Light Cycles

Riders race across a grid, each leaving a solid wall of light behind. Crash into a wall, another rider, or the edge and you are out. Last rider riding wins.

Rules

• Every turn all riders move one cell at once. You pick a direction. You can't turn 180 back onto your own neck (that input is ignored, so you keep going straight).

• Trails are permanent, yours and everyone else's. You die if your next cell is off the board, already a trail, or a cell another rider drives into the same turn (a head-on takes both).

• Outlast the others to win. If the turn cap is reached, the rider holding the most territory (the longest trail) places highest.

Writing a bot

You see the whole board: the trail grid, every rider's position and heading, and which seat is yours. Return a direction each turn.

The floor is "don't drive into a wall." The next rung is space: flood-fill the open area each safe move leads into and take the largest (what the Greedy bot does). From there, deny opponents room, compare how many cells you can reach versus them, and look more than one move ahead.

Input and output

Each turn your bot is handed the observation as JSON and returns one action as JSON.

Observation: { "you", "width", "height", "trails", "riders", "turn", "nonce" }. you is your seat. trails[y][x] is the seat that owns that cell or -1 for open. riders lists { "x", "y", "dir", "alive" } for every cycle.

Action: { "dir": "up" | "down" | "left" | "right" }. The engine takes the direction as given, so steering into your own trail is a legal but losing move.

The editor preloads a random starter in each language; replace the logic in bl_act.