Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 25
6
Revisiting the Scenarios
O
n day three of the workshop, participants were placed into small groups
and given the opportunity to revisit previous scenarios for a second
round of analysis. Loosely following the DARPA Grand Challenge
competition, each group developed a 10-year research proposal on a topic of
their own choosing, using two of the earlier scenario discussions as a starting
point: Hospital Service Robotics and Preparing For and Managing a Major Dis-
aster. In addition to receiving an unlimited research budget, each group was ob-
ligated to rely on the actual expertise of its members. For example, if a group
did not possess a natural language expert, its delivered system could not employ
sophisticated or innovative natural language.
Group 1: Disaster Management System for a Collapsed Urban
Hotel
Moderator: Alex Morison
Group Members: Paul Maglio, Alex Morison, Don Mottaz, Gopal Ramchurn
The moderator, Alex Morison, spoke on behalf of the group. Based on
the large-scale volcanic eruption scenario, he discussed the group’s development
of a Disaster Management System for search and rescue efforts following the
collapse of an urban hotel. As a result of the collapse, people are believed to be
trapped in the rubble within contained cavities that are not navigable by humans
or dogs. The group’s system would make effective use of robots to map cavities
within the rubble (for size, location, interconnectedness) and coordinate the ex-
ploration. Key technological challenges include: mobility, structural stability,
communications, environmental awareness, multi-robot coordination, and “big
data” sense making. The system’s design considerations would very likely in-
clude both staged rescue scenarios and real-world rescue efforts with actual res-
cue personnel.
To provide the mobility necessary for such a system, the group pro-
posed the design of a crawling robot composed of multiple modular sensor units.
This “slug-like” robot would consist of a series of sensor arrays; for example,
one module might be an antenna system to improve communication capabilities.
25
OCR for page 26
26 INTELLIGENT HUMAN-MACHINE COLLABORATION
In addition to navigating confined and unstable spaces, multiple robots—each
with different perspectives—would provide improved spatial awareness; new
reasoning functions could provide 3-D mapping capabilities.
Morison noted that the system’s success will depend on how well mul-
tiple robots can work together as a team. In fact, the group identified multi-robot
teamwork as the group’s most significant challenge, citing the current lack of
breakthroughs in communications protocols and multi-agent coordination. Ef-
fective multi-robot coordination becomes especially critical in post-disaster en-
vironments that are often resource limited and unpredictable. Under some cir-
cumstances, humans would assume a larger or primary role in coordination
efforts—for example, under system failure or when human expertise is required.
In cases where humans and robots share responsibilities, automated reasoning
would be combined with human reasoning.
Lastly, the group observed that as robots develop increased autono-
mous capabilities, there may be a push for increased autonomous decision mak-
ing. The group questioned what if anything might limit such autonomy. For ex-
ample, what ethical considerations exist for human robot rescue teams (with
varying various levels of autonomous capabilities) that triage lost or injured in-
dividuals?
Group 2: Team Clean
Moderator: Michael Beetz
Group Member: Michael Beetz, Andreas Hofmann, Mark Neerincx, Liz
Sonenberg
Michael Beetz, the moderator, provided a summary of the group’s dis-
cussions. Beetz indicated that the group focused its efforts on designing a home
robotic cleaning team, “Team Clean,” composed of multiple machines (e.g.,
humanoid robot, vacuum cleaner, small UAV to “map” the environment), and
potentially a human director. The team would be capable of accomplishing a
number of tasks with varying degrees of difficulty, from cleaning bathrooms to
washing dishes, vacuuming, and doing the laundry.
To do this, a number of research challenges would be addressed, in-
cluding: practical task manipulation (e.g., picking up fragile objects), smooth
locomotion and navigation in a dynamic environment (e.g., going up stairs and
opening doors), safety (e.g., not getting in the way of residents or pets), human-
robot communication, and social robotics. In addition, machines would have to
be able to learn and recover from mistakes and possess sufficient knowledge
intensiveness (e.g., to go from an abstract task “to clean up” to understanding
how clean is “clean enough”).
Some tasks, Beetz acknowledged, would require varying degrees of in-
teraction between machines and residents. In some cases, a robot may request
feedback from the resident. For example, a robot might ask whether a dirty glass
OCR for page 27
REVISITING THE SCENARIOS 27
situated near the resident is currently being used or in need of washing. In other
cases, the system should be adaptable if it is re-tasked by the resident. This
could occur if the resident’s cleaning expectations differ from those of the robot.
The robot may also need to resolve conflicting resident demands—for example,
balancing a parental request to “pick things up off the floor” and a teenager’s
request to “leave the bedroom as it is.” To deal with this situation, the group
proposed a system with one “chief,” as well as an organizational structure to
deal with conflicting goals.
Lastly, the group identified performance evaluation as a significant el-
ement of the system. For example, how many tasks were accomplished and in
what time frame? How well did the team function? As the team evolved, the
system would be scalable to accomplish a wider range of tasks.
Group 3: Biped Hospital Companion Robot
Moderator: Candy Sidner
Group Members: Robert Hoffman, Lakmal Seneviratne, Candy Sidner, Rong
Xiong
The moderator, Candy Sidner, provided a description of the group’s
proposal for creating a biped hospital companion robot (based on an earlier dis-
cussion of hospital service robotics). This robot would undertake personal care
activities (e.g., dressing and bathing patients and picking up laundry) and pro-
vide mobility/balance support by preventing mobility-related accidents and
catching patients who are falling. In addition to physical manipulation require-
ments, some basis for human-robot communication is required and humans need
to be comfortable receiving robotic assistance. For this reason, human-robot
trust is an important systems requirement.
