Neon Lightcycle Duel

Two agents carve neon trails across a grid. Territory is Voronoi-partitioned in real time โ€” the one who claims more open space wins.

Loading interactive simulation...

The territory overlay is a race, not a shape 🖖

The coloured split looks like a Voronoi diagram, and it is one โ€” but the distance it measures is moves through open cells, not the straight line. The moment a trail is down, the two stop resembling each other. Take a 22ร—22 grid, run a trail along the middle column and leave one gap in the bottom row: the cell two squares across is 44 moves away โ€” twenty-one down, two over, twenty-one back up. Twenty-two times further than it looks. A wall you can see over is still a wall you must drive around, which is why the overlay redraws the instant either cycle turns, and why Voronoi sits in the dropdown as a strategy rather than as decoration.

A race for space, not a fight 🖖

In a lightcycle duel nobody attacks anyone โ€” you lose only by driving into a wall, a glowing trail, or a dead end. The real objective is to claim and keep enough open room to keep moving. The practical lesson: hug the edges and fill space in tidy sweeps, rather than slicing across the open middle, which leaves you with fewer escape routes than your rival.

Perfect play is provably intractable 🖖

You might assume a grid this small is easily "solved," but in general it is not. In 2012 Tillmann Miltzow proved that deciding the winner of Tron on arbitrary graphs is PSPACE-hard โ€” a class widely believed to be even tougher than the NP-complete problems behind most puzzles. That is exactly why this tool relies on heuristics like greedy corridors and 2-ply minimax instead of computing the truly optimal move.

References (1)

Example problems

  • Greedy vs greedy - Balanced duel: medium arena, medium AI caution
  • Tight arena - Tight arena forces early corridor decisions and high collision risk
  • Minimax duel - High speed + aggressive AI creates chaotic short rounds
  • Play (weak bot) - Large arena with timid AI gives player more trap-building control