the biggest problem in robotics



MESSAGE RECORDED <03.12.02015>

Hey. Very interesting to meet you.  Keep me posted on the other links that you were talking about.

Yeah, I remember that I was thinking about the spine, and how it was curved (and why) - and was much more graceful than anything we've created in the world of robotics.  I started looking into why that was (back when I was working at the RI at CMU) in 2010.  I spent a lot of bunch of time mulling over it, (as well prestressed structures such as the bicycle wheel).  I started looking into the artwork of kenneth snelson, and buckminster fuller, and then I had a eureka moment back in 2012 where it became blindingly obvious to me that this was the future of robotics.  We basically have most the basic building blocks to create living creatures - we have ways of metabolizing, storing, and acquring energy.  High density concrete has very similar properties to bone; carbon fiber can make tendons; plastic & silicon can be mechanically equivalent to skin and cartilage.  There are a host of sensors (cameras=eyes, microphone=ears, chemical sensors=tastes), and most things that exist in biology have some imperfect analog (or better) in machinery.  Everybody focuses on the difference computers and brains, but mimicking the way animals move is so far beyond us because we just can't move the way they do.  The reason robots "seem robotic" is because we just don't understand the principles of our own motion.

I realized that the biggest problem in robotics was (surprisingly) in something as basic as actuation.  Muscles have two equally important properties - they can independently control tension as well as their amount of deflection.  I realized that the reason most robotics failed was because they attempt to approximate two orthogonal axes of control with one.  Part of the reason this is so counterintuitive, is because we have an animalistic bias towards thinking of strength as coming from being monolithic or rigid.  It's actually the opposite - strength comes from being flexible, adaptive, dynamic, controlled ...

I started imagining robotics systems, and mechanical systems purely in terms of tension and compression, and it was a really enlightening way of looking at the world.  It felt a bit like the first time I discovered object oriented programming, where exactly the same thing is happening under the hood, but you think in ways completely differently.  I couldn't unsee it, and nobody agreed with me (inside or outside the RI).  About a year ago I spent a couple of months researching it, and I designed a few motors / linear actuators, but they weren't very good, and it really seemed to me that the real solutions would come from material science, which I unfortunately don't have a background in.

The ideal solution, is a string, whose which can be induced to change its length (and knows how long it is).  It can be induced to contract, requires no energy to maintain it's deflection, is flexible, can be controlled individually (for redundancy), has a high tensile strength, is highly efficient, stores its own energy, and is dirt cheap.  In short - we need a mechanical equivalent for muscle.

In the world of robotics, I'd say we're somewhere near where computers were in 1978.  Once the problem of actuation is solved, I believe that some seemingly insane things will be possible.  You could have a skeleton made of thin plastic rods, all of which weigh less than a pound.  When you flick a switch, they can tense up their conneting strings, and stand up, or dance.  You could have an spider the size of a building, that can walk across any ground.  You could solve inefficiency of walking around, by piloting various "shells" and walk around the world as 8 different robots at once ... 

Anyway, pleasure to meet you - its great to finally get some validation on this.  You're the first person I've met who has shared this insight.  Great minds think alike :)