Rodney’s Robot Revolution
There was a pretty awesome TV special a few nights ago- Rodney’s Robot Revolution. It was on “The Nature of Things”. Here’s the summary they give on their site:
As former head of the Massachusetts Institute of Technology, Computer Science and Artificial Intelligence Lab, Professor Brooks is once again going to test his robot-building mettle with what has become the most challenging project of his entire career: to build a robot for the Pentagon capable of playing an ancient Chinese board game of GO against a human opponent. Brooks has only five months to build it from scratch, making his challenge that much more outrageous.
The premise of the show surrounded Rodney Brook’s project for the government- to build a social robot that will play go…. in nine months! (For some reason the CBC website says 5 months, but I clearly heard 9 months, three or four times).
I think he did have some students help- at least he did with the hand part that needed to pick up the pieces. The hand consisted of two fingers, with special force sensing resistors on them. The fingers was pretty good at first- it worked often. Afterwards, they placed a cover on the two fingers with bumps on it so that the piece would remain grasped until placed onto the board. It didn’t work often- it couldn’t pick up the pieces.
The other main parts of the robot was the computer vision. It used two off the shelf logitech cameras as eyes, and it had a few more stationary cameras around to get different views of the board. The eyes were mounted on a metal part that served as a face, which could move. At one point, the robot almost fell off the table because the movements were too quick, and not fluid.
The only part that seemed a little wonky was when they mentioned that robotics is a competitive industry. The jist of the message was that other places like Google, Sony, and Honda are competing with MIT’s robotics. I kind of laughed at that… as MIT is a university, not a company. They weren’t just comparing an apple with an orange, they were comparing an apple with a pineapple. Either way, Sony’s Aibo robot is a distant past, and Asimo has sort of been around for a long time- just evolving and improving! As for Google… errrrr. I thought we were talking about physical robots here? Lol! I think what they meant to portray was that there are other places that are R&D’ing robotics.
The part that I found quite interesting was that even though each of the “sections” of the robot worked okay- like the hand, the computer vision, the algorithms, they weren’t combined to work with each other very well. What I mean by that is… there was too much focus on creating the basics perfectly, instead of just building a simple model first, and improving on it with milestones. If it would have been done this way, there would have been something to demonstrate at the end that would work. It’s better to have something that works, even though it doesn’t meet the requirements, than something that meets the requirements but doesn’t work.
First start off with giving the robot the piece, and start off with all the cameras stationary. The robot can still be sociable even if the cameras aren’t in the eyes (for some reason, this seems to be a common misjudgment). Once that stage is working, make a hand that will have to move the piece. Then make it pick up the piece. After that works, then go on to moving the cameras, and finish up by adding a sociable touch.
It’s almost like the difference between the old macbookpros and the new ones. The old ones were created by putting pieces together, the new ones are created by taking material away. My suggested method is like taking all the possible difficulties away, one by one. If you only put pieces together, you’ll miss out on many opportunities to improve and optimize the robot. Additionally, there were no backup plans. If the robot didn’t have a pebble, it wouldn’t go back to pick another one up. It happens all the time in games between humans… the pieces are small and glossy, seriously!
9 months (or even 5 months) is a lot of time- especially when you have all the resources, and people that are eager to help out.
If I would have built the robot, I would have used an octopus-like suction cup to pick up one of the pieces. The location of the cameras would have been different… and the base would be bigger so it wouldn’t tip over. Plus, my method would have been different, as discussed above.
There’s another neat TV show soon, Five Years on Mars! National Geographic channel, Sunday Nov 2!
Don’t forget about Daily Planet too… my fav TV show that’s on at 7pm each weekday (Discovery Channel)