Agenda

• Start planning itinerary update from Scott
• Robot shipping plan
• Self-administered PC and networking update
• List of pre-known objects posted
• Demo of working image processing nodes
• Finally, the list of completed and in progress tasks.

Minutes

• Everyone should contribute to the abstract.
  • It can be checked out like this: svn co svn+ssh://username@cascade.cs.ubc.ca/lci/project/raid1/srvc/SVN/DOC/abstract
• Bug Scott about planning who's going to Vegas
• Start looking into renting SUV/minivan to drive the bot to vegas. In parallel, Catherine will investigate shipping options in case those look better.
• Robot network setup is complete. From any locally administered machine, we should be able to drive the robot
  • Put this in your .bashrc.ros file:

function fraser() {

export ROS_MASTER_URI=http://fraser:11311 }

• • Then type "fraser()" in your shell in order to have ROS use the robot's roscore
  • Test this by typing "rostopic list" and make sure you see the robot's devices

• This is the list of pre-known objects, grouped by our assessments:
  • We can probably do these already:
    • laptop
    • toy car
    • toy Stegosaurus
    • table lamp
  • These objects are really going to need some contours or structure:
    • mug
    • bottle
    • bowl
    • Frying pan
  • Not sure if there's any point working on:
    • pencil
    • toy bed
  • We need to obtain a bunch of these objects for testing. We'll go buy more if needed. [ALL]

Completed components:

• Porting of basic drivers for: bumblebee, cannon and powerbot
• Tower design
• gmapping
• Tilting laser drivers
• Robot coordinate transform code
• Network configuration and development environment
  • Robot router setup
  • Setup self-administered PC's
  • ROS instructions
• WG nav stack
• Basic saliency map computation

Current in-progress task list:

• Capture data from robot for testing
• Tower upgrade:
  • Order material for building a new laser/camera mount and assemble same.
• Attention system:
  • Stereo + saliency combined to identify interesting regions [PV]
  • Tilt laser point cloud segmentation [MM]
  • Choice of where to look
• High-level control functionality such as planning
  • Random walk behavior
  • 3 main high-level planners:
    • Exploring frontiers
    • Find tables [PV]
    • Space coverage
    • Look back at objects
  • Top level state machine to choose between above planners
• Recognition framework (James module directly or something built upon that) [AG and CG]
  • Skeleton framework for the recognition system (inputs: robot images, outputs class guesses)
  • Combining results from different types of detectors (different algorithms)
  • Combining results from various viewpoints
  • Evaluate on the known objects
  • Test data interface
  • Felzenswalb detector
    • MB profiled Kenji's python implementation - most of the time in convolution - promising
    • Will investigate cuda'ing pieces
  • Helmer detector
    • Using point cloud,
  • Mccann
  • Training data interface and additional parameters
  • Load balancing between various recognition algorithms
• Cuda on fraser [MB, WW and TH]
  • Need to get the code compiling
  • GPUSift
  • FastHOG
• Web grabbing module [PF and CG]
  • Add additional sources of info
  • Investigate filtering techniques
  • Integrate output data format with classification
• Speed-up of Felzenswalb training [MB]
  • Data transfer kills several ideas we've had about converting to Cuda
  • Kenji suggested several non-GPU speedups which Matt will work on next

Future tasks pending completion of others:

• Use of 3D models in recognition
• Use of 3D information and context in attention system
• Real time result reporting
• Feeding back classification results to robot planner
• Investigate new cameras which might be faster than the Cannon
• Prioritizing computation done by classifiers towards images which look really promising to the attention system, and based on the classes which have already been recognized.

-- DavidMeger - 14 Oct 2009