Design Principles for Intelligent Environments Coen M. H., AAAI-98

1
Design Principles for Intelligent
EnvironmentsCoen M. H., AAAI-98

• Introduction
• Intelligent Room
• Room Vision System
• Speech Interaction
• Conclusion

2
Introduction

• Intelligent environment (IE)
• Highly embedded, interactive spaces
• used to enhace ordinary activity seamlessly
• Goal
• computers to participate in new activities
• HCI via gesture, voice, movement, and context
• Presents Intelligent Room prototype space
• research question How IEs should be designed?

3
Intelligent Room

• Spaces in which computation is seamlessly used to
  enhance ordinary activity
• The UI primitives of these systems are not menus,
  mice and windows, but gestures, speech, affect,
  and context
• IEs are both embedded and multimodal and thereby
  allow people to interact with them in natural
  ways

4
Intelligent Room

• Rather than to make computer-interfaces for
  people, we want to make people-interfaces for
  computers
• Enable tasks historically connecting computers to
  phenomena

5
Why this isnt Ubiquitous Computing?

• They need many connections with the real world
• IE advocate minimal hardware modifications and
  decorations

6
Intelligent Room

• Half of the room is a conference room
• The other half has room for computation

7
Intelligent Room

• Problem1 Computer vision and speech
  recognition/understanding system development
• Problem2 Interconnecting all of the rooms many
  subsystems and coordinating the flows of
  information
• gt Scatterbrain (Coen M. Building Brains for
  Rooms Designing Distributed Software Agents)

8
Room Vision Systems

• Tracking system
• 2 wall mounted cameras
• 3 steerable cameras to get optimal view
• Pointing
• two overhead LCD projectors which support finger
  and laser pointing interactions
• Finger pointing wo parallel cameras
• Laser pointing othogonal camera

9
Room Vision Systems

• Interactive table
• room detects hand-pointing gestures
• not very natural form of communication
• newly placed documents
• post-it notes assigned to lights etc.

10
Room Vision Systems

• Solutions
• robust vision systems that require little
  calibrations and are self-training
• Embedded the rooms vision systems to reinforce
  one another
• Agents (lighting, drapes etc.) communicate with
  it before changing anything
• In tracking most data is irrelevant important is
  where someone is when she stops moving or ehwn
  she crossed a particular threshold

11
Speech Interaction

• Should support spoken language interactions
• people wear wireless lapel microfones
• wakes up the computer by saying computer
• hands- and eyes-free stly of interaction
• unimodal
• the computer can address the user

12
Room Vision Systems

• Issues
• The tracking network need to be trained
• The system is sensitive to any deviation from its
  training conditions
• Computer vision system that rely on background
  segmentation, can be extraordinarily sensitive to
  environmental lighting conditions
• Shadows are particularly difficult

13
Speech Interaction

• issues
• The size of the recognition grammar
• divided into subsets
• Current context is taken into account
• other systems can help to overcome comptational
  limitations
• Can help other modalities

14
Conclusions

• IEs need to be more than mix on systems
• Communication among modalities can lead
• to synergistic reinforcement
• to a more reliable system
• Modalities need to carefully selected
• easy to install
• easy to maintain
• able to be used under different environmental
  conditions
• IEs success
• self-training and avoiding manual calibration!