This is vid of my simple stickbot -- with modified hind legs ;- -- performing some simple tasks.
The idea behind the exercise is to illustrate the adaptability of even simple "clever" s/w.
The robot runs a problem solver-based "operating system" that can sometimes find a sequence of operations to perform the current goal. It has built-in procrastination (i.e. if the goal is too far away, push it and try for an intermediate goal for now), generalisation (if it looks similar to something I've done before, it is) and differentiation (if X worked in situation A but not in B, then the difference between A and B becomes a precondition for doing X in future).
Solving simple tasks is a co-operative effort. If the robot fails to do X, the human operator can also select a simpler task that will lead to X, and get the robot to do that first.
This is also known as "learning" of a simple form.
Apart from adding the 2 dof "feet" to the stickbot, the only change needed to the simulated robot s/w were additional axioms to handle standing on the hind legs and standing on 1 foot.
In the original stickbot there was a rule "you can lift leg X if the other 3 are on the ground and rhe robot centre of mass is sufficiently inside the triangle thus formed".
In this "footbot" that axiom was generalised to "you can lift X if the robot centre of mass is inside at least 3 other widely-spaced points (even if the points are the corners of 1 foot)".
Both. It is a program, but it can solve problems it hasn't seen before. Look into the area in AI called "planning".
kymhorsell 3 years ago