What Have You Done?
I am deeply interested in intelligent systems that perceive and act
within human environments. The following narrative gives a
chronological, hyperlinked summary of my past research activities. For
the cold facts you can look at my CV
and publications.
At MIT I was a member of Rod Brooks's group. I
initially worked on vision for the humanoid robot Cog
and an independent cog-like robotic head. This was followed by
research with Cynthia
Breazeal that led to our proposal for a new approach to artificial
intelligence (AI) based on robots that
emulate infant development. I then worked with Yuppy,
a mobile pet robot, for which I focused on the actuated vision system
and human-robot interaction. The Yuppy project evolved into the Coco
project, which allowed me to contribute to the creation of a new
robot from the initial brainstorming sessions to the design, the
construction, and the programming. Coco (left image) ended up being
similar to a quadrupedal infant gorilla that could crawl around and
sit up. For my Area Exam, I evaluated computational
models of the rat hippocampus.
As a qualified graduate student, I became interested in how wearable computing
could be used to teach robots to behave intelligently. Paul
Fitzpatrick and I created the first system to use a shoe-mounted camera
to perform gait analysis, obstacle detection, and context
recognition. For my dissertation, I created
the first wearable system to autonomously infer a kinematic model of
the wearer via body-mounted orientation sensors and a head-mounted
camera. I also demonstrated a variety of applications for this fully
untethered system, such as browsing the wearer's activities, acquiring
image segments of the wearer's hand and manipulated objects, and
discovering significant arm postures.
In parallel with my work on wearable computing, I continued to
contribute to robotics research. I became especially interested in robot
manipulation. I worked on mobile manipulation in built-for-human
environments as part of the Cardea
project. Cardea (right image), which was designed to open doors,
used an RMP Base (Segway for robots) as a mobile platform and an arm
that was a predecessor to the arms of the humanoid robot Domo.
As a postdoctoral researcher, I worked with Aaron Edsinger and
his humanoid robot Domo. We
created new methods for manipulation in human environments,
including novel approaches to tool use,
visual-motor learning, and human-robot interaction. We also developed assistive applications that enabled Domo to
help a seated person place everyday objects on a shelf, or
perform everyday bimanual tasks such as pouring
and stirring.
These assistive applications led me to my current research, which
focuses on the development of intelligent robots with autonomous
capabilities for healthcare.