Chess_perception
---------------------

TODO
-------------
l2 norm penalties for filtering ground
l2 norm penalties while clustering
look within a square for the chess piece
Add checks to find if we should really look for a fallen piece or not (Game turn)

Calibration
-------------



Compile and Run
-------------
cd Debug;
make;
./chess_perception

ACH Channels
-------------

* Summary *

Name: chess_perception
1. Somatic__Token->iattr
boardstate[64 + NUMBER_OF_FALLEN+1]

The first 64 elements are encoded as per the craft board.  After that,
a value of 1 indicates that there is a fallen piece.

2. Somatic__Token->fattr
double[(64 + NUMBER_OF_FALLEN+1)*4] in camera ref

Contains offsets of each piece The offset is calculated from the
original initialization of the chessboard.  The 4th element specifies
orientation.  For the first 64 elements, the offset is relative to the
square.  After the first 64 elements, the offset is relative to the
center of the board.

* Details *

There are two vectors (one integer - iattr and one floating - fattr)
Since you cannot encode offsets and orientation in iattr, they are in fattr

For the 64 squares on the board, this is when pieces are nicely placed on the board.
Then each iattr[i] will say if there is a piece in the square. 
If there is a piece it will say which piece according to:
Pawn as 1, Knight 2, Bishop 3, Rook 4, Queen 5 and King 6.

For each piece, 
fattr[i*4+0] = x offset
fattr[i*4+1] = y offset
fattr[i*4+2] = z offset from chessboard
fattr[i*4+3] = orientation around Z axis

It compares the board state with Crafty
--------------------------------------------
If there is a piece missing, it tries to find fallen pieces. 
If there is one fallen piece iattr[64] will be 1 and then iattr[65] = 0. 
for two fallen pieces, iattr[64] = iattr[65] = 1 and iattr[66] = 0.

Corresponding to each piece, fattr will have offsets and orientation.
BUT, the offset will be reported from the center of the board.