Willow Garage's Week-Long PR2 Robot Training Workshop

Last week Willow Garage kicked off its official PR2 Beta Robot Workshop, where approximately 50 robotics researchers from the 11 PR2 recipient institutions gathered to become familiar with ROS on the PR2, hack on the robot alongside Willow personnel, and have plenty of stimulating robotics discussions.    Attending as a grad student in Georgia Tech's Healthcare Robotics Lab, I had a front-row seat to this enjoyable but grueling event.  I'll give my high-level day-to-day overview of the week (below), which included the "PR2 Launch Party" on Thursday night that featured many prominent folks from Silicon Valley (more detailed coverage).  As an aside, I was proud to hear that so many robotics professionals at the event read and enjoy Hizook -- be sure to contribute your photos and impressions in the comments.

The entire event was held at Willow Garage in Palo Alto, just a stone's throw from Stanford.  We spent much of our time in a huge tent in Willow's parking lot that was specially erected for the event.  It was a nice gathering of individuals and included lots of robots!

Willow Garage PR2 Robot Training Workshop


I was fortunate enough to attend as one of six people from Georgia Tech's PR2 team (seven if you include Willow Garage summer intern, Jenay).  From left-to-right:  Advait Jain, Travis Deyle (me!), Hai Nguyen, Jenay Beer, Tucker Hermans, Cory-Ann Smarr, and Prof. Charlie Kemp (our PR2 team's principle investigator). 

Georgia Tech PR2 team at the workshop


It was a fantastic event with lots of hands-on robot time, as shown by some of Willow's Flickr photos.  My photos are still en route (I didn't bring my personal camera), but will be added to the comments soon enough.  If you have any, contributed photos are certainly welcome.

Willow Garage PR2 Training Workshop  Willow Garage PR2 Training Workshop  Willow Garage PR2 Training Workshop 

Willow Garage PR2 Training Workshop   Willow Garage PR2 Training Workshop


Rumors are that the PR2's are going to be boxed up and shipped off to recipient institutions in the coming days / weeks, especially now that a shipping dry-run to nearby Stanford is complete.  

PR2 Unboxing at Stanford  PR2 Unboxing at Stanford  PR2 Unboxing at Stanford


The schedule was pretty grueling... the detailed technical schedule for the week can be found here, and my personal notes follow.




In the morning, we were introduced to the PR2.  This included overviews of software, networking configuration, and hardware as well as a hilarious safety video (soon to be on YouTube?) that warned about the PR2's incompatibility with stairs (among other things).   

The afternoon brought on a moment of future-shock -- all eleven teams initialized their mobile-manipulating robots simultaneously, bringing them to life (an initialization mechanical calibration routine).  It was humbling to see so many mobile manipulators in one place and acting in near concert.  The feeling continued through the rest of the day as robots moved around via joystick teleoperation.




In the morning, we learned how to setup NavStack for online mapping and localization (SLAM) and to command the base via rviz for autonomous navigation.  This was pretty straight forward for some (Advait and I had setup NavStack on our robot, EL-E, the previous week).  The addition of online (versus offline) map generation was an impressive (new?) capability.

The highlight of the day was learning how to control the various mechanisms, such as the arms, head, and torso.   While some focused on the prescribed keyboard teleoperation, Advait and I independently made quick alterations to the head movement script to have the head (and sensor suite) track the robot's right hand.  This (seemingly advanced) servoing functionality was just a few lines of code asking the head to stare at location <0,0,0> in the hand's frame -- yay tf!

The afternoon focused on cameras and lasers, covering the various sensors, data processing, apps, and tools.  This included Radu's awesome PCL (Point Cloud Library) for processing laser scans.  Hopefully we can goad Radu into releasing Python bindings sooner rather than later.  ;-)

The remainder of the evening was open hack-time on the robots.  [Though I retired to the hotel sans dinner for sixteen hours of sleep to recuperate from sickness].




The only workshop session on Wednesday was about setting up the PR2 in Gazebo simulation.  I don't really have much to say about this session -- I so rarely use robots in simulation.

The next 5 hours were reserved for on-robot hack time.  Our group was doing lots of stuff: comparing the PR2 force calculations from motor effort to readings from a 6-axis force-torque sensor, playing with the image and PCL pipelines, tinkering with projected stereo, working on old laser-pointer interface code, and (myself) getting the ICRA 2010 manipulation demo code up and running.  Unfortunately, we got kicked out around 3pm so that Willow folks could prep for the evening's events.

The highlight of the day (and probably the week) was the big press conference and PR2 launch party on Wednesday.  As a Hizook representative, I got to attend as both a member of the press and as a workshop participant.  The former allowed me access to a press conference featuring Willow Garage CEO (Steve Cousins) and founders (Keenan Wyrobek and Eric Berger).  The evening was intense, and has its own dedicated Hizook post.




Thursday was a dedicated hack day.  In the morning, I got the ICRA 2010 manipulation demo running.  It was nice to see the PR2 picking up previously-unseen objects.

In the afternoon, each recipient institution gave a 15 minute talk outlining their intentions with the PR2.  Unfortunately, I cannot recall the details of all eleven presentations.  However, the talks were recorded and will presumably be placed online soon -- keep watching Hizook for updates.

The remainder of the day was free robot-time.  At midnight, the generators were shut down, cutting off power to the computers, robots, etc.  This was probably my single constructive criticism of the workshop  -- we would have pulled an all-nighter working on the robot if permitted. 




On Friday, we wrapped up with a busy day of tutorials covering servicing and maintaining our PR2.  This included tutorials about how to install extra sensors and how to tune the large counter-balance gravity-compensation springs in the arms.  A treasure trove of information will be available on Willow Garage's support website, including videos of the robot's insides (i.e. counter-balance adjustment) and the upcoming PR2 manual.

The workshop closed with beer and pizza, after which I was offered a unique opportunity to drive around a Texai robot -- it was quite enjoyable.  I remain convinced that remote presence is going to be a boon for the (home) robotics industry. 



In the end, I'm really excited to be a part of the PR2 Beta program.  I'll be working diligently with labmates and collaborators to port our work on manipulation, develop point-and-click interfaces, perform studies with older adults, and experiment with my own RFID work on the PR2.