The Bridge - Issue 3, 2021 - 33

Graduate Research SP
Kevin Doherty
Iota Delta, Massachusetts Institute of Technology,
Ph.D. Student in Aeronautics/Astronautics and Ocean Engineering
RESEARCH TOPIC
Robust Object-Based Robot Navigation
Increasingly, robots are deployed with the capabilities not just to measure distances to obstacles
using cameras, sonar, and lidars, but, through recent advances in machine learning techniques,
to also detect and classify objects in the scene and perform other high-level perception tasks.
We expect a robot to know not only where it is and where the obstacles are, but also what those
obstacles are. Combining this semantic information with the geometric information that robots
typically use to build maps is challenging because present techniques for object recognition often
fail in unfamiliar environments. Moreover, even when these systems perform perfectly, ambiguous
scenes can make navigation a challenge. Likewise, we as humans might walk around an unfamiliar
area and think " is this the same building I saw before? "
Kevin's work, supported by an NSF Graduate Research Fellowship, is centered on the problem of
building algorithms or robots to combine object-level information with geometric information during
navigation, while being robust to the inevitable failures that come with long-term operation. A major
focus of this work has been on enabling robots to efficiently keep multiple hypotheses about the
state of the world; e.g. " this could be the building I saw before, or it might not be. " This work is
particularly targeted toward marine robotics, where sensing is often limited and natural features are
repetitive or sparse, so reasoning about ambiguity is critical.
LEARN MORE
https://scholar.google.com/citations?user=IYPhbO0AAAAJ&hl=en
CONTACT
https://people.csail.mit.edu/kdoherty/
Fiona Popp
Beta Eta, North Carolina State University, M.S. in Electrical Engineering
RESEARCH TOPIC
Novel Optical Sensor System for Prosthetic Wrist Rotation Control
Conventional prosthetic wrist control utilizes electromyography (EMG) signals. The two main
EMG control methods are Direct Control (DC) and Pattern Recognition (PR). DC uses an agonist/
antagonist muscle pair to control the direction of one Degree of Freedom (DOF) at a time.
Contracting both muscles allows the user to consecutively switch through the DOFs. PR trains a
classifier based on the EMG signals when performing different arm/hand postures. Both of these
methods impart challenges on the user. DC is unintuitive because the user must consecutively
switch between the DOFs. PR has challenges with robustness due to inconsistent EMG signals from
skin impedance, electrode shift, and fatigue, which cause misclassifications.
Fiona's research, supported by a National Science Foundation grant, consisted of developing
and evaluating a novel sensor system for wrist rotation control. The system utilized an optical
displacement sensor, commonly used in computer mice, to monitor the position of an able-bodied
individual's forearm. The speed and direction of the prosthetic wrist, mounted on an able-bodied
prosthesis adapter, was controlled based on the approximated position from the optical sensor.
This system allows for simultaneous control of 2 DOF, when utilizing the optical sensor for
pronation/supination and EMG control for hand open/close. Additionally, it has the potential of
being more intuitive than DC, more robust than PR, and allow for faster task completion time.
LEARN MORE
https://repository.lib.ncsu.edu/handle/1840.20/38600
TLIGHT
Top left: Illustrated example of an
underwater vehicle performing objectbased
mapping (x: robot poses, l:
object landmark). Top right and bottom:
Mapping pilings with the BlueROV2
underwater robot in Woods Hole, MA.
Prosthesis setup and 3D printed
adjustable mount with sensor.
CONTACT
https://www.linkedin.com/in/fionapopp/
HKN.ORG
33

https://scholar.google.com/citations?user=IYPhbO0AAAAJ&hl=en https://people.csail.mit.edu/kdoherty/ https://repository.lib.ncsu.edu/handle/1840.20/38600 https://www.linkedin.com/in/fionapopp/ http://www.HKN.ORG

The Bridge - Issue 3, 2021

Table of Contents for the Digital Edition of The Bridge - Issue 3, 2021