HRI%20A1%20(Sukhatme)

1
HRI A1 (Sukhatme)

• Monica Nicolescu, Benjamin Kuipers, Cynthia
  Breazeal, Jennifer Burke, Jana Kosecka, Vladimir
  Kulyukin, Brian Scassellati, Shri Narayanan,
  Holly Yanco, Gaurav Sukhatme

2
Define HRI

• Meta-thoughts Difficult to define, accumulate
  definitions as we go?, do a goal-oriented
  definition?
• Keywords and attributes teleoperation is old, is
  it subsumed by HRI ? HRI originally meant
  teleop, now the definition has changed.
  Definition is shifting like the very notion of
  robot. Teleoperation and virtual humans are both
  degenerate cases of modern HRI. Sliding autonomy
  generalizes pure teleoperation. The type of
  interface matters.
• How it is different from HCI? HCI focuses on
  low level issues, interaction with physical
  world is minimal, to qualify as a robot there
  must be physical embodiment (situated and
  embodied)
• Goal-oriented development of computational
  models, and algorithms
• NSF sponsored grad student WS produced the
  following definition what you need to about
  humans, and what you need to know about robots
  for them to be useful to each other
• How we deal with and respond to robots is not
  understood reflects the fact that the
  boundaries between HRI, HCI, and other fields is
  evolving and not clearly understood.

3
Identify and Define Scope of Research Areas in HRI

• Meta discussion
• Where were five years ago ? The debate then was
  whether HRI was a valid field to be working on.
• Should we list the survey topics since people
  only added two new problems/areas to the original
  survey ?
• Where do we want to be in five years ? Articulate
  that and work backwards.
• In any HRI investigation a natural set of axes
  along which to taxonomize
• Methods characterize how human interact with
  data
• Techniques
• Evaluation metrics
• Questions
• How do robots technologies interact with naïve
  users
• How to HRI problems provide context for
  techniques from other areas (give the vision
  community their marching orders)
• HRI is the route to AI explainability (why is
  the robot doing what its doing), trust, and
  sliding scale autonomy are all key parts of this
• How can one design a robot to collaborate with a
  human to achieve goal(s) the degenerate end is
  teleoperation
• How does interacting with a robot affect a human
  being? Can we sculpt the interaction so we get
  the desired effect?
• What is the emotional component of the
  interaction between humans and robots (and how
  does it affect task accomplishment)?
• How does a robot learn by interacting and
  observing interactions (social learning in the
  broadest sense, switching between distinct
  learning modes
• When is it socially and morally acceptable to do
  some of these things (specifically the three
  things before this one)?

4
Emerging and Important Areas

• Understanding people (goals, intentions, mental
  states)
• Long term interactions with robots

5
Any other questions

• Set of research platforms that run for extended
  periods of time, reliably
• Sharing (code and platforms)
• Relationship between knowledge representation
  that robot has, and the languages it can support
• How well does a robot have to perceive the world
  to be effective?
• How to acquire representation (particularly for
  social learning)?