paper.tex

\documentclass[11pt, draft, a4paper]{IEEEtran}
\usepackage[utf8]{inputenc}
\usepackage{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{cite}
\usepackage{graphicx}
\usepackage{textcomp}
\usepackage[left=2cm,right=2cm,top=2cm,bottom=2cm]{geometry}
\author{D. Kelting, J. Maliauka, B. Manuel, C. Pereyda, A. Crandall, M. Schmitter-Edgecombe, and D. Cook}
%Minimized names because they didn't fit.
%Daylan Kelting, Julia Maliauka, Brittany Manuel, Christopher Pereyda, Aaron Crandall and Maureen Schmitter-Edgecombe
\title{Dynamic Robot Placement in Smart Environments}
\begin{document}
\maketitle


\section{Abstract}
As life expectancy rises the proportion of the population above 65 rises as well.
This increases the ratio of people in need of care compared to the number of
caretakers, making in-home nurses prohibitively expensive. \cite{UNProspects2017}
Robotic assistance for the elderly is a promising solution to this problem, 
allowing people to age in-place independently. Robotic assistants are able to 
remind the user of day to day tasks and guide them through tasks if they have 
difficulty. \cite{RAS_Paper} One problem in robotic assistance is the question of where the 
robot should be when users do not need help. In order to be effective, the 
robot must be able to approach the user in a timely manner; it must also stay 
out of any potential paths the resident could use while they transition between 
activities. Here, we describe an algorithm that used smart home data to choose 
locations for the robot to sit, in real time as residence occupy the living 
space. We used a combination of smart home sensor data and SLAM (Simultaneous 
Location And Mapping) map our algorithm reliably found locations that are close 
to users while also being out of the way. We evaluated our algorithm using both 
historical data {from?} and a preliminary survey study. \{begin projected 
results\} Our historical data analysis showed that our algorithm maintained a 
distance of \{...\} from users, even when there are multiple users in the 
space. Our survey study showed a strong correlation between human scores of 
locations and algorithm scores. \{end projected results\} This work will allow 
home and medical robots to live more harmoniously with human users by having 
better reaction time to assist residence. 


\section{Introduction}
Intro after the rest of the paper is written. 

\section{Related Work (DAYLAN)}
\subsection{To Cite}
\begin{itemize}
    \item RAS Paper
    \item Smart home Paper(s)
    \item Nurse to Elderly Paper(s)
\end{itemize}


\section{Methods}


\subsection{Problem Overview}
In our problem scenario, we were given a Simultaneous Location And Mapping 
(SLAM) map of a smart home and real time stream of motion sensor data aligned 
with the SLAM map. Our goal was to design an algorithm that would  take this 
information and produce an optimal location for RAS to idle in. [RAS INTRO]. 
This location was scored on two key elements: ability to move to the resident 
in a timely manner and being out of the way. 


\subsection{Solution Overview}
For our solution we used two sets of data: SLAM map and sensor data from a 
smart home environment. From this data we built 6 of our own maps and a 
weighted average point with the most relevant data to determine the optimal 
placement for RAS. The reachability map was designed to show all the areas 
that RAS could reach, the wall map was made so that RAS could gravitate toward 
walls, the cramped map was used to determine tight spaces. The long-term and 
short-term heatmap were both used to observe traffic in the space; the long-
term map was used to observe total activity and short-term focusing on the 
last 30 minutes. All of the aforementioned maps were combined to determine the 
optimal location for RAS. Once we had this location we conducted an informal 
study to determine how accurate and realistic our result was. [STUDY SUMMARY 
INFORMATION].


