Tracking pieces for robot chess

[theycallmevirgo]

Our school got a robot arm and I was thinking of teaching it to play chess. AI chess is a solved problem, so is moving the arm around. However, I was hoping to at least try to use regular pieces. It would be easy enough to do it with markers, but physical pieces would need to be identified and tracked. So far I thought of

-Computer vision (unreliable)

-RFID (afaik insufficiently accurate at that scale)

-Resistors with physical connections (how would I calculate the piece id and position from resistance?)

Many thanks in advance for any hints

Joe

 

[berkeman]

Sounds like fun.

I think a very near-field RFID system could work pretty well. You'd want to use very low power Tx/Rx antenna structures under each of the squares.

You could back that up with just a keyboard entry that the person playing uses to tell the computer about the move it has just made.

 

[berkeman]

Alternately, you could use a technique that my first computer chess board used (from Radio Shack about 30 years ago). It had a switch under the plastic fabric at each square, and to let the computer know your move, you pressed the piece down on it's existing square to signal the start of the move, and then pushed it down on its new square to signal the end of the move. The board made a "beep" sound with each button push to give tactile/audible feedback.

EDIT -- I found this picture of a similar-looking chess game as an illustration. But I see now that they are advertising some sort of "sensor" capability to track the pieces. It may just be the same pushbutton technique with the addition of magnets in the pieces to actuate reed relays under the squares, or it may be more comprehensive. It may be worth it for you to look more into this game to see what they are calling the "sensor" capability/feature:

http://www.spacious-mind.com/html/sensor.html

 

[trurle]

Our school got a robot arm and I was thinking of teaching it to play chess. AI chess is a solved problem, so is moving the arm around. However, I was hoping to at least try to use regular pieces. It would be easy enough to do it with markers, but physical pieces would need to be identified and tracked. So far I thought of

-Computer vision (unreliable)

-RFID (afaik insufficiently accurate at that scale)

-Resistors with physical connections (how would I calculate the piece id and position from resistance?)

Many thanks in advance for any hints

Joe

Machine vision may be easier than you think - if you attach a small sticker with QR code to each chess piece.

On the other hand, most reliable way may be still an RFID - to hide a passive, trimmable LC resonator inside each piece. You need to place VCO and induction coil in robot arm, and read out both piece ID (from response frequency) and piece distance (from response magnitude).

I would advice against "physical" (i.e. DC connected) resistors - these are notoriously unreliable. Especially whenever a greasy hands are involved.

 

[256bits]

If the arm is moving the pieces it could have a memory of where each piece is on the board.
You would tell it which piece to pickup and where to move it. The new layout is then back in memory.
Your opponent would then do the same.

If you are playing against the computer it will need to have this memory anyways to plan its own moves.
No need for an active method of determining the layout for each play.

The one problem to solve then is how the arm is to grip each piece, as king, queen knight, bishop, rook, pawn all have different "heads" and "bodies".

 

[trurle]

The one problem to solve then is how the arm is to grip each piece, as king, queen knight, bishop, rook, pawn all have different "heads" and "bodies".

The solution is called "universal gripper".

 

[256bits]

As long as we don't knock over adjacent pieces.
I suppose a suction on the top of the piece would do it - another engineering system to design and test.