Javascript must be enabled for the correct page display

Robot Localization and Dreaming using Convolutional Neural Networks and Denoising Autoencoders

Chong, Y. and Kuiper, C.G. (2017) Robot Localization and Dreaming using Convolutional Neural Networks and Denoising Autoencoders. Bachelor's Thesis, Artificial Intelligence.

[img]
Preview
Text
AI_BA_2017_Yiebo_Chong.pdf - Published Version

Download (8MB) | Preview
[img] Text
toestemming.pdf - Other
Restricted to Backend only

Download (78kB)

Abstract

Denoising autoencoders have been used with success for indoor localization based on images in a 3D simulated environment. With the recent successes of convolutional neural networks in image recognition tasks, convolutional neural networks are a good candidate for the localization task. In this thesis, the performance of (stacked) denoising autoencoders and a convolutional neural network are compared in a localization problem using color image information. The networks are provided with images labeled with location and orientational coordinates. The performance of the networks are compared by the mean error in location and mean orientation error. Both networks are trained in two 3D simulation environments (a small and large room) and a real environment. The convolutional neural network performed better on all three environments by a large margin. The convolutional neural network attained a mean error of 6cm in location error and 5° in orientation error on the small environment and 8cm in mean location error and 4° in mean orientation error on the large environment. On the real environment the convolutional neural network attained 51cm in mean location error and 27° in mean orientation error.

Item Type: Thesis (Bachelor's Thesis)
Supervisor:
Supervisor nameSupervisor E mail
Shantia, A.UNSPECIFIED
Wiering, M.A.UNSPECIFIED
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 15 Feb 2018 08:34
Last Modified: 02 May 2019 09:49
URI: http://fse.studenttheses.ub.rug.nl/id/eprint/16298

Actions (login required)

View Item View Item