introduction.tex

\chapter{Introduction}

%TODO: will need heavy rewritting
This report aims to describe the work conducted at the end of the 18 month period by the authors on the use of machine learning to provide a robot with an action selection policy for social human robot interaction, in support of a PhD programme of research. The structure of the report is the following one. The remaining of the first chapter describe the context in which the research in taking place. The second chapter introduce a literature review of the field of action selection for social HRI. Then the work done so far is presented and followed by the future work planned the remaining part of the research programme and is presented in Gantt chart in annexe.

\section{Research Outline} %\section{Research context}

Human-Robot Interaction (HRI) is a

Social interaction between humans rely on a large number of behaviours and features to react to. Furthermore the interactions dynamics follow implicit social rules that we tend to follow unconsciously. With the increasing number of robots designed to interact with humans, their partners will expect them to conform to these social rules or conventions. However, due to the large quantity of possible environment state the robot will have to respond to, it seems unlikely that all the desired social rules could be encoded by hand. 

DREAM only context to test, but PhD larger than it

introduction of Dream, RAT with few ref...

quick motivation and main hypothesis:
in social HRI, combining human supervision and machine learning - human level performance can be achieved whilst relying less on the human overtime, resulting on a lower workload on the human side

\section{Research Scope}

present limitation of the research: presence of a human constantly available, discrete actions, HRHI

definition of metrics tlx - research focus - justification of limits - potential direction - opening?

\section{Research Questions}

consequences from the hypothesis

How to design an interaction setup allowing supervised autonomy and ML

Which algorithm can be used to allow online learning of action policy for social HRI

how to be able to test the setup - not possible to rely on HRHI as it is composed of 2 HRI