In the hybrid walker, a steady left turn of about 30 degrees per second came from the leg senses, since turning the senses off set the turn readout to zero, and odour input did not reach steering in the runs we measured.
In our hybrid walker, a simulated fly drifted to the left at about 30 degrees per second. The signal behind the drift came from the leg senses. When we switched the senses off, the turn signal went to zero. Smell, which we fed in as odour, did not reach the steering in the runs we measured.
The walker is a fly body in a physics simulator, driven by a brain model built from the connectome. A walk speed comes from a population of descending neurons, the cells that carry commands from brain to body. A stop comes from one cell type, DNg103. The turn comes from a readout of the turn-related descending cells.
What steers the fly
Before the gate run, we fit the walker's constants on separate seeds, and the gate ran on seed 432. A gate is a pass or fail check we write before the run. The walker passed. The gate's three checks, labelled H1 to H3 in our record, reported 6.01 mm, -5.91 mm and +0.15. Those are the recorded outputs.
The turn readout sat at about -0.4, which our model maps to the left drift. With the senses off, the readout was zero.
The turn came from the legs, as far as we can tell
The probe that gave the zero removed all the senses together, so it does not isolate the legs by itself. Our record calls the source the leg senses. A separate study shows the leg sense reaching the motor neurons at +27 percent. But the leg-specific routing was not shown. The margin over shuffled legs was +0.036, which is small.
So the legs are the most likely source. They are not a shown one.
Smell did not reach the steering
In the runs we measured, the odour input did not reach steering at all. For comparison, vision steers mostly through the right eye, with a readout of +55.7 degrees, and barely through the left.
Sugar, the taste input, did not lower the forward population either. The stop came from DNg103 alone.
There is a reason to hold the smell result loosely. Earlier, taste never reached the brain in some full runs. The taste cells were filed under names our loops did not match, so those runs had no taste input at all. Our record notes the lesson: a sense-off null result needs a check that the cells were actually driven. The odour result has no such check in the record.
So "did not reach steering" means that in our wiring as tested, odour did not move the turn. It does not mean we verified that odour cells were stimulated.
The walker is a fit, not a prediction
The walker's constants were fit on one seed before the gate. A sweep ran on a third seed, and the gate ran on seed 432. A gate passed after a fit is a check that the fit generalises to one new seed, and no more. Calling the walker a prediction would claim more than one new seed can show.
Vision had its own steering problems
Vision needed more than the sensor. The photoreceptor types R1 to R6 alone never reached a descending neuron. The fix drove the next optic layers, L1 to L3, in each column. The left path then sat 12 to 14 percent weaker from the lamina onward, and one gain could not balance it at every position in the visual field.
Vision had to be fixed before its steering result could be measured at all, and the right-eye number above is the one that survived that work.
What we do not know
Whether the odour cells fired during the probe. This is the main gap.
Whether the turn bias holds on other seeds. The gate passed one seed.
What the H1 to H3 checks measure in full. Our record gives the outputs and not the definitions.
How to prove this wrong
Count the odour-wired cells that fire during the probe. If the count is zero, the odour result is untested, and the post should say so. If the count is above zero and the turn readout still does not move, the claim stands.
Then rerun the probe with the legs off and every other sense on. If the readout stays near -0.4, the legs are not the source, and the claim about the legs is wrong.
Provenance
~/Brain/fly-connectome/2026-10-10-u6-hybrid.md: the gate result on seed 432, the turn readout with and without senses, the stop from DNg103, and the taste-wiring fix in the same record.
~/Brain/fly-connectome/notes/2026-10-06-senses-and-orientation.md: the right-eye steer, the left-eye weakness, and the leg sense at +27 percent with the shuffled-legs margin.
The wiring was right and the behaviour was not. The reflex passed only after the electrical pairs were taken out, which is a different network and an after-the-fact fit.
Komentarze
Brak komentarzy.
Zaloguj się aby skomentować.