Ariel Zeleznikow-Johnston

Heading.

So, what did E.ON actually do?

E.ON took the pre-existing components we just described, the Shuital brain model and the NeuromechFly V2 body, and connected them together into a closed loop.

Sensory events in a virtual world feed into the brain model, and selected outputs from the brain model direct the virtual body.

The loop has four.

Steps.

First, something happens in the virtual environment, the fly's leg contacts a sugar source, or dust accumulates on its antennae, and these events activate specific sensory neurons in the brain model.

Second, the brain model runs for a 15 millisecond time step, propagating activity through the connectome's roughly 140,000 simplified digital neurons.

Third, Eon reads out the activity of a small, handpicked set of descending neurons and translates it into high-level commands, turn left, walk forward, groom, feed, that are passed to pre-trained motor controllers in the body model.

Fourth, the body moves, changing what the fly senses, and the loop repeats.

The result is the video that went viral.

But the behaviors on screen are less impressive than they appear, because the brain model is doing far less of the work than a viewer would naturally assume.

Take the walking.

The brain model does not orchestrate the fly's legs.

It doesn't compute the gait cycle, coordinate the six limbs, or position the joints.

It activates a few descending neurons, ODN1 for forward velocity, DNA01 or DNA02 for steering, and hands that signal off to a locomotion controller within NeuromechFly that already knows how to walk.

The brain is issuing something like a go-forward or turn-left instruction.

The body model handles everything else.

In a biological fly, the detailed work of translating such commands into coordinated leg movements is performed by roughly 15,000 neurons in the ventral nerve cord, the fly's equivalent of a spinal cord, none of which are simulated here.

The same applies to grooming.