Deep Reinforcement Learning for Autonomous Navigation

Abstract

Autonomous navigation in GPS denied environments poses significant challenges environmental knowledge in limited. Conventional path optimization methods struggle with these complexities. The motivation for this thesis is to develop a model-free learning algorithm based on Deep Reinforcement Learning (DRL) that can effectively navigate in unstructured environments, while avoiding collisions and minimizing time and battery consumption. The primary goal is to contribute a novel approach to navigation using DRL. The added value lies in enabling autonomous vehicles to navigate efficiently without requiring precise environmental or pose information. The algorithm's capability to adapt to uncertainties and produce optimized paths under realistic conditions is a significant contribution.

9