Internal:Jbutterf weekly agenda

From Brown University Robotics

Revision as of 15:45, 22 June 2010 by Jbutterf (Talk | contribs)
Jump to: navigation, search

Contents

Agenda for June 22

AR Teleop

I've started adding functionality for the head and a second arm to be used in the ar_teleop node for the pr2. Also making the code more robust, more readable, and more flexible.

HUMANOIDS

We submitted a new draft of the humanoids paper with new experiments comparing to SOGP and a new video with more content and higher production value. Check it out at [[1]].

HDP-HMM

I continue to work on learning the post creation task with HDP-HMM.

Agenda for June 8

SOGP

The SOGP ros node is nearly ready for release. I've created a new how to page Internal:SOGP and a status page Internal:SOGP Instructions charting the changes. The node is pretty much in final form. All it needs is some extensive testing and some updates to the documentation.

HDP-HMM

I'm adding in some functionality for learning a start and a terminate conditions. Previously I've treated all motions as never-ending which creates strange very high velocities in the data between trials. This will give the learner a sense of when a single trial in the training set starts and ends.

Agenda for June 1

Nao Data Collection

I'm continuing to collect data from the Nao for the overhead task, and will begin learning controllers with different perceptual vectors to see what perception is necessary and start seeing what the results look like.

Interns

What would we like the interns to be doing? Certainly it would be nice to get them to do some teleoperation work, or we could have them write some hand coded controllers for the Nao.


Agenda for May 25

Nao Data Collection

I'm in the process of collecting data for the 2-piece assembly task. I need to find a tag for the hand that will not be washed out against the specularity of the hand. I'm getting data both from the camera and from the top down view.

NEMS Presentation

I'm revising my comps talk to become a NEMS presentation. Preparing for practice talk on Tuesday.

BRICS Research Camp

I'm working on my application for BRICS Research Camp.

Agenda for May 18

Hbase

The Hbase data processing functionality continues to grow as I add more features based on what I need. I'll put up some documentation on how to use it and make a stable version available to the lab this week.

PR2 teleop video

Are Sarah and I supposed to be putting together a polished teleop video? If so we need to do video interviews of the students involved soon, because they leave soon. I heard a rumor Dan Byers was back and may be in charge of this. What's your game plan on this?

Nao Data Collection

I've developed a two piece assembly task (a precursor to the building a gate task) which I can reliably complete using a top down view. I'm figuring out how to run and log blobs from two cmvision nodes at once. I hope to have data collected and a video of the task later tonight.

Paper Outline

I've branched a new icra folder in the rlabsvn/trunk/docs/ which contains my new draft of the experimental section. Here's a pdf File:Icra.pdf. The relevant changed section is IV-b.

Agenda for May 11

Collecting Nao data

Currently it is too difficult to grab the object from merely the visual info from the robot. I am experimenting with providing top down views.

Hbase

I've developed a basic class structure for the data processing module. A TableBrowser allows you to see what sessions are available and various meta data about those sessions. It can also load any number of Sessions (a second class) and allows you to call various filtering techniques on all of the sessions at once. I'm still concerned with the problems that occur if the database is not such down properly. I think this is the main obstacle between us and using the system as it stands now.

PR2 teleop video

I redid the PR2 video for the nao teleop. After seeing Sam's video for the ar-tag based control, I'm doing something similar with screen capturing side by side video feeds of the simulator and the operator. Check it out at flikr. It was a huge pain to get this into a format so that my Mac could play it and it could upload to Flikr. I'm starting an Internal:Video Uploading page where we can share tips on how to do this stuff. Especially now that we have the Macs here so we can have a standardized pipeline.

Revising algorithms

Per your suggestion, I'm putting this project on hold.

Agenda for May 4

Hbase

An initial version of the hbase database is up and running with ros nodes for logging and printing data. Obviously printing data is not particularly useful. We'll be working on creating some python modules which will allow the user to sync and map the data. The package is currently checked into brown-ros-pkg in the experimental section.

Collecting Nao data

Currently it is too difficult to grab the object from merely the visual info from the robot. I am experimenting with providing top down views.

Revising algorithms

After an excellent conversation with Eric about his objections to my model, I have a better understanding of the differences between several of the approximations I was making. He's given me several references to read, in addition to rereading several of the papers I've already read. I'm working on a new implementation of the model in my free time, which will allow me to try multiple sampling strategies.

Agenda for April 27

Collecting Nao data

Version two of the bench is now complete to new specifications. After experimenting with collecting data on the original bench, I revised the specifications. I'll be collecting new data tonight with robot bench 2.0.

Revising algorithms

I've been working through the blocked sampling implementation for doing inference over the models. I've emailed Eric to set up a meeting to talk over a few questions that I have about the technique.

PR2 Teleop

PR2 Teleop now works. It allows control of most of the degrees of freedom in the arm, control of the gripper, and control of the head.

Agenda for April 20

PR2 Teleop

The Nao Teleop controller for the pr2 now can control both arms, the head, and the grippers. Using the bumpers on the feet you can control which gripper you are controlling and also can lock either arm in place. There will be video available later tonight.

Collecting Nao data

The bench is almost built, I've cut about half the pieces of pvc pipe I need and the rest will be done by tonight. I'll film video of the Nao building the two simplex for our meeting tomorrow.

Revising algorithms

I intend to talk to Eric about possible revisions to my algorithm which would please him.


Agenda for April 13

PR2 Teleop

I'm working on getting the Nao teleop to control the pr2. I've installed the pr2 message types and I'm working on figuring out the joint translations to replicate the naos joint positions with the pr2. I expect to have this completed for friday.


Collecting Nao Data

I'm also working on spec-ing out the bench for the Nao. We're debating the merits of getting it printed with a rapid prototyper or building it out of available parts from home depot. I'm headed there tonight to shop for parts. My goal for this week is to collect 20 examples of building a 2-simplex using the new desk setup. As I already have existing infrastructure for learning from this data, I'll also segment and run the controller from this data to try and determine how well it covers the space.