineer in the financial services industry. He lives in Richmond, TX just outside of Houston with his wife and his 10 year old son, Zachary. DeepRacer is not his only competitive addiction – he has been playing pool in leagues and tournaments for the past 30 years and when not competing himself he enjoys watching Zach play baseball.
Roger has shared his experiences with DeepRacer:
Been racing since about March 2020. JPMorganChase started an internal league across 15+ cities all at once. Me and Tyler Wooten (a.k.a. JPMC-Driftking-Houston) were on the same team named “Driftkings”. We won Houston and then globally against the other cities. During that time, I started connecting with the wider DeepRacer community in Slack. Once I started using DeepRacer for Cloud, we got much better. I entered the pro division using my own account followed by Tyler with the original account which he renamed. Sairam and Yousef (Rogue and Ace) also joined the pros right after that. While I was able to do well in virtual TT races, I have struggled with OA and physical races. I learn a lot by trial and error and studying not just online but the open source code base, as well.
Doing horrible in OA after a great many failed models, gave me insight to what is required for a generalized model. TT models never change and can be overfit. Overfitting with random boxes and physical races can be disastrous due to the changing nature of the environment. After placing 3rd in 2 summits this year and doing horrible in OA, I feel I have learned a lot of what is needed for a generalized model.
My preparations are going well. I finally can get consistent complete laps physically which is something I have had trouble with in the past.
My hopes for Vegas? To win, of course.
I have known Martin Paradesi since 2019. He’s been active with racing at Capital One and also took part in the AWS DeepRacer Community Experts Bootcamp at re:Invent that year
While not very active with virtual racing, Martin has been around and has been spending more time on physical tracks.
Martin Paradesi (Source: Martin Paradesi)
Here are a few words from Martin:
My name is Martin Paradesi and I’m an Engineering Manager at Capital One. I obtained a Master’s degree in Computer Science with a specialization in Machine Learning.
I live with my lovely wife and wonderful kids in the New York City area. My hobbies are running, bicycling and swimming.
At Capital One we want to ensure our associates have the opportunity to learn and apply new skills. DeepRacer is a fun interactive way to have our associates start to learn this technology but more importantly the program is designed in a way that anyone, technical or not can learn and participate.
In 2019, Capital One had 82 teams participate in the Capital One AWS DeepRacer League, which included a 6 week virtual league culminating in a physical event, with the winning team being sent to AWS re:Invent to participate in the world finals.
We continue to have our DeepRacer league, and we partner with AWS to ensure associates receive training and have all of the resources needed to participate in the race. This year, 76% of our racers were new!
Last year, I worked on a cool open-source project called RoboCat on AWS DeepRacer in my spare time. It was featured on the AWS DeepRacer Robotics page and the AWS Innovators series. This year, I gave a talk about that project at the AWS Summit New York.
I started experimenting with AWS DeepRacer in 2019. I trained several reinforcement learning models and tested them at the AWS Summit New York. Some of my models’ performance was captured on the AWS DeepRacer TV New York episode. I have participated in the AWS DeepRacer Virtual League since it was launched. While I graduated to the Pro league quickly each year, I was unable to finish in the top 16 in the virtual racing format. This year, I took the initiative to improve the performance of my models, specifically in the virtual racing format. Thankfully, I was able to finish third in the Americas Qualifier.
I had the opportunity to race my reinforcement learning models outside New York for the first time this year. I had an opportunity to visit Paris earlier this year and test a few of my models at the AWS Summit Paris. It was nice to connect with the AWS DeepRacer community at that summit.
I’ve completed training reinforcement learning models for the championships. I now have to work on the autonomous driving part. I trained two models based on my experience in Paris and New York summits. As each finalist will get two attempts on the track, I hope to test a newer model in the first attempt and then switch to a model that will be similar to the one I used at the New York summit for the second attempt.
While I will give my best shot during the AWS DeepRacer League Championship Cup, I’m also excited to meet amazing reinforcement learning developers from all over the world. I look forward to learning and sharing knowledge with them.
I’m curious to see how Martin does in Vegas.
Summit Regionals Public Sector
Robin Castro has been racing virtually from the very start of the league in 2019. He qualified into the championship as one of the runner-ups (in 2019 the best racers across all months also raced in Vegas). He has taken part in 65 public races making over 96,000 submissions of models to them.
He has been one of the organisers of a league at DBS Bank in Singapore where he works leading an FX software development team.
This year Robin has been very active. He’s taken part in multiple virtual races and the ASEAN league. He also attended the AWS Public Sector Summit in Singapore where he finished second. This result has allowed him to race in a regional live finale and qualify to the championships.
