Self-playing chess board

What it does.

Play a computer at the strength you pick, a friend online, or watch it replay a famous game. Every move happens on a real board with real pieces.

Sees your moveA magnet sensor under each of the 64 squares reads where every piece is.
AnswersA chess engine on your laptop picks the reply over Wi-Fi. Online play goes through a Lichess account.
Moves its own piecesA carriage under the board drags pieces along the square edges, so knights and crowded boards work.
Handles captures and castlingCaptured pieces slide off into lanes on each side. Castling and en passant are automatic; promotion asks you to swap the piece.
Shows the gameA light under each square marks the last move, check and whose turn it is.

45 mmsquares, 8 x 8

512 x 492 mmfootprint

47.5 mmthick at the deepest point

3.5 saverage move (simulated)

64Hall sensors, 64 LEDs

USB-C20 V power, or a 12-24 V adapter

4circuit board designs

~3.5 hagent time (+37 min parallel); engineer review to come

~$634first unit estimate, incl. one-off fab fees (~$417 parts)

Numbers come from the design and simulation. Nothing here has been measured on hardware yet.

The trade-offs.

Each part of this board fights another one. These are the five fights that shaped it, and the call we made on each.

Magnet pull vs board thickness vs friction

A thinner top pulls harder, but sags and leaves no room for the lights. A stronger magnet also presses the piece down harder, so friction rises with it.

Call: a 7.6 mm gap, two small electromagnets with thin cores so pieces let go cleanly, and felt that has to test slippery enough. If it doesn't, the pieces get PTFE dots instead.

The magnet vs the sensors

A switched-on magnet looks like a piece to the sensor above it.

Call: electromagnets, which have no field when off. The board only reads its sensors with the coils off and settled, and never powers them while it waits for you.

Speed vs noise vs dropped pieces

The piece trails the magnet by a few millimetres. Move too fast and it slips off. Faster also means louder.

Call: gentle acceleration, at least 3x under the slip point. The carriage overshoots slightly, then wiggles to centre the piece. A 244-move simulated game set averages 3.5 s a move.

Rail stiffness vs total height

A stiff aluminium beam would push the board past 50 mm thick.

Call: the beam is a bare steel rail hung low, with a one-belt layout and flat motors tucked under a rear ledge. 2.5 mm to spare.

Square size vs piece size vs gaps

A piece sliding along a square edge has to squeeze between its neighbours.

Call: 45 mm squares and a slightly smaller tournament-style set. The planner checks every path against the real piece sizes and moves a blocker aside if it has to. The capture lanes make it 512 mm wide instead of about 450.

Where it stands.

Designed. First build in progress.

Design: CAD for every part, four board designs with fab files, firmware that compiles, a parts list checked against live stock, and a written log of the design mistakes we caught and fixed.
Checks done: Board electrical and design-rule checks pass (not a test of function). No part collisions across the full travel. Firmware logic tested in simulation over five full games.
Not yet known: Real felt friction, how well pieces centre, how much the top sags, noise and Wi-Fi range. These need the real thing.
Next: Nothing is built yet. First, four small bench tests (magnet pull, felt friction, centring, top sag) before any boards are ordered. Then the first unit. We'll post the results, good or bad.

A chess opponent that plays you on a real board.