"Software will predict robot behavior before they are built"

erco

http://www.rdmag.com/News/2011/11/Information-Technology-Robotics-Accelerating-Robotic-Innovation/?et_cid=2314391&et_rid=54725820&linkid=http%3a%2f%2fwww.rdmag.com%2fNews%2f2011%2f11%2fInformation-Technology-Robotics-Accelerating-Robotic-Innovation%2f

Good luck with all that! Have these people ever tried to get a BoeBot to drive straight down a hallway? Or to get a troupe of Toddlers dancing the Hokey Pokey? When's the last time you built a robot whose initial behavior even mildly matched your intent? Robotics is about mechanics: backlash, deadband, wheel slip, motor variations, error, rugs, dust, friction, obstacles, irregularities, and general unanticipated randomness. I have to chuckle at the idea of Robotic "simulations". The real world introduces so many new variables that I can't believe that anyone seriously thinks that a simulation can represent or replace any real testing.

Sure prototyping and testing is expensive, but it's money well spent. These ivory tower clean room types can do all the simulations they want, but they will still get ugly surprises when they initially test their "virtually optimized" super bot in the real world. But it's unlikely we'll ever hear that follow-up story!

lardom

Quote by erco
When's the last time you built a robot whose initial behavior even mildly matched your intent? Robotics is about mechanics: backlash, deadband, wheel slip, motor variations, error, rugs, dust, friction, obstacles, irregularities, and general unanticipated randomness.

I can stay up late trying to correct errors in my logic and I know what magic smoke smells like.

Martin_H

Hmm, software is a major portion of robot behavior, and sensor data controls execution flow. So in effect they're saying that they can have one program predicting the output of another. That sounds suspiciously similar to the halting problem. Pretty much the only way to know the output of a program is to run it.