He says that because of a late qualification he didn’t have much time to prepare. He has trained a few models and hopes one of them will win.
Yousuf Nizam is the AWS DeepRacer League 2021 runner-up. He has been racing since the 2020 season but had greatest success a year after. This year he has focused on physical racing.
Here are a few words from Yousuf:
I am based in Hyderabad and I am a software engineer at JPMC, involves working on building software that aids the company to procure virtual machines internally.
My hobbies include graphic designing and playing basketball.
The most interesting event this year was my visit to Singapore when I came to take part in the summit. While I loved meeting the community of DeepRacer in Singapore I thoroughly enjoyed the city as well ( was my first time visiting Singapore).
This year my racing has been majorly focused on physical tracks, so I have become more aware of overfitting and trying my best to get a model without actually “overfitting” it as opposed to what I always used to do in the virtual environment.
My prep for the championship is like a roller coaster, sometimes I feel i have the exact thing needed to win and the same thing doesn’t work the next day . Physical has been a major challenge and it’s very difficult to train a model that is stable and fast on a physical track. I am trying my best to get a model that would make me competitive, so far I don’t think I have a model that will make me confident that I am going to win. Only time will tell 🙂
I really hoped I could make it to Vegas
Sadly, Yousuf is one of the racers who cannot come to Vegas. His models will be raced by a proxy racer.
Daniel Morgan is a Community legend. He’s been around for as long as I can remember, started racing in August 2019 and has been racing ever since. He has always presented pretty solid results on track but never qualified into the championships until 2022.
Outside of racing Daniel has always been extremely helpful in the AWS DeepRacer Community, producing training resources, supporting racers on Slack (http://join.deepracing.io) and representing the community for the AWS DeepRacer Student League participants. On top of that he has raced at the Summits and provided commentary at the AWS Summit in Anaheim.
Here’s what Daniel has said about himself:
I live just outside of Washington DC, I’m a System Engineer for NPR and recently took a new position that has a focus on the cloud. Besides DeepRacer, I’m typically designing things for 3d printing, woodworking, or playing video games.
This year I became an AWS Community Builder thanks to my contribution to the DeepRacer League and DeepRacer Student league.
For racing I took a step back from my previous approaches. I spent time on track analysis using leading racers as a baseline. I spent time learning how hyperparameters affect the model and challenged myself by using continuous action space. This led me to two particular reward functions that only learn with specific hyperparameters. Which has led me to further refine these settings.
Outside of racing I spent time building other AI demos using Stable diffusion. Which had a challenge in itself because of GPU limits to generate 1080p images you need in the upper limits of 32GB of VRAM, so I used a video restoration technique to upscale the images generated.
My championships prep is going well, I have a new rig to train on in addition so I have several models training in parallel. I hope to learn a lot while in Vegas. This will be my second time doing physical races so not being last would be my goal right now.
Summit Regionals Australia New Zealand
Adam Niu started racing early this year. He has taken part in one Open Division race in March and then spent the rest of the year getting used to the Pro Division and presenting varied results there. In the meantime he qualified into Summit finale race through an Royal Melbourne Institute of Technology event and claimed his victory punching a ticket for the championships and AWS re:Invent. He has shared a few things about himself and his racing.
I’m based in Melbourne. I’m studying at RMIT (Royal Melbourne Institute of Technology) and interning at NAB (National Australia Bank). I study Electronics and Computer Systems Engineering. I’m a big soccer and AFL fan. Looking forward to the world cup. I like hiking, reading, gaming, and traveling.
I qualified for the regional qualifier at an RMIT DeepRacer event at the F1 in Melbourne. I also attended the AWS Summit in Sydney for a DeepRacer event.
Throughout the last few months, I’ve learnt a lot about DeepRacer from online guides and contestants at events. In transitioning from virtual to physical racing, I’ve had to make a lot of changes to my approach. Virtually, my models are based on racing lines, and are trained over a long period of time to be good at a specific track. For real-life races, I am training much less to avoid over-fitting, as well as trying to expose the model to as many different situations as possible.
DeepRacer has been a very rewarding experience. Through DeepRacer, I’ve had the opportunity to attend many events and meet lots of interesting people.
I have been training my championship models virtually, and have been to the AWS office in Melbourne to test them on a physical track. I don’t have particularly high hopes, as I have not had much experience at all on physical tracks. My goal for the year was to simply make it to Vegas! I’m hoping to make it to the knockout stage of the tournament.
