Formula RL: Deep Reinforcement Learning for Autonomous Racing using Telemetry Data

This paper explores the use of reinforcement learning (RL) models for\nautonomous racing. In contrast to passenger cars, where safety is the top\npriority, a racing car aims to minimize the lap-time. We frame the problem as a\nreinforcement learning task with a multidimensional input consisting of the\nvehicle telemetry, and a continuous action space. To find out which RL methods\nbetter solve the problem and whether the obtained models generalize to driving\non unknown tracks, we put 10 variants of deep deterministic policy gradient\n(DDPG) to race in two experiments: i)~studying how RL methods learn to drive a\nracing car and ii)~studying how the learning scenario influences the capability\nof the models to generalize. Our studies show that models trained with RL are\nnot only able to drive faster than the baseline open source handcrafted bots\nbut also generalize to unknown tracks.\n