Abstract
Introduction
Prior to changing my STS research topic, I focused on how to design autonomous
technologies to responsibly advance society. In recent times, the call for more advanced methods
with security systems demands attention. And with automation and artificial intelligence (AI) on
the rise, there has never been a more appropriate time. Traditional surveillance systems, such as
human monitoring with conventional cameras, do not provide adequate coverage of a secured
area. There is potential for inevitable blind spots in the line-of-sight that increase the overall risk
of ill intention by a malicious actor. In addition, other limiting factors like human-error, slow
response times, and cost play a significant role in developing the scope of my technical project.
Therefore, my capstone established a framework for effective surveillance through the use of
programmable autonomous mobile robots (AMRs) and Wi-Fi sensing.
In contrast, my revised research topic examines the societal impact of low-field (LF)
magnetic resonance imaging (MRI) centered around public perception and trust in AI-based
image reconstruction. LF MRI systems offer a more accessible, affordable, and portable
alternative of receiving vital medical imaging. However, there is an ethical dilemma, with
concerns that diagnostic results may not be fully accurate as compared to traditional high-field
(HF) machines. Therefore, my STS research will focus on how these concerns can be addressed
via analysis of existing sources, literature, and even experimentation of improving hardware to
reduce reliance on AI for greater raw accuracy. In fact, I chose this new topic based on an
ongoing research project in developing a low-cost, LF MRI as well as an associated club.
Technical Project
To address the limitations of modern surveillance systems, my capstone team combined
two emerging technologies: AMRs and Wi-Fi sensing. AMRs enable rapid and more direct
responses when intercepting intruders, while Wi-Fi sensing allows for detection of their
movement through obstacles such as walls. The system deploys three TurtleBots: two acting as
interceptors and one as the malicious actor or intruder. Wi-Fi sensing incorporates ESP32
microcontrollers configured as transmitter (Tx) and receiver (Rx) pairs. These sensors collect
Channel State Information (CSI), which characterizes how Wi-Fi signals propagate between the
Tx and Rx pairs. When an individual moves through or obstructs the path produced by a sensor
pair, CSI data will fluctuate and an approximate location is flagged. Using deep learning
algorithms, CSI data are trained to correspond to specific activity such as presence or absence of
an individual. The resulting output of the binary classification is forwarded to a programmed
AMR to traverse strategically toward the detected intruder within pre-defined secured areas.
Testing demonstrated successful interception of the intruder robot, with success rates
ranging from 85% to 100% depending on the complexity of the room layouts. However, the
system faced challenges related to the reliability of the Wi-Fi sensing, particularly due to the
instability in detection outputs. Efforts were made to reduce the noise, but consistent
performance was not achieved. To address this limitation, theoretical outputs were simulated
using a Vicon motion capture system, which provides sub-millimeter tracking accuracy.
STS Research
Around two-thirds of the entire globe don’t have reasonable access to MRI technology.
The disparity is driven by the associated high costs, lack of trained professionals, and inability to
acquire needed materials to construct the machines. A low-cost and accessible solution is to
make these machines low-field, removing hardware constraints pertaining to the use of
superconducting magnets commonly found in high-field MRIs. Conventional HF MRI scans
provide higher image quality due to the increased signal-to-noise (SNR) ratio as compared to LF.
LF MRI scans resort to post-processing techniques to compensate for the increase in noise,
including AI-based image reconstruction. However, the use of AI raises concerns from public
opinion surrounding data privacy, transparency, and trust in diagnostic accuracy.
With my research, I examined public perception of AI in medical imaging by analyzing
various perspectives from professionals in the biomedical field to more general audiences on
social media. My findings generally indicate that there is an overall optimistic outlook of AI
usage in healthcare. Nevertheless, I also explored methods to enhance LF MRI hardware. A
custom-built LF MRI, provided in collaboration with Johns Hopkins University, was analyzed to
mitigate the residual concerns and to promote a more modular design.
Conclusion
Although my technical project and STS research are not closely coupled, a common
motive is shared: advancing engineering design in automation and AI. Emphasis was also taken
into account on the ethical considerations and implications on society. I gained further ethical
understanding of my topics from the lens of diverse groups of individuals. Their perspectives
helped to shape and fine tune my design choices. For instance, the responsibility for handling
intercepted intruders was intentionally left to appropriate authorities, eliminating potential
criticism of unjustified human-robot interaction. In all, I hope to shed light on the importance of
developing new modern surveillance systems, as well as spread awareness to the impact of
low-field MRI technology, while balancing responsibility.
