IROS 2009 Report

Last week was spent at the IROS 2009 conference in St. Louis, MO. When starting this conference report, I realized that IROS has been a conference of repetitions for me:

• In 2008 in Nice, France, I visited it the IROS together with Sebastian, as the only people from our group. We again visited the IROS together this year.
• Last year, I had to give a talk for Joachim who couldn't go for personal reasons. This year I had a talk instead of Raphael, whose paper is due for a patent.

And as if that wasn't enough, quite a few other things about the conference were also the same:

• The only professor from Bielefeld attending last year was Jochen Steil. Same this year ;-) (though, admittedly, Yukie who only just left for Osaka a month ago was also there)
• The weather was about 5-10 degrees colder than usual for the city and the time of year. (I know that for a fact for St. Louis and it certainly felt like it in Nice ;-)
• Last, but not least, the main reason for our attendance was a workshop.

However, fortunately, some things also changed. This year, the workshop was the TC-SOFT Event-Based Systems for Robotics workshop, which Sebastian and I co-organized. Also, the talk I had to give was at the workshop and so I was free to change the topic (and, actually, given that Raphael couldn't talk because of the patent, required to!) to some of my own.

Anyway. Back to the conference report ;-) I saved some of the data for last -- first about the talks.

I quite liked the first keynote talk about fish biorobotics by George Lauder -- robotic fish may still be a bit out there, but fish are generally quite interesting and it was also novel to see the experimental apparatus required to study them. We in the HRI world always complain about the effort required to build systems but the setups these biologists were using don't seem easy to build either!

Noticeable talks included

• Unsupervised Simultaneous Learning of Gestures, Actions and Their Associations for Human-Robot Interaction<, by Mohammad, Nishida and Okada. They had an interesting Wizard-of-Oz style gesture learning setup. The method was a bit data-mining heavy and seemed overkill for their initial (simple) experiment, but should be usefull in more complex situations.
• Persuasive Robotics: The Influence of Robot Gender on Human Behavior by Siegel, Breazeal and Norton modified the robots voice and did a fairly substantial user study on how that affected donation behavior in a museum setting. Unfortunately, they did not modify the appearance, but voice alone gave some differences. They were not totally clear on whether people knew they were in a study -- first they said people were unaware, but later they mentioned giving them money for donations. Still, interesting.
• My absolute favorite talk at the conference was Active Segmentation for Robotics by Misha, Aloimonos and Fermüller. Aloimons gave the talk and he started by totally bashing the performance of standard segmentation methods based on one frame, because their outputs are unusable for typical robotic tasks such as manipulation. Interestingly, what is liked by recognition (strong edges and varying texture) isn't by segmentation. He then introduced a novel log-polar representation centered not on the image center but the an object fixation and combined this with active camera motion, video or stereo, to result in almost perfect segmentations.
• Influences on Proxemic Behaviors in Human-Robot Interaction by Takayama and Pantofaru seems very relevant to Patrick's work. They also included head gaze and had a very interesting indirect measure for people's comfort level, which to me seems much more indicative than active reporting.
• Vision Based Motion Control for a Humanoid Head by Visser, Carloni and Stramigioli was a follow-up to their head paper at ICRA, only this time they presented the full head. Again, lots of screw theory (no pun intended) which probably isn't as interesting to me as it would be to a mechanical engineer, but certainly a head worth watching.

Of course, there were lots of other talks worth watching and I saw some by former or future colleagues, but I'll leave that for some other time.

Data wise, IROS, which stands for Intelligent RObots and Systems, is the second big robotics conference, next to ICRA. It is a three-day conference, with one day of workshops each at the beginning and end, for five in total. This year, it received close to 1600 śubmissions, of which 936 were accepted, and there were 18 workshops and 2 tutorials.

936 out of 1600 is about 58% acceptance rate, which seems to be fairly typical for the larger conferences. However, packing so many talks into three days required 16 parallel tracks, so I could only visit a small proportion of those. I concentrated mainly on the Human-Robot-Interaction sessions, with a little bit of vision and learning thrown in.