Genotypes should express a range of behavior wide enough, and complex enough, to allow for interesting solutions to arise to a given 'problem' posed by fitness criteria. Currently genotypes consist of a processor tree that computes a new pnode (short for plant-node) position based on the parent-pnode position and the nutrient position. This constrains pnodes to always produce a child when fed. This is rather rigid. (Certainly interesting behavior may nevertheless present itself. That would be the focus of tweaking the fitness criteria and evolutionary algorithm.)
The types of behaviors possible should be wide ranging. Here are some:
'memory' of collision events
sending signals to parent, child
deciding what type of child node to spawn.
deciding how to configure (if possible) the child node
It is unclear if standard GP techniques will allow for this range; indeed the 'memory' behavior suggests the need for retaining state.
Genotypes should express a range of behavior wide enough, and complex enough, to allow for interesting solutions to arise to a given 'problem' posed by fitness criteria. Currently genotypes consist of a processor tree that computes a new pnode (short for plant-node) position based on the parent-pnode position and the nutrient position. This constrains pnodes to always produce a child when fed. This is rather rigid. (Certainly interesting behavior may nevertheless present itself. That would be the focus of tweaking the fitness criteria and evolutionary algorithm.) The types of behaviors possible should be wide ranging. Here are some: