Learn Plans
- CABot3 planned by a neural implementation of a Maes net.
- This had facts from the environment, goals, actions, and constraints.
- All were programmed as orthogonal CAs.
- We're going to do two relatively simple things, and then one open
ended complex one.
- First, we're going to change the associations between the Maes CAs
so that it performs better; we'll use our reinforcement learning for
that.
- Second, we'll use the language system to allow the user to
specify new plans. This will be limited by the size of the
net, but that should be pretty easy.
- The hard thing is to learn the plans in a more biologically
plausible way.
- We'll have overlapping CAs so we'll get hierarchical effects.
- We will have a separate area to activate the plans, test them,
then if they work, learn them.
- This is pretty speculative.
- Note this is really just learning CAs for processing.
