Convolutional Neural Networks Towards Arduino Navigation of Indoor Environments

TL;DR

This paper explores various computer vision and machine learning techniques, including edge detection, floor detection, and imitation learning, to enable a low-cost robot to navigate indoor environments autonomously.

Contribution

It compares the effectiveness of different methods for indoor navigation on a low-budget robot, providing insights into their practical implementation and challenges.

Findings

Supervised floor detection showed promising results.

Canny Edge Detection was less effective in complex environments.

Imitation Learning demonstrated potential but required further refinement.

Abstract

In this paper we propose a number of tested ways in which a low-budget demo car could be made to navigate an indoor environment. Canny Edge Detection, Supervised Floor Detection and Imitation Learning were used separately and are contrasted in their effectiveness. The equipment used in this paper approximated an autonomous robot configured to work with a mobile device for image processing. This paper does not provide definitive solutions and simply illustrates the approaches taken to successfully achieve autonomous navigation of indoor environments. The successes and failures of all approaches were recorded and elaborated on to give the reader an insight into the construction of such an autonomous robot.