Video of robot that balances an inverted pendulum
Eric says
LinkThis video shows a balancing robot moving with such fluidity and apparent emotion, that it almost seems alive.
Magnus Linderoth and Kristian Soltesz, students of Lunds University in Sweden, posted a video last winter of their "inverted pendulum" robot. I only found it recently because I saw a post in MAKE about double pendulums and searched for more in YouTube. The robot in the video tries desperately to keep a rod balanced in the air like a broomstick. As the researcher meddles with the robot, it responds with various tactics.
the latest
latest episodes
It looks alive because our motor systems are partially governed by feedback control mechanisms, of which the inverted pendulum balancing algorithm is a simple example.
I saw this same set up, only bigger, 10 years ago when I was visiting the University of Washington for one of thier engineering open houses. I wish I knew more details about who was involved but it was a very long tima ago.
The grad students would play with it the same way. They would give the pendulum a tap and watch the machine scramble to keep it upright or flip it back up when it did fall over.
While this is neat to watch, I'll note that this theoretical problem (the 'inverted pendulum') is a classic problem in Controls textbooks, either from advanced undergrad or regular grad engineering courses.
That said, it's definitely harder to build the actual thing than to merely do the math behind it for a textbook problem. Some classmates of mine did this same type of project in our grad-level controls course about 2 years ago, and there are always subtle complexities/nonlinearities that come into play when building the real thing such as friction in the components or the delay inherent in the system (mmm... feed-forward control). It's always cool to see when those are overcome in the final project.
Oh yeah, also note that the real-life problem of starting from a dead stop and swinging up the pole is quite non-linear, whereas the textbook problem involves assuming you're starting with the pole up and linearizing around that point. i.e., the text problem is much simpler, even without taking the other non-linearities into account.
So yeah, this is cool. Not new, but still cool.
My final undergrad course (1994) in electrical engineering at Clemson University was to design and physically build an inverted double pendulum like that shown in the MAKE article that would swing up and balance like this, where the actuation was applied to the upper portion (upper portion when inverted). A belt connected our motor shaft to the lower pendulum piece to actuate it. I'll have to dig it out and post the video. It took 3 people a full semester to design and build it. The trick was money. Theoretically, you could use analog electronics to put together the controls for very little, but practically we dropped $1K in digital electronics to control it with a computer.
Step One: Invert the double pendulum
Step Two: Attach an inverted pendulum balancer to the now-inverted double pendulum and another at the joint
Step Three: Sit back and watch two pendulum balancers fight to keep the thing upright.
Optional Step Four: Triple Inverted Pendulum*
*this sounds like a sex position.
Yeah we did the same sort of thing as a project in our "Methods of Experimental Physics Course". We wanted to teach a neural network to do the balancing, but our aparatus was so weak we ran out of time getting a basic algorithm to work...
Video here:
http://youtube.com/watch?v=2zaigbwpX04
The first example I saw of this was David Anderson's nBot. His site has some impressive videos of 2- and 3-wheeled robots balancing an inverted pendulum and rolling down some bumpy hillsides.
Here's another video from classmates of mine; this time the whole thing's made of LEGO. http://video.google.com/videoplay?docid=5412705235458687377
My final design class as an electrical engineering student at Clemson University a few years ago required the creation of this "pendubot":
http://www.youtube.com/watch?v=Lfh9upwZWA0
The first arm was directly attached to a motor on one end and on the other end was a free arm connected by a digital encoder (angular position sensor), feeding back to the motor.