Deep Reinforcement Learning for Goal-directed Visual Navigation

Master thesis (2018)

Authors

M. Kisantal Aerospace Engineering

Contributors

G.C.H.E. de Croon (supervisor 1)
K.G. van Hecke (supervisor 1)
Faculty
Aerospace Engineering
Copyright: © 2018 Máté Kisantal
Deep learning Machine learning Neural networks Computer vision Artificial intelligence Autonomous Vehicles Drone UAV Reinforcement learning MAV Autonomous navigation MAVLAB Deep reinforcement learning
More Info
expand_more

Published Date

23-02-2018

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Aerospace Engineering

Abstract

Safe navigation in a cluttered environment is a key capability for the autonomous operation of Micro Aerial Vehicles (MAVs). This work explores a (deep) Reinforcement Learning (RL) based approach for monocular vision based obstacle avoidance and goal directed navigation for MAVs in cluttered environments. We investigated this problem in the context of forest flight under the tree canopy.

Our focus was on training an effective and practical neural control module, that is easy to integrate into conventional control hierarchies and can extend the capabilities of existing autopilot software stacks. This module has the potential to greatly improve the autonomous capabilities of MAVs, and their applicability for many interesting real world use-cases. We demonstrated training this module in a visually highly realistic virtual forest environment, created with a state-of-the-art computer game engine.