Hold a course in gusty flight
Can the real circuit turn visual motion into corrective steering, the reflex every flying fly relies on?
Learning curve
press TrainInside the circuit
live, current readoutSaved runs
tick runs to overlay their held out curvesNo saved runs yet. Train the real circuit and a control, save both, and compare their curves here.
What is being trained
The fly flies in the middle of a striped drum, like the tethered flight simulators of Götz and Reichardt. Gusts turn it; its own turning is seen as wide field motion by the HS cells of the lobula plate, and a dark bar ahead slides to one side of its view (LPLC1).
A readout of the 21 output neurons sets the steering torque. The score is the mean cosine of the heading error: 1 is straight on course, 0 is lost.
Training uses moderate gusts; the held out test stronger, more frequent ones.
Inspired by: Götz (1968) and Reichardt's flight simulator; Kim et al. (2017) on HS cells and steering.
How to read the result
Fixed: the wiring, the synapse counts, the transmitter signs and the neuron model (Shiu et al. 2024). Learned: 22 readout numbers.
A rising curve alone proves little: a trained readout can often make use of any network. The evidence is in the comparison. Train the real circuit, save it, train a rewired control, and overlay the two held out curves. If the real wiring learns faster or generalises better, the specific map carries information the controls lack.
Senses: Turning right → HS cells right, turning left → HS cells left; Bar drifting left or right of centre → LPLC1 of that side. Readout neurons: DNa01 left, DNa01 right, DNa02 left, DNa02 right, DNa03 left, DNa03 right, DNb05 left, DNb05 right, DNa11 left, DNa11 right, DNp01 left, DNp01 right, MDN left, MDN right, DNp09 left, DNp09 right, DNg13 left, DNg13 right, DNp03 left, DNp03 right, MN9 (proboscis).
Training uses a 0.25 ms simulation step (the lab uses 0.1 ms); firing rates in these circuits differ by about 2 percent. Dataset: Subcircuit of FlyWire v783: inputs HSE, HSN, HSS, LPLC1, LC4, LPLC2, LC16, ORN_DM1, ORN_DA2, sugar/water, bitter; outputs DNa01, DNa02, DNa03, DNb05, DNa11, DNp01, MDN, DNp09, DNg13, DNp03, CB0701; paths with at least 40 synapses per step. About this dataset