in Which Robots Rule the Post

In this post I was planning to talk about the career fair last Friday at which some fellow students and I pulled a great prank (with permission, of course), but then we decided to do the same thing on a bigger scale later so we're going to keep it hushed until then (read: check back in later for an awesome story). Robbed of its primary focus the post became very contrived and a rather dull discourse on how great the career fair is, ra-ra-ra, go Mudd.

That's boring.

Instead, let's talk about something cool: robots. With lasers. Got your attention yet? It only gets better. I'm taking CS 154, robotics, this semester and I have to say, building robots is just as cool as you hope it'll be. First off, the class is taught by Professor Zach Dodds who is just about the nicest person you could hope to meet. Not a particularly hard teacher and he doesn't give the same intimidating impression of towering knowledge and experience as some of the other professors here, but a really nice guy who really knows his stuff -- many of his stories are about robots he built in his college days that didn't end up working, but don't let that fool you: Dodds is a smart man with boundless energy. Dodds' research interest is in computer vision, which couples tightly with robotics. He's got a working NES in the lab with a Duck Hunt cartridge. The Duck Hunt light gun is strapped to a pair of servo motors that can point it at any spot on the screen and they are connected through software to a camera that watches the screen and controls the gun to play the game. It was built by some of his students several years ago and this year there are two teams using the Wii Remote as part of their projects.

He's also interested in cheap robots. While his robotic idols are Stanley and Boss, the autonomous cars that won the DARPA Grand Challenge and DARPA Urban Challange, he likes working with students on projects that cost very little, and still work well. A classic example is using a webcam and laser pointer as a low cost rangefinder -- true laser rangefinders cost thousands of dollars, but by using an off-the-shelf webcam and a cheap laser pointer it is possible to get very accurate range data. Couple that with some servos and you have quite a good ranging system. I told you I'd work lasers into it.

Just last summer three girls, all rising sophomores, did research with Professor Dodds and entered a competition. Their goal was to build a robot that would autonomously seek out and "capture" 6 colored markers. The capture was done by driving up to a marker and reading off the sequence of colors that made up the marker. They threw together an iRobot Create (the Roomba platform) with a Macbook on top of it and some legos holding up little sonars and a camera. I think the whole thing (except the macbook, which one of the students already owned), cost under $200. Almost all of their competition was teams of graduate students and the next cheapest robot was somewhere in the realm of $6,000. Our team of sophomore girls won the competition. And let the bruised egos of the mighty come crashing down.

I mentioned that I was taking the class, but I have yet to mention my project. My team's project is ambitious...possibly too ambitious. We're entering the Semantic Robot Vision Challenge. The goal is to build a robot that can receive a text file containing a list of objects and search for images online of those objects. Once it has a bunch of images it has to autonomously rove around a testing environment which contains many of the objects from the list, taking video or pictures of its surroundings. After touring the environment it has half an hour to sort through its video and select one frame that has each object in it, declare what the object is, and place a bounding box around it. So far we have a laptop on wheels. On the agenda this week is integrating the motors and vision together. For that we need to get the computer talking to the camera and some servo motors, then write software that will make the camera center on a region of a particular color using both the servos mounting the camera and the wheels turning the robot. Since we haven't gotten our short-range IR sensors or sonars working yet we can't really make it autonomous, but we're going to work on making it capture images and process them against a small database.

I'll keep you posted as this gets further along.

~KMarsh