One tiny pilot. Three very dirty rooms.
Meet the world's most overqualified vacuum operator: a fruit fly with a little joystick and a whole apartment to clean.
Watch it work through the living room, kitchen and bedroom, collecting scattered debris and tackling stubborn puddles. Wheels smear liquid into tracks; the side brush sweeps it inward; suction gradually removes it. Every clean patch comes from physical contact.
Follow the little pilot
- Three camera views: orbit the apartment, follow the Roomba, or ride beside the fly in the low-mounted Pilot GoPro.
- Live neural activity: see the simulated fly's activity as a moving texture and motor-spike telemetry.
- Actual travel history: toggle the mint route line to see where the vacuum has been.
- Mess with the experiment: start a fresh mess, retrain the motor circuit, zero its learned weights, or silence its motor neurons.
Controls
Use right-drag to orbit and scroll to zoom in Overview and Follow pilot. Switch cameras with the bottom buttons. Route ON/OFF shows or hides the trail, and Hide UI gives you an unobstructed view.
A real neural simulation
The experiment uses a simulated fruit-fly CNS containing 164,587 neurons. Local depth, floor sensing and odometry build its memory of the apartment. An engineered circuit of native simulated neurons propagates navigation signals, and trained motor synapses turn measured neuron spikes into wheel movement.
This is an engineered intervention in a fly-neuron simulation, not a claim that an intact biological fly has learned housekeeping. The cleaning is measured: a recorded full-house run finished 72/72 mess sites, 24/24 in each room. Silencing the actuator neurons produced zero wheel output.
Comments · 0
Open discussion, no rating attached. Ask questions, swap notes, follow threads.