Autonomous driving using deep reinforcement learning Algorithms on Carla Simulator
Self-driving vehicles have become popular, and need little human interaction to prevent road accidents. The environment must be conscious of the number of sensors mounted on the car. There have been numerous suggestions for learning about the driving strategy for taking regulation steps. Due to the advancement of CNN, the end-to-end tracking of learning is being used increasingly to train the model of the neural network across vast volumes of time-consuming data. Reinforcing learning (RL) can however be trained without the comprehensive labeling of data needed for supervised learning. It was recently seen in the area of self-employed vehicle analysis as a promising strategy for driving policy.
In thesis work, a DRL model is designed based on two different algorithms. First, implementing the DQN (Deep Q-Learning) algorithm to train our model to see whether our agent will take ongoing steps. Also, using a DDPG algorithm that is useful in continuing tracking of behavior in the second Scenario. The goal of the Model is to measure different cases of Self-driving inside the city with different roads and conditions. The model will be tested on the CARLA simulator.