To accomplish these tasks, the proposed biped robot would be designed
with articulated, touch-sensitive hands and somewhat soft bodies with suitable,
nonaversive “skins.” In addition, visual recognition would be integrated with
touch and task-manipulation capabilities. Communication between the robot and
patient would be computer-controlled and employ simple dialogue—for exam-
ple, questions that can be answered with a “yes” or “no” or with a very short
statement.
Sidner added that algorithms, such as those used to predict the move-
ments of rapidly traveling Ping-Pong balls, would be tuned and applied to pre-
dict when a human is falling and to respond appropriately. This would require
the robot to distinguish not only between types of falling (e.g., falling while con-
scious or unconscious), but also between similar actions (e.g., falling versus
bending over to pick something up). For some frail individuals, the group noted,
the line between falling and bending over is thin. By merging robot companion
and robot assistant technologies, the robot could also act as an instructor or
coach for patients. For example, a robot that has learned to balance itself could
OCR for page 28
28 INTELLIGENT HUMAN-MACHINE COLLABORATION
not only help feeble patients cross hospital floors but could also act as a physical
therapy coach.
As a part of its design, the group would also develop a number of
testbeds to assess both communication and trust issues, as well as appropriate
and safe interactions between robots and patients. Questions addressed would
include: How does the nature of human-robot communication and interaction
change when robots are working with patients who may be sick, feeble, or phys-
ically or cognitively impaired? How anthropomorphic should a robot companion
be, and should it be more or less anthropomorphic if it is engaging in a conver-
sation or dressing/undressing a patient? How might robot companions work with
other robot companions in this environment?
Group 4: The Robotic Patient Advocate
Moderator: Michael Freed
Group Members: Michael Freed, Yukie Nagai, Jean Scholtz, Satoshi Tadokoro,
Manuela Veloso
The moderator, Michael Freed, spoke on behalf of the group. Using the
medical service robots as a starting point, Freed described the group’s proposal
for creating a robotic patient advocate that would work either as an intermediary
between the patient and hospital staff or directly with patients. The robotic pa-
tient advocate would keep nurses up-to-date (e.g.., monitor and report changes
in patient physical or emotional states), provide continuity when nurses change
shifts or when patients are assigned new doctors, and communicate with nurses
when patients are asleep or unable to effectively communicate. In addition, the
advocate would directly provide information to confused or forgetful patients
(e.g., asking “Why am I being wheeled to Room 108?” or “Have I taken my
medication already?”). The advocate would also support medical staff when the
patient required encouragement. Lastly, the advocate would run interference
with visitors and people who stay too long or get in the way of medical staff.
As Freed explained, the advocate would leverage the group’s collective
experience in autonomous systems, communications and dialogue, human emo-
tion, machine-human interaction, and performance evaluation of both robots and
humans. Although some of these capabilities were possible using conventional
technologies, six key breakthroughs would be required. (1) Dialog: The advo-
cate should be capable of high-level discussions with people possessing different
knowledge, motives, and cultures. (2) Multimodal Sensing: The advocate should
be able to tap into—via many and complex sensors—a hospital’s data-rich envi-
ronments to access a patient’s medical records, real-time physiological condi-
tions, test results, and schedules. (3) Strategic Planning: The advocate should
take action by balancing a patient’s immediate goals and requests with long-
term patient support that considers legal and safety issues. (4) Safe Navigation:
The advocate should navigate a complex environment of constantly changing
OCR for page 29
REVISITING THE SCENARIOS 29
people, carts, beds, and equipment. (5) Social Understanding: This might re-
quire the advocate to know when and how to “shoo” away visitors who are un-
wanted or have stayed too long. (6) Multi-Persona Negotiation: The advocate
should be a middleman or “middlerobot” among a floating team consisting of
patients, family members, doctors, nurses, and others.
The group acknowledged that evaluation of the system was critical;
thus, the advocate would be developed first with limited capabilities that would
be expanded on the basis of experience and learning.
Group 5: Providing Post-Disaster Basic Services
Moderator: Lin Padgham
Group Members: Tal Oron-Gilad, Lin Padgham, Dirk Schulz, Holly Yanco
Lin Padgham, the moderator, provided the description of the group’s
proposal. As one component of the volcanic eruption disaster-management sce-
nario, Padgham described the group’s design of an information management
system to provide basic services, such as communications, food, power, and
water, in the first week following a major disaster. The system would not pro-
vide total coordination across the entire disaster management value chain, but
rather would provide on-the-ground individuals with decision support.
Such a decision-support system would require data inputs from numer-
ous sources, including cell phones, sensors, weather reports, and UAVs. The
system would take in data in a variety of formats and then organize and share
those data with a range of specialized users. For example, data inputs from
UAVs that show downed power lines could be used to coordinate prompt robot
deliveries of electrical and other power sources to neighborhoods lacking elec-
tricity. Effectively distributing and acting on this information will require simple
yet specialized human-machine interfaces.
Ongoing access to massive amounts of parallel data would allow man-
agement officials to better prioritize their attention and efforts—for example,
whether to immediately evacuate a neighborhood or to first restore basic infra-
structure. Padgham added that the system could be used as a simulation tool in
advance of a disaster to improve emergency management response. By assessing
the efficacy of different communications protocols and of evacuation routes un-
der different environmental and social circumstances, authorities can identify
where critical post-disaster response failures are likely to occur.
The system would also make “individualized” information available to
both specialized users (e.g., UAV operators with specific data needs to survey
for downed power lines) and to untrained users who are stranded in their homes
with limited food and water. Although acknowledging that such a system would
provide complex decision support, the group noted that human judgment will
always remain key.
OCR for page 30
