A week of ICRA'10

Posted on

The first week of May I was in An­chor­age, Alaska at the 2010 IEEE In­ter­na­tional Con­fer­ence on Ro­botics and Au­toma­tion (I­CRA’10). In short, it was an awe­some event amidst great scenery, but for my tastes it was way too large and too far away. (The US cus­toms of­ficer even asked me why they de­cided to hold the event so far away from the rest of the world.)

View from con­ven­tion center bal­cony
View from con­ven­tion center bal­cony


There were roughly 1,600 reg­is­tered at­ten­dees, so the event was spread out over two big con­ven­tion cen­ters. All of An­chor­age’s ho­tels were packed with ro­boti­cists. It wasn’t long until local cafe­tarias were playing into this.

An­chorage wel­comes its new robot over­lords
An­chorage wel­comes its new robot over­lords

More than 2,000 pa­pers were sub­mit­ted, of which 857 were ac­cepted. While that’s an ac­cep­tance rate of only about 40%, it’s still one heck of a lot of pa­pers! In­stead of ac­tual paper pro­ceed­ings we re­ceived a booklet called the “Con­fer­ence Di­gest”, con­taining a single slide for each pa­per. With six slides on each side of each page, this thing is still about the same size as the full pro­ceed­ings to SoCG or Eu­roCG. The ac­tual pro­ceed­ings came on a DVD, and my tablet PC lacks a DVD drive, so I have not been able to look at them yet. But I pre­sume that there is still a lot of space left on the DVD, so I wonder why there is a 6-page limit on the sub­mitted pa­pers. For the ini­tial ver­sions it makes sense, as this re­duces the re­viewing load, but why not allow larger final ver­sions?


Image from the slides of my talk
Image from the slides of my talk

To ac­co­mo­date all these pa­pers there were thir­teen par­allel tracks; but re­ally, it felt more like thir­teen par­allel con­fer­ences. Ro­bot­-building dis­ci­plines like me­chan­ical en­gi­neer­ing, elec­trical en­gi­neer­ing, and com­puter sci­ence come to­gether with ap­pli­ca­tion areas for ro­botics such as the life sci­ences, ge­ol­ogy, lo­gis­tics, and man­u­fac­tur­ing. It seems like you could present re­search on just about any topic (“X”) at ICRA, pro­vided you word it the right way (“X using ro­bots”). Heck, I at­tended one talk that went into an­thro­pol­ogy! While it was nice to get a glimpse of what is hap­pening in other fields, and how they all re­late to one an­other, most speakers as­sumed the au­di­ence al­ready had in­ti­mate knowl­edge of their spe­cific field and its acronyms. I felt a bit left out, most of the time, and won­dered whether that de­scribed half the au­di­ence or just me.


Willow Garage's PR2
Willow Garage’s PR2

On Monday I at­tended the work­shop on Best Prac­tice in 3D Per­cep­tion and Mod­eling for Mo­bile Ma­nip­u­la­tion. My main take-away was that we’ve come far­ther than I thought, and that the state-of-the-art is all freely avail­able as open-source li­braries. You just down­load the ROS robot op­er­ating system and OpenCV com­puter vi­sion li­brary, and you’re all set. In a couple of lines of code you can com­bine a stereo image into a single image where you know the dis­tance of each pixel to the cam­era. A bunch more lines and you can build up a fully tex­tured 3D model of the world from a stream of these im­ages, cap­tured as the robot moves around.

Of course, that’s when the tricky part starts of de­ter­mining what all the ob­jects you are seeing are, and what you should do with them. This is where the open prob­lems are. The rest of the work­shop had some talks on these. All were in­ter­esting tech­niques for dif­ferent kinds of spe­cial cases, but we are still nowhere near dealing with the full va­riety of ob­jects hu­mans deal with on a daily ba­sis.

Tech­nical ses­sions

Tuesday till Thursday were ex­haust­ing. A 12-minute talk every 15 min­utes from 8:30 to 19:30, ex­cept for breaks and the oc­ca­sional ple­nary or keynote speakers who had longer talks. The ses­sions ended ear­lier on Wednesday be­cause of the con­fer­ence ban­quet, but that’s still more than 70 talks I at­tended. For many of them I only vaguely un­der­stood the problem that was being solved, and usu­ally had no idea what the state-of-the-art was and how the pre­sented method im­proved on it. Still, there were a couple of talks I found re­ally en­joy­able.

