Live from the AAAI Conference

[Jerry]

Hi Everyone,

I thought I'ld try something different.

I'm in San Jose, CA for the American Association of Artificial Intelligence National Conference. It is the main conference for AI researchers and academics. I thought it might be interesting to give a day-by-day update of anything interesting going on here. I'll also insert links for more info, like here is the link to AAAI: www.aaai.org.

Besides people presenting their most recent work in AI, there will be exhibitions of new systems, robotics competitions, AI companies showing stuff off (and giving away free stuff), and book publishers (also giving stuff).

In addition to the AAAI conference, the National Conference for Educational Robotics is here too. The NCER is host to the Botball Robotics competition (www.botball.org).

SIUE has a team in the collegiate level of Botball. The team members include: Ross Mead (CS), Carey Lizotte (CS), Brian Glass (CS), and Alex Narconis (ME). They called me late last night and were having major trouble with their light sensors. Apparently the robot is afraid of its own shadow. Well, they have one day of testing before they compete on Tuesday, so I'll keep you updated.

EHS also has a Botball team here at the High School Level. This team won the Midwest Regional in a decisive way. After losing their first match in a double eliminaion tournament, they went on to crush all their opponents by 10+ points winning 12 in a row. I have high hopes for them.

I'm going out for a little site seeing tour and I'll write more tomorrow.

If anyone has a specific questions please post them.

--Jerry Weinberg

[Jerry]

Sunday was day 2 at the AAAI Conference.

There are workshops and tutorials going on today. But these cost extra, so I decided to take a day off, do some sight seeing, catch-up with people I know from around the country.

Downtown San Jose is nice. I'm just across the street from Adobe's main head quarters. Which is interesting because I started typing in this message and had to start over because Adobe's update crashed my browser window  :-? .
Maybe I should go over there and ask them about that.

There is a nice pedestrian mall that leads from about where I'm at to San Jose State University. At one end is Caesar Chevez Park. There was a free concert. I'm not sure of the name of the band, but they sounded and looked alot like Los Lobos. Anyway it was good, the weather was great, and a good time was had by all.

One robot team is practicing in the lobby. They have a very sparse robot. It is a slim wheeled cart with a tricycle design: two drive wheels and one castor wheel. It's cpu is a laptop. The cart is about 4 foot tall, making it easy to program directly.

The only sensor I can see is a web cam. This is interesting. There is a push by some researchers to use only optical input for sensing. This means the robot would have to use each frame to detect obstacles, recognize landmarks, and detect visual odometry. Visual odometry means detecting the robots movement by means of optical flow. There is a scheduled demonstration of optical flow work later in the week. You can read a little about it here:
http://www-2.cs.cmu.edu/~rahuls/pub/aaai2004-demo-rahuls.pdf.

The team appears to be preparing for the robot presenter challenge. Here the robot is dropped off at the front door, it finds its way to the registration desk, registers for the conference, gets directions to the room it is giving a talk in, finds the room, and then gives a talk about itself.
You can read about this and other challenges at:
http://www.aaai.org/Conferences/National/2004/Robot/aaai04-robots.html

I'll tell you more about the challenges as I see them.

Ok, I'm going off to the conference. Today it is mostly activities associated with the National Confernce on Educational Robotics. Tonight is the openning reception for AAAI.

--JW

[Jerry]

I’m a day behind my reports due to a small tech glitch.

First the big news: SIUE Collegiate Botball Team took 1st place in yesterdays seeding rounds!  An impressive showing given the big name competition. Hopefully they can repeat their performance at tonight’s double elimination. Last night they were trying to convince the other teams to go out for beer and sushi hoping to get them drunk so they wouldn’t work on their bots tonight.

The EHS team did not fair as well. They had carried their bot onto the plane, but it wouldn’t fit in the overhead compartment. So they had to check it in. Well, apparently the baggage guys decided to play catch with it. The bot made it into Cali in nearly as many parts as it has lego pieces. The team spent most of their time putting it back together, but you know what they said about Humpty.  Their double elim is this afternoon.

KIPR is going to try a live feed of the tournaments. There is a link at www.botball.org . Remember all the times are Pacific time.

Monday was all about the National Conference on Educational Robotics.

This is a small co-located conference mostly dominated by High School and Middle School educators. During the morning I sat through a couple of sessions of talks from teachers describing ways they are using robotics to teach math, science, and language arts.

