FFLYLAB01
NEURAL BEHAVIOR WORKSPACE / PROJECT / SIMULATION
Connecting

Browser demo · manual arena physics; neural training requires compute.

LIVE ENVIRONMENT

Food seeking

Each session continues from its saved brain. Stop gives food and saves.
MANUAL INPUT One press = one decision
READYSet up a scenario to begin.LEARNING
ARENA VIEW 100 × 100
Drag to orbit · Scroll to zoom · Right-drag to panLoading anatomical fly mesh…X 22.0 / Y 50.0
FoodBarrierAversive zone
EPISODE— / 5
DECISIONS0
EPISODE REWARD0.00
FOOD COLLECTED0 / 1
NEURAL TIME0.0 s

Training history

Measurements appear after the first neural decision.
REWARD PER DECISIONNO DECISIONS YET

Learning & checkpoints

Continue from a saved brain to accumulate practice.

Internal state & cue recall

Episode Task Seed Decisions Reward Food Outcome
No completed episodes

Inside the simulation

In the browser, choose an arena preset and start the demo. Use W, A, D and S, or the action buttons, to advance one decision at a time. Your actions drive the environment directly; the browser does not simulate neural activity or learning.

  1. Design: click a placement tool, then click the arena. Tune rewards and episode settings under Training.
  2. With compute: memory persists across episodes. Turn learning off for a frozen-memory comparison.
  3. With compute: pause at a decision boundary, advance one decision, and compare behavior with the exact sensory input and measured neural rates.
  4. With compute: save the brain, then choose its checkpoint under Training to start a new episode with that memory.

The fly moves on a plane displayed in 3D. Its body and wings illustrate behavior; this is not a biomechanical flight model. The sensory projection, behavior readout, and reward rules are engineered interfaces. Reward or changed weights alone do not establish learning.

Browser scenario saves stay in this browser. Configuration and demo summaries can be exported. The browser runs one manual arena episode; randomized food, other task execution, neural training, assays and checkpoints require compute. Other task scenes show their default layout.