LABS FLIGHT EXPERIMENT / 04
< QUEST TERMINAL
NEW SITE
EXPERIMENT 04 / NEURAL FLIGHT

FLY
LANDER

A tiny brain. A lunar landing.
Watch the fly, take the controls, or let it learn.

LANDING TARGET

Memory copy unavailable

Stationary terrain pad · original mission

ATMOSPHERE

Calm · no wind

Sideways gusts rise, fade and reverse.
Applies when you start a flight.

PREPARING FLIGHT SYSTEMS…
Flight logs stay in this browser.
+ MARE TRANQUILLITATIS 00° 40′ N   23° 28′ E
FLIGHT MANUAL / 01

FLIGHT MANUAL

This experiment uses a simplified controller built from measured fruit-fly connections: 272 neurons, not a full simulated brain. It receives numerical terrain and motion sensors rather than camera images. The illuminated pad shifts to a new site each flight. Enable Moving platform on the flight deck for flat ground and full-width travel; speed zero holds the platform still. Moving missions use a saved fly without live learning. Enable Short-term memory + velocity to use the separately trained copy: it sees the last 0.20 seconds of relative position and craft velocity, plus platform velocity. Turn it off to use the current fly without observation memory. Clear the ridges, slow your descent and stay upright. Only the marked pad is safe. Random wind gusts add a sideways push, adjustable on the flight deck. The current saved fly handles both calm air and gusts. Changing the weather does not switch the brain. The WIND PUSH readout shows direction and acceleration in simulation units/s²; zero strength is calm. Wind affects stationary and moving missions. The fly senses its effect through craft velocity; Watch fly keeps its saved brain frozen. One tank supplies every engine; empty engines stop firing while gravity continues.

↑ or W
Main engine
← / A   → / D
Left / right attitude engines
Space or P
Pause / resume
R
Restart flight
F / Esc
Fullscreen / exit

Safe touchdown: horizontal speed relative to the platform ≤ 0.20, vertical speed ≤ 0.25, tilt ≤ 11.5°, spin ≤ 0.30 rad/s, and fit within the lit pad. The altitude readout is clearance above terrain. Distances use simulation units.

Watch fly uses the saved neural policy. Train fly lets a separate copy learn from each flight; its updated connections are saved in this browser after every completed attempt. The fly senses the target and remaining fuel. Full-terrain policies also see the highest ground in 32 sections across the map and the tallest ridge between the lander and pad. Gameplay runs entirely on this device. Completed flights and learned weights stay in this browser; use Export flights on the flight deck to download them. Other devices have separate logs. GPU batch training remains a separate desktop workflow.

Data attribution and software licenses