Neural Pursuit

Behind the game

Neural Pursuit is built around an uncomfortable advantage: you are faster than any one chaser, but there are three of them. Running directly away feels sensible for a few seconds, then the pack begins to occupy the exits you planned to use. A sixty-second survival clock makes each turn meaningful. You are not trying to defeat the hunters or clear a level; you are trying to keep enough open space that the next turn still exists. The best survival time is a personal record, but every restart is also a new lesson in how a crowd can become a trap.

Resist the urge to spend your speed immediately. A long straight sprint gives the three chasers time to arrange themselves around your destination. Instead, watch the gaps between them. Cut across a gap before it closes, make a sharp direction change near the edge of their reach, or double back through the space a leading chaser just vacated. These moves are not magic escapes: a turn that is too late simply puts you closer to another hunter. The satisfying runs alternate calm repositioning with one deliberate burst, preserving room rather than treating every second as an emergency.

The pack is driven by a policy learned through self-play, with the same policy controlling each AI seat from that seat's own viewpoint. No handwritten pincer rule tells one chaser to flank while another follows. During training, behaviors that made contact with a runner became useful, and coordinated-looking convergence emerged from that pressure. Sometimes the group trails; sometimes two bodies close the obvious lane while the third arrives from the side. That variation is why memorizing a route does not solve the game. Read the shape of the pack each moment, not just the nearest dot. The arena is deliberately spare so the drama is legible: three bright pursuers, one quick runner, and a shrinking set of honest choices. Last one more second, then ask where the next escape was hiding.