Yee Vien Ng is a bit of a mystery racer! The first public race records that I have on her are for a September Open Division race and then for the October Pro Division where she ended in an impressive 67th place (and trust me, getting into the top half of the leaderboard on first approach is an achievement). Yee Vien’s company, Quantium and Telstra Helix (joint venture) organised a competition in early September, it was the first time she heard about DeepRacer, she joined the competition and eventually won the first prize with her teammates!
Here’s a great self-introduction from Yee Vien:
Yes my name is Yee Vien , no middle name. Haha. I grew up in Malaysia, came to Melbourne, Australia in 2009 to study Mathematics, and eventually got a job here working as a Data Analyst/Modelling Analyst. You can say I’ve been in the data science industry for 10 years now, but because the technology keeps improving, I’m still learning about Machine Learning.
Before joining Quantium, I used to work in water industry and engineering consulting firm which mainly involves building optimisation and simulation models, perform data analytics, data mining and management, they’re quite engineering heavy. I’ve also worked on a couple of Machine learning and Deep learning problems, eg: defect detection in sewer pipe, predict pipe failure, those information can be found in my linkedin page. I only joined Quantium this year in January so I haven’t had a lot to say yet, but it’s quite a different industry to my previous roles, Quantium focus a lot on customer analytics, ie conducting analysis on our partner’s dataset and identify customer purchasing behaviours to generate insights and provide recommendation to the clients, and my client is Telstra (Telecommunication Australia).
Outside of work, I love to run, and sometimes do cycling or climbing. I’ve done 15 marathons and 1 ultra-marathon so far, and I’m planning to run the 90km Comrades Marathon next year. You probably realised by now, yes I love to participate, but I’m not competitive :D. I also enjoyed learning new languages. I’m currently preparing for the Japanese exam (N1 level) which will be held on the 4th Dec, yep, it’s on the same week as RE:Invent! (ps: I registered for the exam way earlier before I knew about DeepRacer, I did not plan to go to Las Vegas at all, but since I won the ticket, I should go.). Apart from that, I do like to code and build models during my spare time (only if I have time!), I love to build something that I don’t get the chance to do in my work, such as chatbots, augmented and virtual reality. Once I tried to build my own apartment and immerse myself into it.
As for my DeepRacer journey, it’s pretty short but amazing I’d say. I only got to know about the DeepRacer when my company, Quantium and Telstra are hosting a competition internally in Sept 22, and our team won the first prize. What’s interesting is that 3 of us never met or work together at all, I just ask to pair up randomly as I don’t have a team and that’s how we got in as a team, but I was lucky that one of the team member (Benil, he) has some experience in DeepRacer and he pretty much help us out throughout the competition. At that time he insists to use my model because it looks stable (completed 5/5 in the evaluation) and it should perform well in the physical race. Having no experience or what so ever, I wasn’t sure if my model will work, and it’s not the fastest in the sim, but I trusted him, and I’m glad everything worked out in the end.
After winning the Telstra DeepRacer event, I thought this is the end, but one of the guys from AWS team told me about the upcoming Australian championship race and the ANZ regional race, and if I got into top 3 in both events, AWS will fly me to Las Vegas. It got me motivated, so I decided to try out and see how far I can go. I’m also tempted to win the DeepRacer car (in fact more so than the Las Vegas trip), and I thought if I can get to top 3 in any competitions, I can get the car. But after 3 races, no car so far .
Anyway, during the Aus Championship race, I just use the same model that won the Telstra race, I got 2nd place and qualified for the ANZ regional race. Prepping for regional race is where I learned most about the DeepRacer, because the race is in virtual, which I have almost zlitch experience, I have to do a lot of research to try and improve my time, and from there I learned about the DeepRacer league, the community slack channel and log analysis (thank you!). I also met Karl from NAB at the Aus championship race and he gave me a lot of useful advise for the virtual race. While prepping for the regional race, I follow closely on other regional race, and quickly realised that my best model is not very fast, so I did quite a lot of practice prior to the main event to try and work out the best settings in each corner of the track, and eventually got the best timing out of my model and earned me a ticket to the Las Vegas.
As for prepping for the Las Vegas event, the AWS team has been kind enough to provide me more credits to train my model on the console and we have a practice session at the AWS building just last Monday on the Summit Speedway 22 track, so I took this opportunity to build models with different approaches, most of them which I’ve never tried it before, and test it out on the track. Even though they’re completely different track, it still gave me a lot of insights, some very important ones. I also had a chat with one of the Executive from Quantium who specialises in Machine Learning and he gave me some ideas on how to tackle this new track. Although it’s difficult to know whether my approach will work in the physical track , fingers crossed! In addition to that, Thireindar Min (Tboxboxbox in DeepRacer) and Karl are happy to borrow their DeepRacer car to me, while I don’t think I will have time to build the track to test it out, it’s still important for me to understand how the car works, the mechanics behind and how to calibrate it. So I’m really grateful to have this amount of resources to help me prepare for the race.
