Start with one intact run, make one change, and compare the response. This guide explains the controls and what to watch in each part of the lab.
| Control | What it does | Useful observation |
|---|---|---|
| Visual environment | Moves the sphere head-on, on the left, or on the right. The stationary option is a negative control. | Compare left and right feature drive. A stationary target gives no expansion to the resting model. |
| Circuit intervention | Clamps the chosen neuron types to zero. | Distinguish visual input, descending pathways, jumping, wing power, and the steering readout. |
| Visual input on | Enables the geometric signal entering LC4/LPLC2. Unchecking it starts from rest with no external drive. | The circuit and body should stay at rest. |
| Motor output connected | Connects the motor readouts to the body’s force rules. | With it unchecked, neural activity can remain while powered body movement disappears. |
| Camera view | Switches between Follow fly, Whole arena, Top view, and Side view. | Use Whole arena to follow the target and trajectory; use Side view for altitude. |
| Playback | Sets 0.15×, 0.35×, 1×, or 2× display speed. | The simulated equations and time steps stay the same. |
| Start / Pause / Resume | Controls playback and ongoing integration. | Pause when an event or rise in activity deserves a closer look. |
| Step 20 ms | Advances one body step, containing ten 2 ms neural updates. | Useful near the start of a response. Pause first. |
| Review recorded time | Scrubs through the history already calculated. | Resume replays that history before continuing. You cannot scrub into uncomputed future time. |
| Reset | Restarts activity, pose, and history with the currently selected settings. | Repeat a deterministic experiment. |
| Assumed lift strength | Changes the motor-to-lift gain under How neural activity becomes movement. | Test how much a trajectory depends on the body assumption. |
The green trail records the fly’s calculated trajectory. Green force arrows indicate lift command and amber arrows indicate thrust command. Their lengths are qualitative. Grid spacing is 2 au. The target’s path is prescribed and it exerts no contact force on the fly; the fly’s path follows the motor readouts and body rules.
Height above floor is measured relative to the body’s resting height. Horizontal speed uses motion along the floor plane. Wing power command is the transformed motor readout sent to the body; 50% is neither a measured muscle force nor a probability of flight. Assumed yaw command controls the modeled rate of turning.
Each dot is a population. Colors use a logarithmic activity scale so small signals remain visible. Brightness differences are therefore not proportional to activity differences. Selected links are highlighted; the layout groups populations by role and does not show their physical positions in the brain. All {{ARENA_EDGES}} included links are simulated, while 450 strong links and selected connections are drawn.
Readout rows summarize useful populations. They do not claim that each row connects directly to the next. Activity values are bounded between 0 and 1. The exact numbers are more useful than color when comparing two runs.
The disk illustrates the target’s bearing and angular size from the current pose. Numeric left/right drives show the last applied 20 ms sensory step. A rapidly changing pose can therefore produce a small difference between the current view and those values. The model uses geometry to produce features; it does not process this canvas as an image.
The three traces show visual drive, wing motor activity, and altitude. Each has its own scale. Similar heights in different rows do not imply equal quantities. The time cursor keeps the graph, body, and readouts aligned with the frame you are reviewing.
The event list marks modeled input onset, threshold-triggered launch, leaving the floor, and returning to it. These timestamps are model results. They are not experimentally measured reaction times.
| Open this setup | Question | Default-model expectation |
|---|---|---|
| Wing power silenced | Can a jump remain when wing power is removed? | TTMn can still trigger a brief jump; wing-power activity and forward powered travel disappear. |
| Visual input closed | Does movement require external drive from rest? | No activity or body movement. |
| Motor output disconnected | Are neural activity and body movement separate? | Activity can propagate, but the body receives no powered command. |
| DNp01 silenced | Is this the only route to movement? | The default early threshold launch disappears; wing-driven lift can still occur later. |
| Steering readout silenced | What depends on the b2 difference? | The adapter’s yaw command vanishes. Launch and wing power remain available. |
Each setup opens paused. These are expectations under the included equations and defaults. Once two bodies take different trajectories, their subsequent visual inputs can also differ. The results article explains that feedback effect.
Open Circuit experiment when you want a prescribed input without body feedback. It uses the original {{CIRCUIT_NODES}}-population graph. Its settings are independent of the arena.
Under Model assumptions, signed activity uses transmitter-based signs. Unsigned structural spread makes every included sign positive. The glutamate control tests an inhibitory, excitatory, or omitted effect. These controls test model sensitivity; changing a global sign does not identify the actual receptors of a neuron.
Export run in the arena downloads a JSON file containing settings, body-adapter coefficients, node identities, recorded neural activity, poses, events, and data provenance. It exports the recorded run, including any recorded frames beyond a currently selected review time.
In the circuit experiment, Download results CSV exports population summaries, and Export experiment includes trajectories and settings. Under Model & data, you can also download the prepared graphs, full-CNS comparison, and code. The downloaded package includes Python reproduction instructions.
Runs are not stored as a persistent experiment collection. Download a result you want to keep before leaving or resetting the page. Reading pages and lab links do not require you to supply an API key.