All ETDs from UAB

Advisor(s)

Arie Nakhmani

Committee Member(s)

Amy Amara

Earl Wells

Karthikeyan Lingasubramanian

Leon Jololian

Document Type

Dissertation

Date of Award

2020

Abstract

We propose a self-supervised, model-free deep reinforcement learning architecture with cascade reward for unmanned aerial vehicle navigation and target interception in a 3D environment. The first contribution of the dissertation solves the problem of partial observability when non-linear function approximators are used for learning stochastic policies. The second contribution optimizes the problem of maximizing the total expected rewards. The third contribution trains the agent in a photo-realistic environment with a real physics engine. To achieve these goals, a deep Q-network that combines double and dueling architectures is adopted as a value function approximator, and the prioritized experience replays the sample independent and identically distributed random variables from the experiences with the highest relevance to the interception and obstacle avoidance tasks. A new simulator is developed for training drones to track and intercept a target. We demonstrate that our approach achieves better results in learning policies compared to state-of-the-art deep Q-network algorithms.

ProQuest Publication Number

Document on ProQuest

ISBN

9798645468705

Comments

etdadmin_upload_722762

Included in

Engineering Commons

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.