Robots doing as they're told: a flexible task execution system taught through human-robot dialogue

Doornbos, M. (2014) Robots doing as they're told: a flexible task execution system taught through human-robot dialogue. Master's Thesis / Essay, Artificial Intelligence.

Preview	Text AI-MAI-2014-M.Doornbos.pdf - Published Version Download (4MB) | Preview
	Text akkoord_DoornbosM.pdf - Other Restricted to Registered users only Download (24kB)

Abstract

Domestic robotics has become a key part of present-day AI research, with clear real-world problems which robotic systems could solve. However, much of the research in this field still focuses on laboratory environments and the created systems generally cannot cope with the complex circumstances within the real world. There are several notable issues with operating in an unknown environment and three of them will be addressed here: The first is that the creators of a robotic system cannot ve sure which way of executing a task will work best in the environment a specific robot will end up in. Here, a system was implemented which allows the robot to learn which way to execute a task proves most effective by trying each option several times. This system was implemented as a core part of the robot control architecture used. The consistency in results and statistical basis of this system were tested. This was done by ececuting three ways to perform a navigationless search task a set amount of times. This was repeated several times, with consistent results. A statistical distinction between the versions used could also be seen in the data gathered most of the time. This demonstrates the viability of this system for certain types of tasks. A second notable issue in domestic robotics lies in the quality of human-robot interaction using natural speech. A third issue is that robots cannot generally be taught new things once in the hands of an end user. A dialog system, which allows the user to give orders to and teach robotic systems, is implemented in the previously mentioned robot control architecture. This makes it one of the first domestic robotic systems which can actually be taught by its end users. This system is subsequently tested through interaction with human operators to determine the ease of use. Improvements are made based on the user feedback, followed by another round of testing. The final version appears to be well-liked by its users, although there are still issues, mainly with speech recognition.

Item Type:	Thesis (Master's Thesis / Essay)
Degree programme:	Artificial Intelligence
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	15 Feb 2018 07:57
Last Modified:	15 Feb 2018 07:57
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/11704

Actions (login required)

View Item