/generally_intelligent_robot

generally intelligent robot

Why

The goal is to build a robot with common sense & the ability to generalize to any 3 dimensional goal with a time estimate.

How

The following are generated images using Python 3 & C & various libraries like matplotlib, networkx, ... The VCC robotic arm is coded using https://robosuite.ai

Visualizations of general intelligence models

Visualizations of HTM & HTW [1]

HTM node x=1,y=2 level=1 time_step=12.png markov graph

t=12

HTM node x=1,y=2 level=1 time_step=50.png markov graph

t=50

HTM node x=1,y=2 level=1 time_step=50.png temporal groups dendrogram

t=50

Visualizations of GVM [2] [3]

16 edge filters visualization

filter_scale=4 0_num_oriented_edge_features=16_display= 0c1,0c13,0c13 =

filter_scale=4 0_num_oriented_edge_features=16_display= 0c2,0c13,0c13 =

filter_scale=4 0_num_oriented_edge_features=16_display= 0c3,0c13,0c13 =

...

filter_scale=4 0_num_oriented_edge_features=16_display= 0c15,0c13,0c13 =

gaussian filter

gaussian_filter_scale=4 0_num_oriented_edge_features=16_weights=False_gaussian_size=13 0

gabor filter radian angle = 0

radian_angle=0 00_gabor_filter_scale=4 0_num_oriented_edge_features=16_weights=False_gabor_size=21

radian_angle=0 00_gabor_filter_scale=4 0_num_oriented_edge_features=16_weights=True_gabor_size=21

gabor filter radian angle = 0.39, incrementing 16 times to 5.89

radian_angle=0 39_gabor_filter_scale=4 0_num_oriented_edge_features=16_weights=False_gabor_size=21

radian_angle=0 39_gabor_filter_scale=4 0_num_oriented_edge_features=16_weights=True_gabor_size=21

For handwritten digit "0"

cross_orient_max

label=0_num_oriented_edge_features=16

16 competitor_maxs

label=0_competitor_maxs_i=0_num_oriented_edge_features=16

16 gabor_filtered

label=0_gabor_filtered_i=0_num_oriented_edge_features=16

16 localized[i] = competitor_maxs <= gabor_filter_at_i

label=0_localized 0 _num_oriented_edge_features=16

localized[cross_orient_max > gabor_filtered] = 0

label=0_i=0_num_oriented_edge_features=16

updated background threshold = 0.001

label=0_i=0_num_oriented_edge_features=16

label=0_i=1_num_oriented_edge_features=16

14 more images representing bottom up messages for handwritten digit "0"

label=0

label=0_num_landmark_features=88_num_close_pairs=395_perturb_factor=2 0_max_lateral_connection_pixel_length=15_tolerance=4.png

label=0_num_landmark_features=88_num_close_pairs=395_perturb_factor=2 0_max_lateral_connection_pixel_length=15_tolerance=4

how to find

optimal perturb factor? optimal max_lateral_connection_pixel_length? optimal tolerance?

cross channel pooling visualization

channel_offset = 1 factor = 1

label=0_i=0_num_oriented_edge_features=16

channel_offset = 2 factor = 2

label=0_i=0_num_oriented_edge_features=16

Visualizations of VCC 2D

Virtual environment with robotic arm & multiple camera point of view data to implement visual cognitive computer (VCC) architecture

Screen Shot 2021-03-09 at 10 37 20 AM

Visualizations of VCC 3D with recursive calls

Visualizations of VCC 3D with recursive calls + grounded language understanding

Sources

  1. how_the_brain_might_work.pdf
  2. GSM_summary.pdf
  3. GVM_details.pdf