🦞 NeuroLobster

World's first embodied emulation of the stomatogastric nervous system

Overview · Demo · Quick Start · Architecture · Roadmap · References

---

"The lobster doesn't know it's in a computer. Its stomach still churns at 1Hz. Its legs still move in the same metachronal wave they've used for 400 million years."

---

Overview

NeuroLobster is a closed-loop brain-body simulation of Homarus americanus — the American lobster. We implement the stomatogastric ganglion (STG) using conductance-based Hodgkin-Huxley neurons validated against Marder Lab experimental data, and connect them to a physically simulated body running in MuJoCo 3.6.

No reinforcement learning. No reward functions. No scripted behavior. Pure biological circuit dynamics.

The same circuit that has been driving lobster stomachs for 400 million years now drives a simulated body in real time on a MacBook Pro.

STG Neurons (Brian2) → CPG Oscillator → Motor Commands → MuJoCo Body → Proprioception → STG
        ↑_______________________________________________________________|
                         Closed sensorimotor loop · 15ms sync

---

Demo

MuJoCo Simulation	Neural Activity	3D Brain Point Cloud
Lobster walking, turning, reversing	AB/PD · LP · PY live traces	STG · CoG · VNC · Brain
8 legs · metachronal wave	Pyloric rhythm ~1Hz	317 neurons · rotating

📹 Demo video coming soon — watch on Twitter

---

Quick Start

# 1. Create environment
conda create -n neurolobster python=3.11 -y
conda activate neurolobster

# 2. Install dependencies
pip install "brian2==2.5.4" "numpy<2.0" mujoco matplotlib scipy trimesh pycollada

# 3. Run full simulation (STG + MuJoCo + Brain visualization)
mjpython neurolobster_full.py

# 4. Run STG only (neural dynamics, no body)
python stg_pyloric.py

Requirements: macOS 12+ or Linux · Python 3.11 · MuJoCo 3.6 · 8GB RAM

---

Architecture

neurolobster/
│
├── 🧠 Neural
│   ├── stg_pyloric.py          # Hodgkin-Huxley STG (AB/PD, LP, PY)
│   └── stg_find.py             # Parameter search utility
│
├── 🦞 Simulation
│   ├── neurolobster_full.py    # Main: STG + CPG + MuJoCo closed loop
│   ├── neurolobster_sim.py     # CPG-only locomotion
│   └── lobster.xml             # MuJoCo body definition
│
├── 👁️ Visualization
│   └── neurolobster_brain3d.py # 3D neural point cloud + wireframe
│
├── 🗂️ Assets
│   ├── lobster.obj             # 3D body mesh (Homarus americanus)
│   ├── lambert1_albedo.png     # PBR texture — color
│   ├── lambert1_normal.png     # PBR texture — normals
│   ├── lambert1_roughness.jpeg # PBR texture — roughness
│   └── lambert1_AO.jpeg        # Ambient occlusion
│
└── 🌐 Web
    └── neurolab.html           # Lab website (Three.js + WebGL)

---

Neural Model

The STG pyloric circuit implements three identified neurons with full conductance-based Hodgkin-Huxley dynamics:

Neuron	Role	Firing Rate	Ion Channels
AB/PD	Pacemaker	~127 Hz	Na⁺, K⁺, H (Ih), Leak
LP	Lateral pyloric	~5 Hz	Na⁺, K⁺, Leak
PY	Pyloric	~1 Hz	Na⁺, K⁺, Leak

Synaptic inhibition matrix (from Marder & Bucher 2007):

         AB    LP    PY
AB  [  0.0  -0.04 -0.03 ]   ← AB inhibits LP and PY
LP  [ -0.02   0.0 -0.03 ]   ← LP inhibits PY and AB
PY  [ -0.01   0.0   0.0 ]   ← PY provides feedback to AB

Validated parameters: w_fwd=0.04 mS/cm² · tau_off=80ms · I_app=[14.0, 15.5, 15.2] μA/cm²

---

Body Simulation

<!-- lobster.xml — MuJoCo model summary -->
Joints:     17  (8 hip + 8 knee + 1 freejoint)
Actuators:  16  (motor-driven, gear=1-2)
Sensors:    8   (proprioceptive joint position)
Mesh:       Homarus americanus · 454KB OBJ · PBR textures
Physics:    timestep=2ms · gravity=-9.81 · viscosity=0.01

CPG metachronal wave phase offsets:

phases = [0.00, 0.25, 0.50, 0.75,   # Left legs L1-L4
          0.50, 0.75, 0.00, 0.25]    # Right legs R1-R4 (antiphase)

---

Emergent Behaviors

All behaviors emerge from circuit dynamics — zero behavioral code:

Behavior	Mechanism
Forward walk	Symmetric CPG · metachronal wave
Backward walk	Reversed phase offsets
Turn left	Right hemisphere amplitude × 1.6
Turn right	Left hemisphere amplitude × 1.6

---

Stack

Layer	Technology	Purpose
Neural		Spiking neural network simulation
Physics		Real-time physics engine
Math		Numerical integration
Visualization		Neural activity plots
3D Web		Brain point cloud
Mesh		3D model processing
Language		Primary language

---

Roadmap

2025 Q3  ████████████████████  ✅  Project initialization · literature review
2025 Q4  ████████████████████  ✅  HH neuron model · STG pyloric rhythm verified
2026 Q1  ████████████████████  ✅  MuJoCo body · CPG locomotion · closed loop
2026 Q2  ████████░░░░░░░░░░░░  🔄  Phase 2: Full 25-neuron STG · gastric mill
2026 Q3  ░░░░░░░░░░░░░░░░░░░░  📅  Phase 3: CoG integration · olfactory navigation
2027     ░░░░░░░░░░░░░░░░░░░░  📅  Cancer borealis full connectome

Phase 2 targets:

• Complete 25-neuron STG (pyloric + gastric mill)
• Neuromodulation switching (dopamine / serotonin)
• Feeding rhythm embodied alongside locomotion

---

References

@article{marder2007,
  author  = {Marder, E. and Bucher, D.},
  title   = {Understanding circuit dynamics using the stomatogastric nervous system},
  journal = {Neuron},
  year    = {2007}
}

@article{prinz2004,
  author  = {Prinz, A.A. and Bucher, D. and Marder, E.},
  title   = {Similar network activity from disparate circuit parameters},
  journal = {Nature Neuroscience},
  year    = {2004}
}

@article{ayers2010,
  author  = {Ayers, J. and Rulkov, N. and Knudsen, D.},
  title   = {Controlling Synchronized Oscillations in a Lobster CPG},
  journal = {Neurocomputing},
  year    = {2010}
}

---

License & Citation

MIT License · © 2025–2026 NeuroLobster Lab · Open Science

If you use this work in research, please cite:

@software{neurolobster2025,
  author  = {0xNickdev and NeuroLobster Lab},
  title   = {NeuroLobster: Embodied STG Emulation of Homarus americanus},
  year    = {2025},
  url     = {https://github.com/0xNickdev/neurolobster}
}

---

NeuroLobster Lab · Open Science · MIT License

The lobster is running. Come look.