Choose where to go.
Your controls set the drive going to Brian’s left and right legs. The banana is a destination you can see; nothing in the simulation smells it.
Field trip no. 001 · Brian's lunch break
Body: NeuroMechFlyMovement: programmed walking
Loading
Brian is getting his legs together.
Loading his little world…
How do six legs work together?
You chose a direction. A programmed rhythm coordinated Brian’s legs, and his simulated body responded to forces and to contact with the table. The program doing that coordination is called a controller.
Coordinate the steps.
Coupled rhythms carry each leg through a stepping pattern. This is a central pattern generator — a controller built from simplified mathematical rules, not a reconstructed circuit.
Meet the world.
NeuroMechFly supplies the anatomically grounded body. MuJoCo computes how that body moves and where it makes contact with the table.
NeuroMechFly and flybody are not the same thing.
Both are simulated fly bodies, both run on MuJoCo, and both need something to drive them. They are different models built by different groups for different questions — and neither one settles how a fly’s nervous system produces behaviour.
NeuroMechFly
NeLy-EPFL- Body
- Its own anatomically grounded fly model, originally derived from micro-CT imaging.
- Physics
- MuJoCo.
- Movement
- Rhythmic stepping, leg coordination rules, and controllers that combine stepping with sensory corrections.
- Research role
- Exploring how sensing, control, body mechanics and surroundings interact.
flybody
DeepMind · Janelia- Body
- A separately developed, anatomically detailed fly body model.
- Physics
- MuJoCo.
- Movement
- Reinforcement-learning tasks and trained controllers for walking and flight.
- Research role
- Exploring body mechanics and embodied control, including realistic locomotion.
Brian is walking on NeuroMechFly’s body under its programmed rhythms. FlyBrian’s DigiFly interactive view uses flybody, so you can feel the difference between the two for yourself.
A moving fly gives us a behaviour to explain.
In a living fly, movement involves the brain, the limb-coordinating circuits of the ventral nerve cord, and sensory feedback from the body. A controller that walks well has reproduced the behaviour — it has not shown that it reproduced the circuitry. That gap is the interesting part.
FlyBrian is a workspace for building and comparing neural simulations on Janelia connectome releases, inspecting the activity they produce, and investigating the mechanisms behind behaviour. Brian’s afternoon is the front door.
See what FlyBrian isThe rest of Brian’s afternoon.
02
The driving testOn the drawing board
03
Brian after hoursOn the drawing board
04
Wall Street, somehowOn the drawing board
Body model and walking controller adapted from NeuroMechFly (NeLy-EPFL, Apache-2.0), simulated with MuJoCo (Google DeepMind). Pinned versions and the exact changes are recorded in the model notice.