I think they do this with some automated manufacturing robot arms and
What they do is have an identical (or scaled down) replica of the robot's
arm, with feedback sensors but no motors, and lightweight enough to be
manipulated by hand. During programming, the sensory feedback from the
training replica arm is passed to the actual arm, which mimics the action
exactly, and the training operator uses the arm to walk the system through
one complete production cycle, the complete list of motor commands and
relevant sensor feedback being logged by the computer. The training arm is
then disconnected, and the computer attempts to reproduce the sequence
repeatedly in full production. Obviously, if this were all that was needed,
a computer would be overkill, and a signal record and playback system would
be enough. However, it is quite possible to then add error-correction
algorithms and things to the computer so that, say if the arm overshoots, it
can recover the correct sequence and not end up performing its actions
relative to an incorrect origin after that point, for example.
Theoretically, even more complex artificial intelligence could be added to
analyse the required actions and improve on the original set of motions, or
come up with a better process altogether...
Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here.
All logos and trade names are the property of their respective owners.