  • My own talk was on pushing a disk with a disk-shaped ro­bot, and it turns out there were two more talks on the sub­ject of pushing (though all in dif­ferent ses­sions, for some rea­son). In A Di­pole Field for Ob­ject De­livery by Pushing on a Flat Sur­face, Takeo Igarashi pre­sented a very simple and el­e­gant al­go­rithm for the same problem I looked at, and even showed how it worked with ac­tual Roombas. The al­go­rithm cannot (yet?) deal with ob­sta­cles, but I had a nice chat with him af­ter­wards about fu­ture work. In Dy­namic Pushing Strate­gies for Dy­nam­i­cally Stable Mo­bile Ma­nip­u­la­tors, Pushkar Kolhe (how awe­somely ap­pro­priate is that name?) studied how a dif­fer­ently de­signed robot should push or pull in order to exert the most force. It turns out that pulling is ac­tu­ally better than push­ing.

  • In Adap­tive Mul­ti­-Robot Co­or­di­na­tion: A Game-The­o­retic Per­spec­tive, Gal Kaminka talked about a new pro­tocol for moving ro­bots to avoid col­liding with each other. They could easily have had the paper ac­cepted after just showing through ex­per­i­ment that their method works, and I would have thought nothing spe­cial of the talk. In­stead, they went the extra mile in using game theory to also show why the method works.

  • In Avoiding Zeno’s Paradox in Im­pulse-Based Rigid Body Sim­u­la­tion, Evan Drumwright ex­plained how physics sim­u­la­tion li­braries such as ODE and Bullet get resting con­tacts wrong, and presents an al­ter­na­tive method that has been im­ple­mented in a new physics li­brary called Moby.

  • My fa­vorite talk ac­tu­ally won the best paper award in its cat­e­gory. In Ges­ture-Based Hu­man-Robot Jazz Im­pro­vi­sa­tion, Guy Hoffman pre­sented his jam ses­sions with Shimon, a marim­ba-playing ro­bot. Rather than play a pre­pro­grammed piece of mu­sic, it im­pro­vises based on what the human is play­ing. The re­sult sounds great, and the ad­di­tion of a head that head­-bangs to the beat was a nice touch. It re­ally looks like it gets into the groove!


On Friday morning I went to the work­shop on Guar­an­teeing Safe Nav­i­ga­tion in Dy­namic En­vi­ron­ments. All great talks on how to avoid col­li­sions among moving ob­sta­cles, each il­lus­trated with pretty videos. Jur van den Berg showed an es­pe­cially im­pres­sive video on sim­u­lating human move­ment in crowds.

Friday af­ter­noon I at­tended the tu­to­rial on Re­al-Time Plan­ning in Dy­namic and Par­tial­ly-Known Do­mains. It em­pha­sised how al­most any plan­ning problem can be re­for­mu­lated as finding a path in a graph where edges are la­beled with a cost and/or prob­a­bil­ity. The ob­vious way to do that is using A* search, but I had no idea how many dif­ferent vari­ants of A* have been de­vel­oped to deal with dif­ferent kinds of prob­lems. Al­most a dozen of them were ex­plained, having funky names like Fringe-Saving A*, Life­long Plan­ning A*, and Any­time Re­pairing A*.


The A* tu­to­rial at the end was ac­tu­ally the first time during this con­fer­ence that I saw someone use the­o­rems and com­plexity the­ory. What’s per­haps a little dis­turbing is that that ac­tu­ally made me feel re­lieved, as if I had come home from a long ar­dous journey through the waste­lands. That’s not to say that the other talks at the con­fer­ence were bad, but they were rather dif­ferent from the kinds of talks I’m used to. I think com­pu­ta­tional geom­etry may have spoiled me a bit in that re­gard. I’m used to talks with clear, pre­cise de­f­i­n­i­tions and the­o­ret­ical analy­ses. It’s easy to forget that in the “real world” one deals with vague con­cepts and must rely on ex­per­i­mental val­i­da­tion.