Steering Model for Autonomous Driving System using Deep Reinforcement Learning
An important part of the Autonomous Driving System (ADS) is the steering system which supposed to emulate the behavior of human drivers as a self-driving car controller. This eliminates the need for human engineers to anticipate what is important in an image and to foresee all the necessary rules for safe driving. The most mature machine learning framework that can be put forward to do such task is Deep Reinforcement Learning (DRL) due to its ability to work and interact with virtual simulation environment.
In thesis thesis work, a DRL model is designed based on asynchronous advantage actor-critic (A3C) algorithm. The asynchronous nature of the method enables running multiple simulation threads in parallel which is important given the high sample complexity of deep reinforcement learning. Also, we try to build our model using Deep Deterministic Policy Gradients (DDPG) algorithm to solve the problem of continuous action space. Finally, we try to compare between the two algorithms to see which is better in continuous action control. The objective of the Steering Model is to perform complex tasks such as; lane keeping, lane changing and overtaking.