Aimbug
Browser FPS: full MaleCNS steers via DNa02 / DNp53 and fires one shot per pIP10 courtship-song bout while LC10 “sight” is open. No trained readout.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The LIF constants are the published Shiu et al. 2024 parameters, stated in source with the reference.
https://github.com/slickdomi/aimbug/blob/7af0899e938d52801459343f9f340e3da0c6ffcb/web/src/config.ts#L6-L21The model is documented as Shiu et al. 2024 LIF plus adaptation, with a separate graded rate model for the optic lobe.
https://github.com/slickdomi/aimbug/blob/7af0899e938d52801459343f9f340e3da0c6ffcb/README.md#L139The authors state there is no trained readout and no aim logic; behaviour comes from fixed wiring.
https://github.com/slickdomi/aimbug/blob/7af0899e938d52801459343f9f340e3da0c6ffcb/README.md#L14The simulation parameters uploaded to the GPU are membrane and synaptic decay constants, threshold, reset and refractory period — no weight update term exists.
https://github.com/slickdomi/aimbug/blob/7af0899e938d52801459343f9f340e3da0c6ffcb/web/src/sim/brain.ts#L208-L234
A well-documented neuron model over the real MaleCNS graph: leaky integrate-and-fire with Shiu et al. 2024's published parameters (0.275 mV per synapse, tau_m 20 ms, tau_s 5 ms, 1.8 ms delay, 2.2 ms refractory), plus spike-frequency adaptation and a graded rate model for the 95,501 optic-lobe cells, with the constants in config.ts mirroring a NumPy reference. Synapse signs come from predicted transmitter. Not L2: no weight is written anywhere in the source — a repo-wide grep for plasticity, learning rates, eligibility or optimizers finds nothing, the connectome edge weights are read-only GPU buffers, and the README states outright that 'There is no trained readout and no aim logic'. Aim comes from fixed wiring (LC10 -> AOTU019/025 -> DNa02), which the authors support with silencing ablations and mirrored-gain controls rather than with a learning claim.
Record
Nearby
All projects →MaleCNS simulation connected to ViZDoom through modeled visual inputs and a fixed button readout. Includes plasticity experiments and negative validation results; learned survival has not been demonstrated.
Research framework comparing a MaleCNS-constrained sparse recurrent controller with rewired graphs, conventional neural networks, and other controls on simple VizDoom tasks.
Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.