DevelopmentNotes.txt

Controller and ArticulatedBody are tightly linked, and their
creation is intertwined.  To support this coupling, creation is
handled by the EvalutionFactory.

An evaluation bundles together all the components needed to
evaluate a given genome.  (For now, we assume that each
genome is to be independently evaluated, so co-evolution
is not supported.)  Thus the Evaluation consists of:

- ArticulatedBody
- Controller
- FitnessFunction, which is part of the PopulationFitness

The ArticulatedBody requires a fixed-sized set of "inputs"
that are used to calculate the torques applied for all the
controlled degrees of freedom of the character.  There may
be controlled degrees of freedom that do not directly map
to inputs, but it's better if this set is small-- calculation
of the control should take place in the controller.

However, ArticulatedBody control inputs are in the range (0,1),
and so must be mapped to appropriate torques, this is one
of the key responsibilities of the ArticulatedBody class.

In addition, the simulated body generally has a set of sensors
capable of quantifying the state of the body.  These sensors
are scaled to (0,1) and provided as the ArticulatedBody's
outputs.

The Controller is a consumer of the ArticulatedBody's outputs,
and a provider for its inputs.  Controllers may use one or
more evolved neural networks.  In general, all of a Controller's
neural networks will be built based on the single genome used
to create the controller.  If co-evolution is needed, a different
Evaluation subclass would be needed, and it would pass multiple
genomes to the controller (or one to build the controller, and
another to build the articulated body).


Generally, the data flow is:

ABout -> controlIn -> Controller.evaluate() -> controlOut -> ABin


Creation.

Generally, the main program creates an Evolver (e.g., NeatEvolver).

Usage:
  NeatEvolver->go()
    NeatEvolver->singleRun()
      for every gen
        evaluatePopulation()
          for every genome
            new Evaluation( )
            evaluation()
        myPop->epoch()

  runIndividualGenome( FitnessFuncPtr, PhysicsSimulatorPtr, argUseHarness )

Controller Factories
-- 
 - use ab to get controller IO counts




------------------------------------------------
TODO-
- What should Crossover inputs/outputs do for symmetry > 2?
- ArticulatedBody should be owned by evaluation, not sim
-