Back in October 2020 the co-founders of our autonomous team reached the Grand Prix of the F1Tenth competition. Senan Stanley, Andrew Dai and Jakub Pyzska represented Trinity College Dublin on their team R2-DU, competing against 17 teams from nine different countries.
To quickly summarise, F1Tenth is an open source project involving racing an autonomous buggy that’s about one tenth of the size of an F1 car. It was founded and developed in 2016 by researchers at UPenn, and racing competitions are generally run on a bi-annual basis with one in April and another in October. In the last few years, the F1Tenth community has grown enormously with hundreds of students and researchers getting involved globally every year. The core learning material has been kept completely open-source in order to facilitate learning about autonomous vehicles to whoever wants to expand their knowledge.
In the past, the autonomous buggy racing competition has involved time trials where teams aim to achieve the fastest lap around the track, as well as head-to-head battles where a team would compete against an opponent on the track. In a non-COVID world, last year’s competition would have warranted a team trip away to Las Vegas to race on the track however it almost goes without saying that the 2020 competition had to be held virtually. It was held entirely online and streamed on Twitch for the general public to spectate. The team members of R2-DU worked tirelessly on incrementally improving their baseline solution, before entirely scrapping it and building a drift bot instead.
Since competing in F1Tenth last October, our autonomous team has grown to 26 members, all of which are eager to expand their knowledge in the areas of artificial intelligence, machine learning and autonomous vehicles. The Formula Trinity autonomous driving project shares the same core goals of F1Tenth’s developers, with both of our missions being to promote critical thinking skills and to foster an interest in autonomous systems. Many AI concepts that are crucial to autonomous driving are not taught at an undergraduate level in computer science or engineering courses. In order to provide our members with the opportunity to gain practical problem solving experience in autonomous driving research we have taken inspiration from F1Tenth’s online competition and we have decided to host our own in-house virtual autonomous racing competition based on the F1Tenth framework.
Over the past few months our team has been building up their knowledge in the areas of reinforcement learning, perception, state estimation, control and path planning. We have been running interactive labs in breakout rooms to allow our members to ask any questions they may have regarding anything from software setup to implementing their solutions. So far these lab sessions have covered an introduction to ROS and the follow the gap autonomous driving method which is an object avoidance algorithm, footage of simulation shown below.
The members of our four teams competing have been working together to combine their individual skill sets to develop their solutions. Earlier this year, five members of the autonomous team competed in the MineRL competition and are now applying their knowledge of reinforcement learning to our in-house FTAI competition. This involves training a model to explore a simulated race track environment, and after colliding into the walls time and time again it eventually learns to control the velocity and steering angle in order to drive around the track. We’ll be catching up with the rest of the teams in the coming weeks so keep an eye out on our social media for any updates.
We intend to run our in-house racing competition on a Python-based interface and we would like to encourage anyone to get involved. If you’d like to take part check out our crash course on GitHub. With regards to the initial software setup, this YouTube tutorial is a good place to start as our state estimation lead, David Nugent, walks you through how to work on F1Tenth projects using Docker. For more information on the competition make sure to contact us at formulatrinity.ai@gmail.com.