Talks were ok, but the most interesting to me was a talk given by two high school students who started a summer robotics camp for middle schoolers to fund their robotics team. Quite impressive. They hold 3  six day camps per year, 12 sudents per camp, $125 per camper. After re-imbursing the school for start up cost of $1500 all the funds are profit to their group.

This is something I would like to try and start-up for SIUE and EHS robotics teams.

The afternoon was Botball practice. Also there are a number of teams practicing for the other robotics events. More on that later.

In the evening there were two invited talks, Richard Simpson from the University of Pittsburgh on assistive & rehabilitation robotics and David Miller from the University of Oklahoma and NASA Fellow on recent design of off planet rovers.

Richard Simpson talked about the state of the art of intelligent prosthetics, particularly bionic legs, robots used for limb rehad after an accident or stroke, and robotic assisted wheel chairs â€ââ,¬Å" lots of cool videos.

One of the most interesting was a robot called Cosmo â€ââ,¬Å" it helps children with limb rehab. The basic idea is that sensors are placed on the child’s limb. Cosmo responds to movement of the limb dancing or playing Simon Sez or other games. This keeps the child interested in the difficult activity of moving the affected limb through the long tedious rehab exercises.

David Miller discussed some of his recent work on rover design. The most interesting part of his talk was how President Bush’s directive to return to the moon has effected NASA’s Robotics Division.

The communication delay to Mars is 7 to 40 minutes. At times Mars’ orbit takes it on the opposite side of the Sun from us making the delay about 40 minutes. The communication delay to the Moon is about 2.7 seconds. With a 7 to 40 minute delay rovers must be autonomous in their second-to-second control. With a 2.7 second delay, rovers could be more tele-operated. So NASA is focusing more on shared control systems that allows human intervention into the robotic control and less on pure autonomy.

David Miller also showed cool video of the new rover design he has been working on. The three Mars rovers used a six wheel design called rocker bogey. The design ensured that all the wheels remained in contact with the ground regardless of the terrain (within reason). His new design simplifies this to four wheels. The design uses a split joint differential between the front and back half of the rover. The back and front can move nearly independent of one another. Really cool stuff.

Well, more later,

JW

[Jerry]

Ok, I’ve been busy judging the robot competition but I’ll try to catch up.

First an update on our Botball Teams.

Our collegiate team, that calls themselves â€Ã...“Fishtank Assassinsâ€Ã,, won their first match decisively. Unfortunately it was all downhill from there. Their next match they lost to a team that had a spoiler bot on defense. Our team was all about offense. It was a single bot entry that went for the highest scoring item. The other team had a spoiler bot whose only intention was to block our bots path.

Our team’s final loss came from last minute code changes. They noted a winning strategy of another team who had a similar bot as their own. If you can’t find the basket to dunk the ball head for you end zone. It doesn’t score as high, but low scores were winning. So the team made a quick code change between games. Unfortunately there was no way of testing it. The bot grabbed the ball, eventually headed to the end zone, made it there but bumped the end of the table. This unfortunately knocked the ball loose and the bot dropped it off the end of the table.

Oh well, they had a good showing so if you see any of them, congratulate them, and ask them about it. Just a reminder our team was: Ross Mead, Brian Glass, Carey Lizotte, and Alex Narconis.

The EHS team faired better. They made it into the sweet 16 undefeated (out of 56 teams). I was busy judging so I didn’t get to see the details of their loss, but it appeared that they had a similar problem as the Fishtank Assassins. Their bot was all about offense, no real defense. It appeared that they lost to teams that had some sort of spoiler strategy. This was a particularly good showing given that they had to re-build the bot almost completely after it was smashed into the baggage compartment of the plane on the way here.

Everyone on the EHS team are juniors, so they have vowed to stay together and take next year’s nationals.

A few notes on some talks.

Though I’ve been busy meeting people and judging I did try to get to some of the talks.

Dan Clancy from NASA Ames Research Center game the opening keynote address. This was attended by about 2000 people. He had three themes:

1. Systems integration is as important to AI as basic research. He noted that most of what AI researchers do results in algorithmic innovation. He had a nice list of what he called â€Ã...“AI Everywhereâ€Ã, such as suggestion systems of Amazon and other web retailers and search engines such as Google. But for AI to continue to advance there also needs to be architectural innovation. How do we get our programs and methods to work together?  A major problem is that our systems have different semantics: two AI programs may both be able to reason about complex ideas such as program debugging, but they do it in such different ways that they cannot share results.

