Drone Obstacle Avoidance and Navigation Using Artificial Intelligence

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2018-05-14

Department

Major/Subject

Embedded System

Mcode

SCI3024

Degree programme

Master's Programme in ICT Innovation

Language

en

Pages

95 + 7

Series

Abstract

This thesis presents an implementation and integration of a robust obstacle avoidance and navigation module with ardupilot. It explores the problems in the current solution of obstacle avoidance and tries to mitigate it with a new design. With the recent innovation in artificial intelligence, it also explores opportunities to enable and improve the functionalities of obstacle avoidance and navigation using AI techniques. Understanding different types of sensors for both navigation and obstacle avoidance is required for the implementation of the design and a study of the same is presented as a background. A research on an autonomous car is done for better understanding autonomy and learning how it is solving the problem of obstacle avoidance and navigation. The implementation part of the thesis is focused on the design of a robust obstacle avoidance module and is tested with obstacle avoidance sensors such as Garmin lidar and Realsense r200. Image segmentation is used to verify the possibility of using the convolutional neural network for better understanding the nature of obstacles. Similarly, the end to end control with a single camera input using a deep neural network is used for verifying the possibility of using AI for navigation. In the end, a robust obstacle avoidance library is developed and tested both in the simulator and real drone. Image segmentation is implemented, deployed and tested. A possibility of an end to end control is also verified by obtaining a proof of concept.

Description

Supervisor

Jung, Alex

Thesis advisor

Ramirez, Enrique

Keywords

drones, obstacle avoidance, autonomous navigation, artificial intelligence, computer vision, deep neural network

Other note

Citation