FlyBrain · Drone Lab

NEURAL READOUTForest · seed 42

00:00.0SIMULATION TIME

Ready for takeoffThe neural model is loading. You can try the vision baseline while it loads.

ALTITUDE3.0 m

AIRSPEED0.0 m/s

DISTANCE0 m

THROTTLE45 %

Fixed seed · target altitude 3 m

Vision supplies avoidance decisions. An attitude and altitude controller keeps the simulated quadcopter stable.

FLIGHT SETTINGS Controller Environment

Forest seed

Added command delay50 ms

0 ms200 ms

COMPOUND-EYE SAMPLES800 / eye

LEFT

RIGHT

Perspective samples; approximate eye geometry.

NEURAL ACTIVITYL / R

Silence circuit LPLC2 LC4

0 active neurons— ms / step

LC4 and LPLC2 receive engineered looming features. Bars show simulated rates, including directly stimulated cells. LC11, descending and motor cells are observation probes.

About this experiment

MaleCNS v1.0 supplies connections across the brain, optic lobes and ventral nerve cord. A bounded rate model propagates activity through that graph. A hand-tuned adapter converts bilateral LPLC2/LC4 activity to steering and speed; it has not been trained.

The nerve cord's circuits are simulated; descending and motor activity is monitored. Steering still reads LPLC2/LC4. Soma-side labels (nerve-entry side for sensory cells without a soma label) are a coarse routing assumption, not measured visual receptive fields. The stimulus benches hold the drone still.

The delay slider adds simulation-time command latency. Measured compute time and simulation speed are shown separately. Setting changes restart the trial for comparison.

Methods and sources ↗

A fly's view of flight.