In recent years, the field of artificial intelligence has attracted numerous enthusiasts, particularly through the AWS DeepRacer competition, which is held annually at the AWS re:Invent event. In the 2024 edition, one participant stood out for their renewed approach and implementation of improved strategies based on their previous experiences.
The author, who had faced significant challenges in the 2022 and 2023 competitions, reflected on the instabilities of their previous models. Aware of their limitations, they decided to adopt a new approach that would improve the stability of their vehicle both on straightaways and curves, thus avoiding track exits. Their goal was to develop a model that could ignore the visual noise surrounding the competition environment.
To achieve this, they adjusted the car’s steering geometry, limiting steering angles and adding stabilization angles. These modifications allowed the vehicle to correct its trajectory more effectively, especially in tricky segments of the track. Additionally, they conducted multiple tests to diversify their model training, ensuring it could adapt to various curve configurations on the track.
During the AWS re:Invent 2024 competition, the author faced a challenging track marked by visual distractions due to large illuminated screens. Despite these obstacles, their car demonstrated a remarkable ability to stay on course, even in difficult turns.
At the end of the race, the author managed to position themselves in 14th place, with an average time of 10.072 seconds. This performance allowed them to reflect on their performance and consider future adjustments in maximum speed and action angles to better align with the actual characteristics of the car.
With the conclusion of this edition of AWS re:Invent, where it is also contemplated to be the final DeepRacer championship, the community continues to show great interest in physical races. There are initiatives to expand access to this exciting activity, thus keeping alive the educational and recreational potential offered by this technology, promising a bright future for DeepRacer competitions and its community.
Source: MiMub in Spanish