\subsection{Using the SLAM Map}
Explain a bit more about the SLAM map, and the processing we did before we did 
the below maps.
[SLAM FIGURE]

\subsubsection{Reachability Map}
DAYLAN
[REACHABILITY FIGURE]


\subsubsection{Wall Map}
The wall map was created by looking at each point within the SLAM map and 
"waterfall" it out by a rate of 70\% until we've reached a value of 10 or 
under. This waterfall process is best explained in the info graphic below. [
NUMBER FIGURE]. This process was repeated for every point in the SLAM map. We 
wanted this information so that RAS could gravitate toward walls and stay out 
of the residents way. This map was used in our final placement map by adding 
its value weighted at 0.5. [WALLMAP FIG]


\subsubsection{Cramped Map}
The cramped map was extremely similar to the reachability map. The difference 
was that the [KERNEL] size was increased to a 1.116 meter square. If any area 
was smaller than this square we would mark that area as cramped. We wanted 
this information so that RAS would be aware of small areas so they could be 
avoided. This map was used in our final placement map by subtracting its value 
weighted at 100. [CRAMPEDMAP FIG]


\subsection{Using the Sensor Data}
Explain the sensor data, what kind there is and what specifically we used from 
the sensor data. I think we just used motion sensors. Double check that. [
SENSOR FIG?]


\subsubsection{Long-Term Heatmap}
Our long-term heatmap used historical data collected from the smart home 
environment. The specific set of data we focused on was the montion sensors, 
this was used to determine approximate traffic patterns of the resident. We 
wanted this information to have RAS be near the resident. [LTHM FIG]


\subsubsection{Path Map}
DAYLAN
[PATHMAP FIG]


\subsubsection{Short-Term Heatmap}
Similar to the long-term heatmap, the short-term heatmap also used montion 
sensor data from the smart home. However, the short-term heatmap was 
constantly updated when a sensor was triggered. When these events would occur 
the map would update with a heavier weight at that location. We wanted this 
information to ensure that RAS would be near the resident as much as possible. 
Since we wanted this map to be current, it had a decay rate of 50\% every 
thirty minutes. [STHM FIGURE]


\subsubsection{Weighted Average Point}
To create the weighted average point we used the short-term heatmap. This 
would produce a single point that represented the average location of the 
resident in that time frame. To obtain this value we would give each point a 
weight multiplier. This multiplier was determined by how active that point was 
in the short-term heatmap. We then summed[WORD?] these points to get a 
singular average location that RAS would gravitate toward. This point was used 
in our final placement map by subtracting its value weighted at 0.1. [FIG?]


\subsection{Putting it All Together}
Our lurking algorithm used the above maps to suggest the optimal location for 
RAS. The weights we used for each map have been manipulated multiple times to 
optimize our final point. We noticed some predictable trends: centralized 
around the weighted average, remained close to walls, avoided tight areas, and 
ignored regions forbidden by the reachability map. By having altered the 
importance of each input, we could allow RAS to prioritize different regions. 
Our final iteration of the algorithm attempted to balance the inputs to 
maximize availability and minimize annoyance to the resident. [FIG]


\subsection{Analyzing Results}
[STUDY INFORMATION -- Data, compare, figures]


\section{Discussion}
While this project produced the results we wanted, it still had a number of 
limitations. These limits included: time, informal study, one map, and lack of 
generalization. This project time line was 8 weeks, because of this a lot of 
elements of the project were condensed. Originally, we wanted to do a formal 
study; however, in the time allotted we would have not been able to get 
approval from [STUDY PEOPLE]. This led us to conducting an informal study 
which had the limitation of most participants being fellow research students. 
These results were limited by the small group of participants. One of our 
biggest limitations was that we worked with one smart home layout, it 
prevented us from generalizing this algorithm enough to be adapted to any map 
currently. 


\section{Future Works}
In the future we want to move into working with a variety of maps, as well as 
unknown maps while keeping the same level of accuracy. Working with a variety 
of maps would also increase the generalization of our code. Given more time, 
we would restructure our study into a formal study that included a more 
diverse group of participants and locations. 


\section{Conclution}
Delay until finished.


\section{Acknowledgements}
\begin{itemize}
    \item Mentors: Crandall, Holder, Cook, Schmitter-Edgecombe
    \item Chris?
    \item National Institutes of Health (NIH)
    \item National Institute of Aging (NIA) grant R25AG046114  (different than NIH?)
    \item Other?
\end{itemize}


\section{References}
\begin{itemize}
    \item SLAM Paper
    \item RAS Paper
    \item Nurse to Eldery People Paper
    \item Other Paper
\end{itemize}

\bibliography{database}
\bibliographystyle{ieeetr}


\end{document}