I've been exposed to the robotics academic world in depth. Even academics I know are growing tired of the academic mode of work. There is a persistent and growing respect for making real and complete systems. This it often antithetical to the unit of academic progress: published results.
A great test for whether you have a real system is whether you can sell it. That's why iRobot is so awesome. They make robots that people actually use.
Some other efforts that come to mind help illustrate the territory. Honda made a horribly unscalable walking mechanism in the ASIMO. It will never be more than PR. The best work from CMU & MIT's Leg Lab got spun out into Boston Dynamics. You've probably seen their "big dog" video.
http://www.bostondynamics.com/content/sec.php?section=BigDog
Boston Dynamics is part boutique research firm and part startup. They'll probably be the first to sell a walking robot.
Honda makes a clunker, and the hungry, small companies make something that will actually work.
I read a recent Popular Science article where it says Trevor helped with the unicycle motorcycle. This article here is essentially what it said in the Popular Science article; so it seems Trevor is quite the go-to guy when it comes to these types of things.
A great test for whether you have a real system is whether you can sell it. That's why iRobot is so awesome. They make robots that people actually use.
Make something people want :)