Besides reading cool papers, my Twitter account is mostly used for catching up Formula 1 news. Very excited about Lando Norris 's great performance recently. Now we combine Formula Racing to AI research: The following video shows we train a Reinforcement Learning policy to drive a Dallara F317 in... Monza and Barcelona Circuits autonomously in simulation. You can think of it as optimizing a Qualifying lap: You want the car to run a single lap, no opponents, with the fastest speed, and shortest lap time. We open-source the code to train RL in a racing simulator with Assetto Corsa, a widely deployed platform for esports. We also train with human demonstrations with multiple expert drivers including a professional e-sports driver. Interestingly AI is not beating the professional yet, but close. See our open-sourced code, simulator, and data here:show more

@LandoNorris Simulation plays a critical role in training drivers, one of the reasons why Red Bull has not been dominating recently:

And drivers play games as well:

Now go back to our paper. We open source and evaluate RL and imitation learning on 4 different circuits with 3 different cars. We have humans drive and collect 927 laps across these environments. Participants included a professional e-sports driver, four experts, five casual drivers, and five beginners.

The following table shows lap time (in seconds). Reinforcement Learning agent is able to beat the professional driver in 3 different cases, but still loses to the guy in most cases.

We collect human data from a fancy simulator in Austria.

And also not so fancy one in my lab at @UCSDJacobs

Check out our paper here: Everything is open sourced. A very fun collaboration with Adrian Remonda, @ncklashansen , Ayoub Raji, Nicola Musiu, Marko Bertogna, Eduardo Veas

@F1 @LandoNorris One day, it will be fun to see Lando race an autonomous F1 car!

@F1 @LandoNorris 10 sec penalty for Ocon NEEDS to be implemented as a bit of fun