IRR@IU Update

1
IRR_at_IU Update

• Team
• Tasks
• VCS
• Vision
• Sensor Network
• AI

2
IRR_at_IU Team Changes

• Faculty Steve Johnson, Florin Cutzu
• Staff Bryce Himebaugh
• PhD Arnab Dhua, Todd Holloway, Alex Leykin
• MS Matt Ira, Corey Bowers
• Project Seminar added for Spring, soliciting
  more student involvement
• Some current participants may return to the
  project in April.
• Discussions with PTL ongoing
• Indications of help () from PTL, Informatics
• Caution Johnson Himebaugh have increased
  teaching/support loads in Spring.

3
VCS Task

• Control I/F between navigation system and
  physical system
• VCS node network communicaiton
• VCM card digital interface
• (RT)Linux, C, ASM, hardware
• Integration Testing on model, preparing for Jeep.
• STATUS operational but primitive functionality.

4
VCS VCS Node

• Application level Xcvr handles network
  communcation and message translation.
• Command process does low-level command
  processing, enforces control rules, monitors
  constraints. Strictly periodic, real-time.

5
VCS Vehicle Control Module

• MSP430 based controller
• Mostly C
• Still being modified for Jeep, model.
• Control rules, timing will migrate down over time.

6
VCS Test Harness

• Main purpose regression testing (nothing fancy
  minimize development effort)
• Alternate purpose enable AI experimentation
  (more later)
• Goal close GPS loop in coming weeks.
• Goal set up distruted integration testing.

7
VCS Outlook

• Integration process has been typical
• Test, test, test
• As the Jeep comes on line,
• Failure mode analysis
• Redundant shut-down functions
• Test site management
• Benchmark (simulation) model is useful
• Needs to be functionally consistent
• Does not need to be accurate
• Additional vehicles?

8
Vision

• Preliminary results from Cutzu, Leykin
• PG stereo- mono-scopic cameras (just arr.)
• Priorities
• Obstacle detection
• Road finding
• Orientation (tower finding, etc.)
• Results posted on IRR_at_IU hub site
• Far to go
• Possible data acquisition excursion in 2-3 weeks.

9
Vision Road Finding (Cutzu)

• Simple nearest-neighbor scheme
• no geometrical information
• no intensity information
• slow

10
Vision Road Finding/Vanishing Point (Leykin)
X
X
X
X
11
Vision Horizon Line (Leykin)

• it definitely runs in real time
• example does not extract a feature just draws
  a line

12
Vision Stereo Ranging (Leykin)
13
Sensor (Video) Network

• one sensor, one node, one person

14
Sensor (Video) Network

• Bumblebee configuration two image streams, self
  synchronized via Firewire
• Stereo differencing in software may consume most
  of a CPU
• Other stereo configurations may require video
  stream fusion, at least.
• Ranging is just one filter, from which multiple
  features are derived.
• And this is just video

15
Sensor (Video) Network

• One or more sensor networks are needed.
• Partial functionality will be needed by the April
  visit.
• At present, IU team lacks the means to accomplish
  this
• This is a potential IP development area for IRR
  LLC
• It also appears to be fund-worthy on a limited
  scale.

16
Sensor (Video) Network Approach

• Data synchronous data flow guarantees fusion
• Demand oriented flow governs acquisition rate
• Opportunities for innovation in specification
  languages, design/analysis tools, implementation
  processes.

17
AI Control Support

• Prototyping Experimentation still far off.
• Need vehicle, realistic sensory input, and a
  means to train
• Used NNs and will use CBR in simulation model
  (Holloway)
• Pf-of-Principle demos using XPlane (Holloway)
• Need to keep AI in the game. It is where IRR
  LLC has commercial prospects