CMU’s BallBot - Scientific Advancement or Just-Another-Robot?
On Friday CMU announced via press relase the creation of the "Ballbot" (original, I know). It’s a robot that dynamically balances on one steel ball. It was picked up by several robot websites and was also featured on RocketBoom.That’s all fine and good, until someone cries "fowl!"
I just want to note that they bought their own pre-assembled IMU. That takes nearly all of the challenge out of this project.
Theo goes on to say:
Don’t believe me? Take a look at the papers. The entire control system is simple LQR. They’ve added some PI control to correct for some frictional effects, and both coefficients were tuned experimentally.
Translation: It’s not that hard. A PI controller takes about four lines of code, and experimental tuning involves grunt work, not real science. Here is a link to the IMU they use.
While I’m not sure if I’d be as harsh as Theo is with them, I definitely don’t think it was worth all the hype or the press release. Come on CMU, show a little humility!
Update: Theo, made some good comments below.
- Ballbot website
- Robotics Research Symposium paper (2005)
- IEEE Robotics And Automation Conf. paper (2006)
Read Similar Posts:
TrackBack URL
June 9th, 2008 at 9:16 pm
Well, they do do a little bit of science in the design of their LQR controller. They build a simple model of the system and show that their controller is a valid controller for a system modeled as such. The experimentally-tuned PI controller was just added afterwards to get rid of some frictional effects. (the “I” is the important part; it serves as an internal model that deals with the frictional damping by accumulating extra control energy)
However, none of this is cutting edge control science. A truly sophisticated handling of this robot would include an input (what they built was purely for stabilization/regulation; it is not meant to track an input) and build an appropriate nonlinear controller. (note: for an inverted pendulum like this, the nonlinearities aren’t that great. Gravity adds a sinusoid function, which is globally Lipschitz and looks pretty darn linear when you’re very close to the equilibrium; thus, this problem isn’t even sufficiently nonlinear to be interesting)
So some real science was done here… But it was classroom science. There was no contribution. Nothing *new* is being done here.