Coming to Las Vegas, I hope I can get as far as I can so I will have more story to tell to my peers , the further I go the better, but even if I drop out from the first round, I’d make sure to have plenty of fun, get to know other participants and learn from each other, and of course wonder around Las Vegas! It’s been an amazing journey and I’ll definitely miss it when this is over!
Yee Vien has confirmed that she’s only been racing for 1.5 months now and started at the company event. That said, she has worked on object recognition, machine learning, optimisation and simulations before and though she has never worked on reinforcement learning, the concept is almost the same. She has done some solid preparation and I think I will be following her results at re:Invent as she may present well on the track.
Ethan Phan doesn’t have much virtual racing experience to show having raced (and won) in the Open Division in September and reaching 37th place in October’s Pro Division, but then how many racers have reached top 40 in the Pro Division in their second race on record? To make this intro complete let’s just add that he qualified into the Australia & New Zealand regional qualifier and then ended it on the podium which gives him the ticket to Vegas.
Here are a few words from Ethan:
I’m currently working as a data engineer at Rio Tinto, based in Brisbane, Queensland, Australia. I don’t have a lot of hobbies, but I know now DeepRacer is one of them.
I really like working using AWS services, especially serverless technologies. Understanding and developing data, Devops pipelines in AWS environment are my main daily duties. About Rio Tinto, I really love working here since I get to work in a community with energetic vibe and always love to adapt to new ideas, and not afraid of challenges.
I think ranking #1 in the September Open Qualifier was a huge event for me since I only started learning DeepRacer in September. And during this period, sometimes it was lonely but as I get to learn more about the communities, I realise that racers come from all over the world and it’s great to make new friends and learn new things. ( I get to know a high school student who beats me so many times, and he’s been a huge courage for me to improve myself).
My racing has evolved from using what it’s there in the examples, to using and understand tools like notebooks to analyse logs, produce heatmaps, etc.
I think DeepRacer is a great tool for anyone who loves to learn about AI/ML with intuitive UI. I get to introduce about DeepRacer to my colleagues and everyone has been pretty excited. Thanks to DeepRacer, I have great bonding time with my colleagues as well given that I was a new employee here.
I have couple of models that I want to test in the championships, it’s been going well, and I will try to do everything I can. I hope to get the AWS DeepRacer jacket haha. The AWS DeepRacer jacket was my initial motivation to join the game because it looks so good.
October Pro League qualifiers
I met Phil Jaquenoud at a London DeepRacer meetup not too long before re:Invent 2019. From what I remember he’s been almost hypnotised by log analysis while his models were zooming around the track like crazy. Those models helped him take the second place in a wildcard race in Las Vegas. Last minute qualification is Phil’s superpower. With the exception of the 2020 outlier he has always qualified through the last possible race.
Phil is based near London and is working as a Senior Director for AWS Product Engineering at Ensono – a managed service provider where he helps large enterprises with their IT estates, transformation and modernization. His hobbies include golf, snowboarding, piano, DeepRacer and DIY.
When asked about an interesting thing that happened to him this year, he said:
I celebrated 20 years of marriage to my amazing wife.
Out of 65 virtual races on record to which Phil has submitted models over 90,000 times, he has finished majority in top 10. He is also the only racer to have taken part in all 16 live finale races and only twice he was not in top 10 in those. He also added:
Having reached my 16th consecutive monthly finale without coming top 3 has tested my perseverance!
Phil has a solid record of improvement in the championships:
- 2019 – Reached the last 64
- 2020 – Reached the last 16
- 2021 – Reached the last 8
Therefore even though he claims his prep is going badly and he hasn’t built his garage track yet, he definitely isn’t someone to ignore in the championships. When asked about his hopes for Vegas, he said:
To have a few beers with the DeepRacer community, enjoy the racing, and meet some friends and colleagues in person for the first time.
Parnell White first experienced DeepRacing in December 2020 and has joined the 2021 season finishing in places 44-66 in the Pro Division that year. This year he has intensified the participation and improved greatly finishing always in top 30 and making it four times into the finales. In October he claimed second spot and this way we will meet in Las Vegas.
I have asked Parnell to introduce himself:
I’m in Ohio, USA where I’m a research engineer. My hobbies are DeepRacer (obviously), gardening, and some light game development.
