Modelling the situation of driving on the grip limit with DDPG algorithm

The number of papers in the topic of autonomous vehicle research is growing exponentially. This paper addresses the problem of self driving a car on tire grip limit, in other words it gives a simple model of a race car driver. Driving is transformed into a simple deep learning problem, where the agent has one action, that is the direction in which the actual speed vector needs to be modified for the next step, and the environment state contains of the actual position and speed. The environment models the race track as a two colour map, to decide on and off track positions, and the car as a point mass with maximal possible acceleration according to the so called GG diagram. Results show that the agent can learn how to drive on the track under the described circumstances.