30 Oct 2008

I am currently on the BNAIC 2008 conference (The 20th Belgian-Netherlands Conference on Artificial Intelligence) in Bad Boekelo.

As I was (along with several others) greeted with a sign at the train station for the first time in my life, I might as well also engage in this new thing they call live-blogging. Here are some first impressions from the first day (I present my paper tomorrow):

In general:

  • The community of AI researchers here seems to be very nice and interesting. I'm looking forward to taqlking to even more people.
  • Some people have these new, fancy mini laptops. Nice... I want one (of course)
  • loads of formulas and text on the slides. Ten sessions in one day are hard to follow this way. The importance of presentation should be known by now, but here the scientific community is late.
  • Everything happens in a nice resort in the dutch countryside. Maybe I play some pool later :)
  • They have free Wifi. And it's completely open (granted, here in the countryside there a few abusers)

Here are a few notes of things I found interesting today:


Nao, the Robot

A french startup company managed to build a robot so versatile and so programmable that it got chosen to be the new platform for the RoboCup robo-soccer world contest. And - they brought some to Bad Beokelo to play. He walks like a robot god, has hands, Wifi and is designed for looks very well.

Isn't he cute? Notice especially how he gets up after being knocked over (but first he says: "I'm okay. I think I'd better get up"). The motion looks fabulous. He has 25 degrees of freedom and is bipedal, helped by an inertia sensor.

He is not smart just by itself, but very easily programmable, as I saw live. See the company website for more. On the dinner tonight, we had a band playing folklore music and they put a Nao on stage and he danced a japanese dance someone programmed for him. A-dor-a-ble!

The guys said they already used Nao with autistic people. Maybe it's also nice for elderly people and children, but first it will be sold to researchers, since it still gets developed.


Keynote by Ruth Aylett: "Interactive Storytelling - emergent narrative or universal plans?"

Her group works on Interactive Storytelling. It tries to integrate common planning theory with all these ideas.
What are the necessary ingredients of a story? You need a world and people, and they have to change each other. And a dramatic trajectory.
The levels of a framework for this are the media used, the plot (the tale) and fabula (everything that actually happened). In interactive storytelling, all levels are updated simultaneously. This needs heavy planning.
They also use "Emergent Narrative" as a bottom-up approach. To only have interesting characters is not enough for global goals... (but they can take part of the burden of generating new plans for actions).

Nice sources:
Riedl & Young: Plot repair (when user threatens plan). Is repair believable? If you want to shoot Robin Hood and try 20 times, and the system makes you miss each time because he can't die, that's not believable.
Mateas & Stern: Library of Story fragments: Facade Beats
Cavazza: Plan Trees

Goals are not enough for drama. You need emotions. Being rational alone is not interesting (drama). They elicit world-directed and/or emotion-directed responses. Ortony et al (1988) have a nice grouping of emotions.

Mihail  Mihaiylov: Collective Intelligent Wireless Sensor Networks

In Sensor Networks, nodes decide themselves what to do and they have to be careful about their energy. A network should have good data responses, but also live long - a tradeoff.

A waste of energy are the following situations: Idle listening time, Overhearing, Collision of signals, control traffic. Just going to sleep sometimes to reduce the listening cost is bad for response times. Basically, the intentions of single agents  (live longer) are bad for the system.

Solution: A node only cares for a set of "affected nodes" - i.e. it doesn't care about everyone, but only about those whose cooperation you'll need soon (because they route your signals to the base station). Learn their energy level (yes, that's control traffic, but not much).
Nodes learn their own tradeoff between sleeping and taking actions. As a result, the system is then a little less responsive, but the lifetime is much longer.


Update 01.11.

I gave the presentation and everything went well. I got a lot of questions and some were really interesting for further research. I uploaded the slides.

Also, I must say I noticed something else: While the people here in Netherlands are pretty diverse skincolor-wise, this wasn't the case at the conference. All in all it was 155 people: a lot of white men, some white women and three Asians

# lastedited 01 Nov 2008
You are seeing a selection of all entries on this page. See all there are.