DeepRacer has been a lot of fun this year, except when the car just refuses to learn how to take a hairpin turn.
My racing this year evolved in through a few stages. At first I set out to learn more about the vehicle dynamics to better inform action space design. Then I began to focus on better reward functions for time trials and increasing speeds. The last couple of months have been spent trying to get better at obstacle avoidance while still putting in just enough training to qualify in the time trials.
Championship prep is proceeding fairly well but I’m definitely not as far along as I’d like. Below is a summary view of my own internal analysis graphs after my last training restart.
Vegas will be my first physical race and I don’t have enough time to get as familiar with the actual car and the handling differences. I’m training in ways that should help generalization and I’m being conservative with speeds. Honestly I’ll be happy if I make one single lap without going off-track.
Jouni Luoma has been racing since 2019 when he officially was the first person training DeepRacer as a job role. He has been in the Community pretty much from the beginning, always curious and always helpful.
This year is the fourth time Jouni is taking part in the championships. He has agreed to share some information about himself:
I’m based in Tampere, Finland. My hobbies besides the nerdy stuff are music in all forms (playing, listening, going to concerts) and bicycle traveling / bikepacking.
I am working as Lead AI Engineer for company Silo AI. I develop solutions involving AI/ML components. Also working as a part time researcher in TurkuNLP group at the University of Turku.
This year I have not had that much time to put in to developing my DeepRacing skill and it shows. The competition is getting tougher every year, and this year I qualified from the last possible virtual race. The old tricks are not cutting it anymore, and I have had only limited time to dedicate to coming up with new ones.
Championship prep is now ongoing, but I have to admit it is pretty much flying in the dark. I have not trained models for physical cars in 3 years and have no place and time to test the trained models on a real track before re:Invent. I just have to trust ,y intuition, but I know from experience that those models may fail miserably on a real track.
My hopes for Las Vegas are to meet the people in DeepRacing community and have fun in the competition. I am also looking forward to participate in some sessions regarding AI/ML solutions development on AWS.
JPMC – Chris Griffin (JPMC-London-WackyRacers)
Chris is a software engineer in London, UK, working for JP Morgan Chase. He will be representing the JPMC-London-WackyRacers team who won the JPMC Global DeepRacer virtual final. While he has started racing virtually only this year, he has has some good results already, coming joint fourth in The Guinness Challenge virtual race for the largest machine learning competition participation.
Reinvent will be the first time Chris will race on a physical track and as things stand he fears he will struggle to complete a lap. Let’s hope it’s not the case. Chris says the best thing about DeepRacer to him was getting to know JPMC Pro racers: David Mutton-Hughes, Roger Logan and Adam Pye. Taking part in the championships strongly suggests that Chris should also consider himself a Pro racer. Hope he enjoys meeting also non-JPMC competitors in Vegas.
Accenture – Doug Wozniak (Doug-Racer)
Doug Wozniak is a Senior Manager at Accenture, specialising in serving clients in the utilities industry. He’s been racing since 2021 and participated in last year’s championships. Compared to last year he’s made great progress, qualifying into the Pro finale races six times.
He has kindly shared a few words about himself:
I am from Chicago, USA. I’m a proud father of two: ages 6 and 2, and they are my biggest fans for DeepRacer. They are also my biggest critics, pointing out every time my car drives off track!
This year I’ve made some enhancements to my reward function and overall strategy that have boosted my performance in the time trial and head-to-head formats, helping me qualify for each month’s finals since May. I’m hoping to continue improving in object avoidance and physical racing formats.
I love learning and I love competition. DeepRacer has been a great combination of both.
Championship prep is ongoing… I’m seeing good results in the simulated environment, but without access to a physical track it’s difficult to predict the results. It will be interesting to see how others perform on a new physical track.
In Las Vegas, I’m looking forward to networking with the top DeepRacer competitors in person.
Other enterprise wildcards
Truth be told there may be a few more enterprise wildcards. This is the amazing thing about DeepRacer being run by AWS – you never know for sure who is going to join. There may be some partners that we didn’t know of. And as always, I’m sure they will be great people that we’ll have lot’s of fun with at the tracks.
This is pretty amazing that we are having racers from so many places around the world joining the championship races. Many participants have not experienced what it’s like to gather around a track to race so this will be our time to enjoy the company in person.
But the stakes are high: $10000 for the first place is not a joke. Who do you think will count in the race for the prizes?
Check how you can follow the races and participate in them in https://blog.deepracing.io/2022/11/10/aws-deepracer-at-reinvent-2022-the-what-the-where-and-the-when/