USING DEEP LEARNING AND REINFORCEMENT LEARNING COMBINATION IN AUTOMATIC VEHICLE APPLICATION OF ALLEE EFFECT

Abstract

Convolutional Neural Network (CNN) can detect images from cameras installed on self-driving cars.
First, we drove a car on a simulator and recorded frames from three cameras: left, right, and center.
These frames were recorded at the rate of 30 frames per second. Additional data recorded were the
distribution of steering angles, average velocity, etc. were passed through a CNN to train a self-driving
system. CNN is like the eyes and visual area in the brain, so CNN's achievements in autonomous vehicle
control are somewhat limited. Therefore, this paper proposes the use of algorithms based on Deep
Learning (DL) combined with reinforcement learning (RL) in the control of autonomous vehicles. We
call this algorithm Deep Reinforcement Learning (DRL) which can send control commands to the vehicle
to navigate properly and efficiently along a defined route. CNN tracks multiple objects while RL predicts
the environment or assesses the current condition of the vehicle to make the safest decision. DRL-based
algorithms have been used to solve Markov Decision Processes (MDPs), where the scope of the
algorithm is to compute the optimal policy of an autonomous vehicle for choosing actions in a
environment with the goal of maximizing a reward function.