On your marks, get set, go!!
It is interesting and fun to participate in the first Westpac AWS DeepRacer Championship which more than 50 teams signed up for the race. Along the journey, I took the crash course of AWS deepracer and the reinforcement learning which is the foundation of deepracer. I did lots of experiments on how to develop an effective reward function, optimise the action space, and fine tune the hyperparameters.
We were fortunate to have a session with two renowned experts Ray and Robin from DBS who have been competing at the world level over the last several years. They shared their valuable experience such as optimal racing line, action space, log analysis with a data-driven approach, local training, and auto submission.
I have spent most of the spare time in the evenings and weekends in training and fine tuning the models during the past three weeks. The effort got paid off, I was ranked 7 in the final and 4 during qu.
