f.Fly Habitata living connectome

An open experiment in embodied intelligence № 001

A small world.
A whole brain.

A measured fruit-fly connectome inside a simulated habitat. Observe its activity. Shape what it learns. Follow what survives the next generation.

Enter the habitat Explore the circuit lab Checking WebGPU…
139,255
measured neurons, one pixel each
15.09 M
signed synaptic connections
78
actuators in the jointed body
2
reinforcement learners, inherited at death
Habitat

The house

A 3.5 × 2.5 cm home with a fruit kitchen, water garden, dark refuge, pollen garden and nursery. Real MuJoCo contacts, real walls.

Brain

Every neuron, live

A 3D map at measured positions, a pixel matrix of the whole brain, a spike strip of named descending neurons, and the population-level connectome matrix summed from the measured wiring.

Learning

Two real learners

Steering is a policy over left−right brain activity trained by REINFORCE. Goal choice is Q-learning over needs and population rates. Freeze, reward, disconnect cells, scale synapses, and watch generations.

Honesty

Measured versus authored

Wiring, weights, positions and body are data. Senses, life rules and rewards are models, and the page says so on every panel that uses them.