How can you teach a Robot how to Juggle?

Question

greater or equal than 3 balls.
what goes through your brain when juggling and how can, if ever, a robot learn this?

How would it cope with errors?

Answer 1

This sort of task can be approached in at least two ways - one would be to develop 'self learning' software - the other by 'pre-programming' the solution.

Since you ask how I would do it, given the very limited nature of the task I would go for the second method.

I will assume the use of 'arms' and 'hands' similar to human shape & size (with 'touch' sensors) and a binocular vision system.

Step one should be 'dead easy' - using 3 balls of exactly the same weight & shape in 'still air', each ball can be thrown upward in exactly the same way, so will follow the same 'path' through the air and so should be 'caught' at exactly the same position as it comes down. In fact it may not even be necessary to employ the vision system at all (can a blind man juggle ??)

ADDITIONAL.
To cope with minor errors the vision system would need to compare the upward path of each ball (as it is thrown) against the 'ideal' path in order to predict where it will come down & thus where to move the 'catch hand'.

After each error I would get the software to modify the next 'throw' (of that ball) in an attempt to compensate for the error.

Answer 2

The way people do it is with very sophisticated sensors and very fast processing of those sensed signals.

A robot would learn it if you could write the algorithm required to sense ball position, capture and launch the ball on the correct path for the next cycle.

Good luck. I think that you will have to wait a while on both sensor technology and computing power.

Your robot will also have to have binocular vision (3D, you know...)