CMSC 691 Spring 2016 Bookkeeping Piazza: .ny.cc/hri-piazza - - PDF document

2/11/19 1

CMSC 691 Spring 2016

• Dr. Cynthia Matuszek

Principles of Human-Robot Interaction

an introduction to robots in our daily lives

Bookkeeping

• Bookkeeping
• Piazza: .ny.cc/hri-piazza
• Signup sheet: .ny.cc/hri-signup
• Journals!
• How to Read a Research Paper
• Goodrich 2007
• Next &me: Journal! (Due Wednesday at 11:59PM)
• What you wonder about extending this taxonomy?

Coming soon!

Reading a Research Paper

• What does it say?
• Central ques.on / problem the paper addresses
• Answer or conclusion
• Approach / how do they do it
• What do you think?
• Do you like this paper?
• What do you like most about it?
• What do you like least about it?
• Ques&ons
• What did you not understand?
• What do you want clarifica.on about?

Reading a Research Paper

• Be cri.cal
• Are they solving the right problem? The right way?
• What are the limita.ons of the approach?
• Is the data correct? Well-gathered? Interpreted?
• Be thoughSul
• What are the good ideas in this paper?
• Do these ideas have other applica.ons?
• Are there important improvements?
• Understand
• Summarize the main ideas; compare to other papers

www.eecs.harvard.edu/~michaelm/postscripts/ReadPaper.pdf

Human-Robot Interaction

• What is an interac.on with a robot?
• What is a robot?
• What counts as interac.on?
• Robot(ic)s, for our purposes, is where

computa.on meets the physical world

• Sensing: seeing, hearing, range-finding…
• Actua.on: moving, manipula.ng objects
• Physical interac.ons with humans
• Speaking, hearing, gaze contact,

holding hands...

HRI as a Field

2/11/19 2 Design Axes to Consider

• Autonomy (or “neglect tolerance”)
• Can it func.on by itself?
• Informa.on Exchange
• Who knows what, when?
• Team Structure
• Humans, robots, so^ware agents…
• Adapta.on and Learning
• “Shape of the Task”

Sometimes we don’t care! Do we need to know what it does, and vice versa? Harder than autonomy When shouldn’t a robot learn?

Competitions and Areas

• Driving
• Search & Rescue
• Useful near-term
• Hors d’oeuvres, anyone?
• Crowd naviga.on, direct social interac.on
• Assis.ve Robo.cs
• Social, physical, and cogni.ve support
• Vulnerable popula.ons
• Space and Extreme Condi.on Robo.cs
• Extreme opera.ng condi.ons
• Par.al, delayed, or no communica.on

Robocup Search and Rescue

• fewkahbc

Search and Rescue Space Robots

Valkyrie, NASA Curiosity, NASA Mars InSight, NASA Dextre, Canadian

Assistive Robotics

HAL exoskeleton: construc.on, disaster response www.youtube.com/watch?v=Zd9WhJPa2Ok www.enablemart.com/obi-feeding-device

2/11/19 3 Assistive Robotics

Cyberdyne rehab exoskeleton

Assistive Wheelchairs

www.youtube.com/watch?v=t2SHKyq5yCU- Chiba group

Assistive Robotics

RIBA-II Transfer Robot

Assistive Robotics

Casper Cognitive/Social Assistant

Assistive Robotics

Intelligent System Corporation PARO social robot

Social Robotics

Leonardo, Personal Robot Group, MIT

2/11/19 4 Types of Interaction

• Non-interac.ve work
• S.ll doing tasks for humans
• Teleopera.ng
• Supervising
• Correct errors, give low-level instruc.ons
• Providing goals
• From “Pick up the block” to “Cook dinner”
• Non-supervisory interac.ons
• With people
• In environments (e.g., cleaning)

Low High Interaction Directness

Types of Interaction

• Engineering:
• Building, maintaining, (re)programming
• Teleopera.ng
• Supervising
• Correct errors, give low-level instruc.ons
• Providing goals
• From “Pick up the block” to “Cook dinner”
• Non-supervisory interac.ons
• With people
• In environments (e.g., cleaning)

Low High Autonomy There are a number of formal autonomy scales!

System Design

• How autonomous should (this) robot(s) be?
• What kind of group is the robot in?
• Other robots? Humans?
• How is it structured? Who’s in charge?
• Informa.on management
• Who knows what? Is it transferred?
• How adap.ve or predefined is the task?
• What is the task or task domain?

Autonomy

• How autonomous should this robot(s) be?
• For what kind of interac.on?
• What is the target problem(s)?
• Are their human collaborators?
• Par.al; full; social; correctable; hap.c; feedback; …
• What kind of autonomy?
• Goal-based
• Behavior-based
• Probabilis.c

In practice, these are incredibly high-level simpli6ications of all of arti6icial intelligence – but useful!

Information Exchange

• Who knows what? Who needs to know what?
• When informa.on transferred:
• When exactly?
• From whom to whom?
• How slow? How costly?
• How informa.on is transferred:
• Using what media?
• In what format? (Scripted, one-way, free?)
• Do you need a cogni.ve model
• f the agents?

Many human- robot interactions can also be human- human, robot- robot, or a more complex

• mix. This is

why I often say agent.

Information Exchange (II)

• The medium
• Seeing
• Hearing
• Touch

“Put the pot here and hand me the sugar, please” OKAY

2/11/19 5 Group Makeup

• Humans and robots
• 1-1, many-many, 1-many, many-1, 1-0, 0-many, …
• Related to autonomy, task
• Who has authority? Who has responsibility?
• When does the human not have pure authority?
• Timing? Task-based?
• What is op.mal?

Autonomy Authority

Adaptation

• What can be learned?
• Task domain; task ac.ons
• Changes over .me
• Communica.on style and kind
• Authority and communica.on flow
• What can change?
• The task? The authority? Agent capabili.es?
• Over what span of .me?
• In response to what informa.on?
• What agent(s) should do the learning?

Progressively more!

• Search & Rescue
• Assis.ve Robo.cs
• Transfer
• Social
• Wheelchair
• Space/Extreme

Condi.ons

• Surgical
• Hors d’oeuvres, anyone?

Practical Discussion

• Autonomy
• Group Structure
• Informa.on exchange
• Adapta.on and Learning