Penerapan Model Pembelajaran dengan Metode Reinforcement Learning Menggunakan Simulator Carla

Abstract

Artificial Intelligence is the study of how to make machines or computer programs have the intelligence or ability to do things that humans can do. The application of AI is currently in various ways, one of which is for self-driving cars. To be able to do a self-driving car, the AI that is implanted in a car must applied to the method to be able to walk on its path and be able to adapt to its environment. Reinforcement learning is one type of machine learning where agents learn something by doing certain actions and the results of those actions and try to maximize the gifts received through interactions with the environment that are reward negative or positive. In this research, we applied of the reinforcement learning method on the Carla Car simulator. The simulator is used to collect data using an RGB sensor, then modeling experiments which produce several models to be used in simulation experiments. The model is obtained by using the Convolutional Neural Network (CNN) algorithm with the NVIDIA architectural model. From the results of research based on experiments conducted obtained the best model obtained from the experimental model by comparing the maximum reward value, high accuracy and low loss is model 1 in the experimental model A with 100 episodes and model 4 in model B experiment with 150 episodes