Design and Experimental Validation of RL-Based Decision-Making System for Autonomous Vehicles

In autonomous driving, different Reinforcement Learning (RL) methods have been implemented to deal with different challenges. One of its advantages is the capability to deal with unexpected situations after an adequate trained environment. The inclusion of RL algorithms is considered as a solution for autonomous driving called “agent” that gathers the environmental information and acts according to this from one state to the next one. This paper proposes a solution for a specific environment that is trained with Deep RL and then is tested in simulation and in on experimental platform.