Human-robot interaction using decision theoretic planning via Markov decision process /
Designing interaction for a multi-agent system when only computational agents are involved has been extensively studied in several literatures and many methods exist to model them. The reason is, in such cases it is possible to define the world or environment entirely where the agents are expected t...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Gombak, Selangor :
Kulliyyah of Engineering, International Islamic University Malaysia,
2017
|
| Subjects: | |
| Online Access: | http://studentrepo.iium.edu.my/handle/123456789/4861 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Designing interaction for a multi-agent system when only computational agents are involved has been extensively studied in several literatures and many methods exist to model them. The reason is, in such cases it is possible to define the world or environment entirely where the agents are expected to perform their actions. But the whole scenario changes when humans become part of the environment, as it alters several important aspects. People make decisions in ways that are not completely under the control of the system designer. It's also impossible to confine human decision-making to any normative theory (utility theory, game theory, etc.) as they are influenced by a variety of cognitive biases and heuristics, as well as psychological cues, social norms, their mood as well as their physiological state. These aspects of human decision-making must be considered for designing successful robotic agent as a social actor. For socially interactive robots to be successfully implemented, it is essential for them to have the capability to interact naturally with human. However, reaching the level of “naturalness” of human-human interaction in human-robot interaction (HRI) is very difficult to achieve requiring deep implementation of artificial intelligence for recognizing and understanding human modes of communication (i.e. gesture, speech, facial expression, behavior, etc.) as well as producing an appropriate response to human with artificial synthesis. This research attempts to address the problem of uncertainty involved in the human-robot interaction by producing appropriate policies for action to be executed by a robotic social agent performing social tasks in situated environment. Unique characteristics of this class of tasks are that the situation-oriented goals in the physical environment, coordination of behavioral actions with other social actors, and established protocols of interaction leading towards goal known to all participants. This work proposes a novel approach for modeling and planning a multi-agent interaction system in situated social environment where participating agents pursue their individual goals while adhering to a set of behavior guidelines. We propose to approach this problem of modeling the interaction system for this socially capable human-robot by means of Decision-Theoretic Planning using Markov Decision Process (MDP). This research specifically focuses on developing computational model of social interaction for an emotionally expressive robotic head – AMIR-III in helpdesk situation. This robot would take human face as input to the system, analyze the face to classify its expression into a finite set of emotional states, and finally attempt to react with its own facial expression - actions produced according to the generated policy via MDP, to engage with its human-agent counterpart. The performance of the facial expression recognition system is found to be better in terms of accuracy than Fukuda (Fukuda, Myung-Jin et al, 2004) and closely comparable to Yang (Yang, Ge et al. 2008) and Leijun (leijun, XIzhong et al. 2009). The classified expressions are used as input states of the MDP for the proposed HRI system. The sociality of the system is verified through experimentation with human agents. The results achieved are compared with existing systems for benchmarking which demonstrated the efficacy of the proposed method for intended social setting. |
|---|---|
| Item Description: | Abstracts in English and Arabic. "A dissertation submitted in fulfilment of the requirement for the degree of Doctor of Philosophy (Engineering)."--On t.p. |
| Physical Description: | xvi, 187 leaves : ill. ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 166-182). |