mebedavis

# PhysicsFixed.py; revert to Aquarium2.py

# Untested, but in place:
# Advance to training the cybernetics connecting sensors to neurons to motors;
# Try testing for motor activations that tend to reduce angle and distance to target.
# Reassign network connections when they are maintained near 0 weight after multiple
# opportunities to update.

# Agents will reproduce when they have accumulated sufficiently many primitives 
# to constitute the new structure; new structure will be modified by random adjustment of all relevant

mebedavis

# revert to atleastnotdestructive

# Current training method at least does not destroy the bespoke network,
# but with random initial map, connections do not reassign productively
# or else acquire sufficient weight to become fixed or something.

# Advance to training the cybernetics connecting sensors to neurons to motors;
# Try testing for motor activations that tend to reduce angle and distance to target.
# Reassign network connections when they are maintained near 0 weight after multiple
# opportunities to update.

mebedavis

# TrainingTake3.py  Based on cybernetics.py Training method appears to work but is counterproductive 
# owing possibly to distance changes lagging too far behind motor inputs.
# Advance to TrainingTake3.py to try predicting negative deltas on angle to 
# target instead.

# Revert to TrainingTake2.py; construct network updates and valid tests of success.
# Diagnose whether or not network updates including reassignment are effective.

# Advance to training the cybernetics connecting sensors to neurons to motors;
# Try testing for motor activations that tend to reduce angle and distance to target.

mebedavis

# RudimentaryTraining.py.  Revert to TrainingMethod.py

# TrainingMethod resulted in ostensibly correct connection weight adjustments
# to penalize counterproductive connection assignments and augment good ones
# based on comparative target acquisition compared to baselines for optimal (21 acquired),
# severly wall-eyed (14), and trained by the method from random assignment (16).
# Deficit from optimal network map hypothetically explained by the mere fact that some
# connections now return zero, being suboptimally connected, and therefore debit total 
# motor thrust.

mebedavis

# Cybernetics.py; revert to BACKUPSentinelDrones.py

# Advance to rebuilding the cybernetics connecting sensors to neurons to motors.

# Vision logic is now totally redundant.  I'm iterating over the flat list of primitives to populate a dictionary
# that's now sorted in geometrical order, when I could just be sorting the flast list OR the
# dictionary.  It works, so I'm not fucking with it right now, but I should find out which
# is faster and eliminate the other method at some point.

# Rebuild connectome, imposing optimal connections between sensors and neurons, neurons with 

mebedavis

# SentinelDrones.py

# Advance to rebuilding the cybernetics connecting sensors to neurons to motors, but first
# refine the test environment by creating the target class and populating the gamespace
# with instances which move to demonstrate the agent consistently interacts with targets.
# Create a collision method to indicate when agents have approached targets within
# some reasonable interaction distance, perhaps d < 5.

# Then, rebuild connectome, imposing optimal connections between sensors and neurons, neurons with 
# sensors, motors, and other nerurons, and motors to neurons.  This will provide the basis for an 

mebedavis

# Revert to FirstRotation_Updated.py.  Advance to Aquarium0.5.py with magnified graphics.

# Demonstrated complete rotation on appropriate pitch from vertically opposed motors.

# Next: magnify graphics and depth-scale.

import math
import numpy as np
#import random
import pygame

mebedavis

# Full rotation demonstrated.
# Re-implement sensors and networks.
# Revert to physics4.

import math
import numpy as np
#import random
import pygame
import time
from scipy.spatial.transform import Rotation as R

mebedavis

# Clean Slate rebuild of the physics core.
# Exists Physics3 backup to revert to.
# Intent is to strip the physics core out of this,
# preserve useful graphics functions, and replace the physics core,
# fully leveraging the rotation math in scipy in hope of solving 
# problems with reference frame inversion and other problems with any
# rotation 2pi/4 causing singularities in the quaternion method.

import math
import numpy as np

mebedavis

# Revert to FirstRotation_Updated.py.  Advance to Aquarium0.5.py with magnified graphics.

# Demonstrated complete rotation on appropriate pitch from vertically opposed motors.

# Next: magnify graphics and depth-scale.

import math
import numpy as np
#import random
import pygame