Arena view
A neural control experiment
The full spiking network drives an engineered two-wheel decoder. The rendered camera is available for visual inspection; FlyBrain receives simplified synthetic left/right sensory signals, not raw camera pixels.
Action and observation log
What this experiment does and does not establish
This is a kinematic simulation, not a measured digital twin of your robot. Wheel speed, track width, body collision radius, servo axes, friction-free motion and arena dimensions are nominal. The robot does not lift its feet or grasp objects. The camera is a rendered chest-camera viewpoint, not a physical feed.
The pinned FlyBrain graph contains 139,255 LIF neurons and 2,698,236 edges. The moving condition uses experimental recurrent gain 50 and 1,305 annotated descending neurons. Synthetic beacon/contact signals are mapped to coarse sensory pools; the two-wheel decoder reads only neural outputs. Original-gain, edge-disconnection and output-silencing controls are selectable. This is not biological calibration or evidence of reliable robot navigation.
Live chat mode is a separate experiment. A private SG MCP connection permits this assistant to observe the simulated room, request a bounded motion, and inspect the result. The public page is a read-only spectator. Motion expires after at most two seconds without another command. An open chat tab does not create a continuously running assistant. Replay is explicitly labelled and uses recorded states, not new decisions.
Local mode computes in this browser with vendored assets, no AI-provider calls. Live chat mode runs the small physics service on GETH. This chat is not an on-device language model. No real Bluetooth connection, camera or motor has been activated.