ARCADE / NEURAL PILOT: DUEL

Neural Pilot: Duel

on-device AIshooter1P vs AIkeyboard + touchtrained adaptive AI

WASD / arrows · touch stick on mobile

Neural feed — standby

Duel

HUMAN 00 NEURAL-V4

Policy readout — ms

You and a PPO-trained neural network fly identical missions side by side — same seed, same waves, no excuses. The policy readout below the stage shows every decision it makes, in real time, on your device. It trained for 14 million runs; you get unlimited retries. Switch to ADAPTIVE mode for a second trained opponent — an LSTM policy that watches your live position and recent moves and reacts to them, not just the fixed wave schedule.

How to play

Rules

Same seed, same enemy waves as the AI — outscore it over 90 seconds. You score by destroying enemies; getting hit costs a respawn, so survival is the real metric.

Controls

WASD / arrow keys to fly, or the on-screen stick on touch devices.

Tips

The Policy readout panel shows the AI's decision every frame — read it to spot safe lanes. Staying ahead of the waves beats chasing score. The HUD mode toggle switches opponents: PPO flies a fixed policy, ADAPTIVE reads your live movement and reacts to it.

Behind the game

Neural Pilot: Duel makes a score chase feel fair in a very particular way: you and the AI are given the same sky. The enemy waves arrive from the same seed in the same order, so neither pilot receives a lucky pattern or an excuse. You can see the difference in decisions instead. A tempting cluster of enemies may offer points, but a hit costs a respawn and lets the next wave advance without you. The 90-second mission rewards a pilot who keeps a route through the pressure rather than one who simply flies at every target.

Start by treating the center of the screen as a resource, not a home. Drift only when a wave gives you a reason, leave yourself room to turn, and avoid entering a corner just to collect one more elimination. The score grows when enemies fall, yet survival is what lets that score compound across the whole mission. The readout beneath the stage is useful study material: it shows the neural pilot's current choice as it is made. Compare a safe lane you spotted with the lane it chooses. On a replay, notice whether it begins moving before the danger looks urgent. That anticipation is often more valuable than fast reflexes.

The standard PPO opponent learned a policy for these repeated aerial situations through training, balancing movement and attack across many runs. It does not know a human-perfect script; it reacts to the state of its own flight and the wave around it. ADAPTIVE mode changes the conversation. Its recurrent policy also watches your recent position and movement, so it can respond to how you are flying instead of only to the fixed schedule. That is why a favorite evasive loop can stop feeling safe after you repeat it. The glowing duel is less about proving a machine wrong than about reading the same storm more clearly. You have unlimited restarts, a visible opponent, and the chance to turn every crash into one sharper decision on the next identical mission.

MLP 48-64-64-9 / ADAPTIVE: LSTM 106-96-9 · infer 0.4 ms / ADAPTIVE: 0.4 ms

More games

NEW
Gate Weave

Slide the colour carts to their matching gates — read the order and the whole board unravels.

NEW
Crate Aperture

Slide the crates, open the exit. Solve each board in fewer moves than par to chain a multiplier — and keep your one undo for a bonus.

Triad Grove

Swap blossoms, grow a cascade, and make the Bloom cell burst.