Amazon DeepRacer Experience
My experience with the Amazon’s self driving autonomous model car.
Hi everyone 😎
Myself Kubera Prijesh Devanand, This is my first blog ever😊 in my life, Here I am writing this blog to share my experience in AWS DeepRacer.
AWS DeepRacer is an autonomous 1/18th scale race car designed to test Reinforced Learning (RL) models by racing on a physical track. Using cameras to view the track and a reinforcement model to control throttle and steering, the car shows how a model trained in a simulated environment can be transferred to the real-world.
AWS DeepRacer is an emerging technology, by which we can train a model in simulation. The trained model is then transferred to the Amazon Web Services (AWS)’s DeepRacer car to traverse a physical track.
First of all, I thank ashok bakthavathsalam, our beloved MD sir for gave me this wonderful opportunity.
To start with, I have not even heard about the AWS DeepRacer let alone know anything about it. But in July 2021, our MD sir called me and assigned me into this project. I just thinking that what’s going to happen — will I do it or not? …and several other questions arose in my mind. But he gave me an idea about it and included me a in a group of students who also has no idea about the DeepRacer.
“Learn Deep Learning using the DeepRacer”
Idid not understand the meaning of this sentence, but later I realized the meaning inside this sentence! We all got added in Amazon DeepRacer whatsapp group and got some faculty mentors too.Then we started working and step by step we started learning about different factors like gradient descent batch size, discount factor, episodes, epochs, learning rate, and some terms as called as Hyper parameters. More importantly, we got familiar with PPO algorithm and the SAC algorithm.
Actually this is self learning concept that we have to explore and learn about it and train our models. Our mentors will help us to learn more and conduct some online meetings to clear our doubts and track our progress in knowing more about AWS Documentations, Github, Medium and blogs, Communities on Telegram, Linkedin and Slack.
Then I started training my model on the AWS DeepRacer website by
- Defining the name to my model
- Selecting the track on which to train my model
- Selecting the type of algorithm as PPO or SAC, and then
- Defining values for Hyperparameters, and
- Set the speed limit and steering granularity, selecting the car as (car with single front facing camera/ stereo camera/ lidar sensor).
Finally, we will have to write our version of the Reward function in PYTHON language that will show the progress of our model, and makes our model train and we have to set timing for training model.
My first model trained graph and evalution with results that it complete average of 28% of the track. This is my first model with some better results that completed with one off track and two completion without going off track. With this I gained hope that I was proceeding in the right direction.
Prijesh-Dev7 — this is my best models which got some more better results, with good training graph and full completion of the track. It was also performed well on the track with good speed.
Do I stop here? Is there anything else further I can do?
My teammates just stop here that they trained model in simulation and just relaxed when they got some model with some best results. But I decided that I will not stop here!
Meanwhile, there is an another option called AWS DEEPRACER LEAGUE, conducted by AWS where they conduct an online monthly contest. This culminates with them hosting a Final Event in AWS Headquarters every year. I just moved further and submit my models as each and every model to AWS DEEPRACER LEAGUE. Different tracks are provided with some more difficulties — I feel this is an golden opportunity for me! , I resolved to explore more, learn more, practice more, and enrich myself more in AWS DeepRacer League. Then I started learning more and trained model each and every day and worked more to compete better in the AWS Deepracer League. I started submitting models in Open Division category and…
BOOM…
I got 98th place out of 1287 participants🤩 in AWS DeepRacer League Open Division. Then I got promoted to PRO division😎, Now I am a Pro Racer.
Then I started participating in pro division leagues in each and every month… The Pro Division is so hard to compete and everyone is getting back in leaderboard with differences in some milliseconds of each other. But I just don’t lose hope and participated in pro division leagues with eager and interest.
This is my results in september qualifier as, I got 150th place out of 208.
This is my October qualifier results in which I got 166th place out of 231.
This is my result in Re:Invent 2021 open race, that I got 875th place out of 1273.
After I participated well in open division and pro division my profile got noticed and, I rewarded with some rewards by AWS DeepRacer, and I got on to the WELL built LEADERBOARD 😎😎😎😎😎😎
Still I am working with DEEPRACER and leagues hosted by AWS DEEPRACER. One day I will get selected to finals and get First place in the AWS DEEPRACER LEAGUE.
My message to everyone who wants to learn AI and Deep Learning is this:
Don’t miss the opportunity, use it and groom yourself towards your goal that will take you to your goals!!!!!!!!
About Me
I am an Tech Enthusiast pursuing my B.Tech Information Technology, in the college KGiSL Institute of Technology in Coimbatore, Tamil Nadu, INDIA.
