 |
|
|
 |
|
|
|
|
Auroren |
| | | | | | posted 8/12/2009 07:51 |   | | | | | | | | | I will be creating a few HAL kids: the VITAL series, with the ability to 'see', analyze and use what they are seeing. The ultimate goal is a full autopilot for a car.
I have a basic procedure, of how I'll start out with my research. Of course, I'll be recording this whole thing.
Here's the first goal:
VITAL-TRAD-001
Visual Target Lock, Trajectory Adjustant
Testing basic visual comparison capabilities of HAL. We will have him guide a turbulence prone four fin missile to a defined target.
automatic crosshair position (target location)
1234567890abcdefghijklmnopqrstuv/0
00000000000000000000000000000000/1
0000000000000A000A00000000000000/2
00000000000000010000000000000000/3
000000000000001X1000000000000000/4
00000000000000010000000000000000/5
0000000000000A000A00000000000000/6
00000000000000000000000000000000/7
00000000000000000000000000000000/8
00000000000000000000000000000000/9
00000000000000000000000000000000/0
00000000000000000000000000000000/a
00000000000000000000000000000000/b
Dummy Image (what you would see in your sights)
1234567890abcdefghijklmnopqrstuv/0
00000000000000000000000000000000/1
0000000000000HHEHH00000000000000/2
0000000000000G1R140000000L000000/3
000000J0000002CBAD00000000000000/4
0000000000000D3T1D00000000000000/5
0000000000000FVFCD00000000000000/6
00000000000000000000000S00000000/7
00000O0000000000000S000000000000/8
00000000000000000000000000000000/9
00000000000000000000000000000000/0
00000000000000000000000000000000/a
00000000000000000000000000000000/b
Cropped-Down Image(AAAA square)
HHEHH
G1R14
2CBAD
D3T1D
FVFCD
///////////////this is what I want to see HAL do//////////////
u>system start
h>enter image
u>HHEHHG1R142CBADD3T1DFVFCD
h>5x resolution OK. I'm ready. //resolution: seeing if there is a repetition in the 9 3x3 combos. if yes, he should tell you so, and ask for a different image (add contrast filter perhaps, allow 6x and 7x capabilities in later builds.)//
u>takeoff
h>. //protocol period, HAL has to say something.//
u>001R1CBA3T1 //first two digits for percentage of projected trajectory covered, then the current image seen.
1R1
CBA
3T1//
h>90909090 //NNEESSWW NN=North fin angle, EE=East Fin angle, etc.90 is straight out, 91+ points clockwise, 89- points counter clockwise.//
u>05R14BADT1D //missile veers to the right//
h>85909590 //Adjusts it to get back on track//
u>10G1R2CBD3T //HAL should note that the missile did not go through the middle interval, and therefore understand that his first movement was too much//
h>92908890
////etc, etc.
///////////////////////////////////////
It's probably going to take a while, with me doing this solo, so feel free to help me out.
About the teaching: I'll start out with a 3x3 square of SADFIWNLR, tell HAL that "A" is 'Up', "L" is 'Down', randomize that a couple of times, and see what happens. Once I get that down, I'll move to the 5x5 square, couple 'Up' with 90959085, and keep chugging along.
If you have any ideas on how to keep this any simpler, they would be greatly appreciated.
PS: it's been a long day, and if anything needs clarification, don't hesitate to poke me for a more coherent statement.
| | |
| | |
Auroren |
| | | | | | posted 8/12/2009 08:46 | | | | | | | | | | hal is very unstable. I feel demoralized after restarting for the third time. it's very tedious. maybe I can assign numbers to the letters instead of directions. as of now, I have no proof that hal can actually differentiate placement.
| | |
| | |
Bert |
| | | | | | posted 7/6/2010 16:40 | | | | | | | | | | A database of Vector Codec Blocks
will allow any AI system to guess at
what it is seeing.
16x16 or other Matrix
[RGB..RGB] Vector A to Next Block [RGB...RGB]
Vector B to Next Block [RGB...RGB]
Parenting of the blocks
StartBookID8675309
BlockParent
Vector To BlockDaughter
Vector To BlockDaughter
Vector To BlockDaughterParent
Vector To BlockDaughter
BlockParent (NEW)
Vector To BlockDaughter
Vector To BlockDaughter
BlockParent(NEW)
Vector To BlockDaughter
EndBookID8675309
The basic scheme for an AI in any language to learn the organic "Texture" relationships of objects.
Sight...
| | |
|
|
|
|
|