2. AI needs to begin thinking about Software Engineering. We need to get to the point that none AI experts, that is just any programmer, can write or put together an AI system.

3. We should be researching shared controlled systems. He particularly focused on human-robot interaction because that is the area of NASA he works on.

Besides these themes Dan Clancy showed some current work and recent success at NASA. One of the more interesting points is that the next Mars Lander that is scheduled for 2009 will be nuclear powered and they expect it to operate on the order of 4 to 5 years.

He also showed a mobile agent project for astronauts. When we send people back to the Moon they will be involved in building a habitat. NASA is working on an intelligent agent that essentially is a robot on your back. It is a back-pack with sensors that provides assistance and helps the astronaut watch out for problems or dangers.

Robot Gopher

Sebastian Thurn from USC gave a talk on work he is doing in Pennsylvania to help map out abandoned mines. Apparently there are hundreds of old coal mines, many that are on fire, where maps no longer exist or are severely incorrect. It is important to know where the mine shafts are to help prevent injuries from building over a collapsed mine or lung disease from building near a mine on fire (they can burn for years). At USC they developed a 3-D mapping system that uses two lasers. So they are working on autonomous control robots to navigate the mines without putting themselves in danger. The reason they are focusing on autonomous systems is that wireless communication is lost fairly quickly underground, and these mines can stretch for miles. In fact it is a very dicey proposition because they essentially send the bot in and soon loose communication. So they have to hope it comes back out. Still very early work, but some cool pics.

As part of the talk he presented an interesting result between metric mapping and topological mapping. Sebastian Thurn is well known for his ground breaking work in robot mapping with a metric representation. That is the maps are represented by hundreds of thousands of points. This has many serious limitations because the methods require a global representation. An alternative is a topological map that represents landmarks and their relative position to each other. Its like when we give directions to someone we say â€Ã...“go about a mile and you’ll see a Quick Mart, turn left and go to the next stop lightâ€Ã,. This representation allows for better scaling because it is all local information, but it not as reliable because it is at a higher level abstraction to what the robot senses.  Sebastian Thurn showed how metric representations can be mathematically converted to a kind of topological representation.  Its not a complete representation because all it can tell you is that obstacle A is near obstacle B (no way to identify them and it wasn’t clear if there is distance and direction data between them). Still, this may open up an interesting hybrid system that uses both topological and metric representations.

He also talked a little about their entry into DARPA’s grand challenge: have a robotic vehicle that can maneuver 150 miles across open country. You can check out their entry at www.stanfordracing.org .

I also saw some cool rescue robots.

One was essentially a radio controlled VW bug with mounted stereo cams (two sony web cams), a snake like robot with oval wheels that made it possible to get over debris, a simple robot that had a PDA as its cpu, and a robot that was jointed between the left and right sides with a bar in the middle that could rotate 360 degrees about its axis. The operators used the bar to help flip the robot up and onto high places. It was really cool. However, the design only allowed for a single camera as its only sensor. Plenty of standard robots too, like the ones in the robot lab.

I’m really being strongly encouraged by the organizers to have a rescue robot entry next year. If anyone is interested send me an email. To be competitive though we would need to start working this fall.

Well, I’m on my way to the airport. Lots more interesting stuff, so I’ll try to have one more post when I get back.

--JW

[Jerry]

My last update post.

I'm back in the office. There were a lot things I didn't get a chance to report on because it was a busy week for me.

For example the academic argument that broke out about the nature of intelligence and its impact on building large integrated intelligent systems. One camp held to the theory of multiple intellgences. That is there is no real general intelligence rather we are made up of very specific intelligent areas that borrow from each other. The other camp holds to a general intelligence. This group argued that research should focus on developmental theories of intelligence and that learning is the key to interconneting knowledge rather than engineering knowledge directly into programs.

Anyway, you might be asking yourself "how can I go to next year's conference to check out this stuff myself?".

Well, each year AAAI seeks out student volunteers to help put on the conference. You volunteer 6 to 10 hours of your time and they will pay for your registration. Student housing is usually availabe at $30 per night, and one meal is free. So the main cost would be transportation.

If you are interested next year's conference site is at:
http://www.aaai.org/Conferences/National/2005/aaai05.html

There are also other ways of going such as doing a project and submitting a student abstract.

--JW