refactor(topics): 멀티 에이전트용 지식 재편 — _Common(공통 기본기) + Domain_* 구조
에이전트 8종(대화형/프로그래머 C·S/디자이너/설계자/기획자/QA/PD/PM)에게 [공통 기본 능력 + 롤별 Specialty] 2층으로 지식을 주입하기 위한 재분류. 문서 내용·포맷은 무수정, 폴더 이동만 (6,372개 문서 수 보존 확인). - Topic_Programming → Domain_Programming (내부 구조 보존) - Topic_Graphic → Domain_Design - Topic_Business → Domain_Product - Topic_General → Domain_General - _Common 신설: Math(구 Topic_Math_Specialty), Reasoning(구 General/From_Thinking & Reasoning), Reasoning_Creativity(구 General/From_창의성), Communication(Poetic_Blog_Writing + From_writing) - 타 도메인의 From_* 폴더는 유지 (출처 표기일 뿐, 이미 도메인에 맞게 분류된 문서) - 빈 폴더 정리 (memory/procedures) - 에이전트→폴더 매핑은 workspace의 .astra/agent-knowledge-map.json (9개 에이전트) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Antigravity Agent
@@ -0,0 +1,143 @@
|
||||
---
|
||||
id: wiki-2026-0508-neuromuscular-control
|
||||
title: Neuromuscular Control
|
||||
category: 10_Wiki/Topics
|
||||
status: verified
|
||||
canonical_id: self
|
||||
aliases: [신경근 조절, Motor Control, NMS Control]
|
||||
duplicate_of: none
|
||||
source_trust_level: A
|
||||
confidence_score: 0.9
|
||||
verification_status: applied
|
||||
tags: [neuroscience, biomechanics, motor-control, robotics, biomedical]
|
||||
raw_sources: []
|
||||
last_reinforced: 2026-05-10
|
||||
github_commit: pending
|
||||
tech_stack:
|
||||
language: python
|
||||
framework: opensim
|
||||
---
|
||||
|
||||
# Neuromuscular Control
|
||||
|
||||
## 매 한 줄
|
||||
> **"매 brain plans, spine reflexes, muscle executes"**. Neuromuscular control 은 CNS 가 motor neuron 을 통해 muscle 활성화를 조절해 movement 를 produce 하는 hierarchical process. 2026 perspective 에서 EMG-driven simulation, exoskeleton control, BCI prosthetics 의 핵심.
|
||||
|
||||
## 매 핵심
|
||||
|
||||
### 매 hierarchy
|
||||
- **Cortex (M1, PMC, SMA)**: motor planning.
|
||||
- **Cerebellum**: timing, coordination, error correction.
|
||||
- **Basal ganglia**: action selection, gain modulation.
|
||||
- **Spinal cord**: reflex circuits, central pattern generators (CPG).
|
||||
- **Motor unit**: α-MN + muscle fibers (final common pathway).
|
||||
|
||||
### 매 control principles
|
||||
- **Size principle (Henneman)**: small MN 먼저, large 나중.
|
||||
- **Co-contraction**: agonist + antagonist 동시 활성화 → stiffness 조절.
|
||||
- **Stretch reflex**: muscle spindle Ia → α-MN monosynaptic.
|
||||
- **Equilibrium point hypothesis**: descending command = desired length.
|
||||
|
||||
### 매 응용
|
||||
1. Prosthetic / exoskeleton control.
|
||||
2. Rehabilitation robotics.
|
||||
3. Surgical motor mapping.
|
||||
4. Sport biomechanics.
|
||||
|
||||
## 💻 패턴
|
||||
|
||||
### Hill-type muscle model
|
||||
```python
|
||||
import numpy as np
|
||||
|
||||
def hill_muscle(activation, length, velocity,
|
||||
F_max=1000, l_opt=0.1, v_max=10):
|
||||
f_l = np.exp(-((length - l_opt) / (0.5 * l_opt))**2)
|
||||
if velocity <= 0:
|
||||
f_v = (v_max + velocity) / (v_max - 4 * velocity)
|
||||
else:
|
||||
f_v = (1.8 - 0.8 * (v_max + velocity) / (v_max - 7.56 * velocity))
|
||||
return F_max * activation * f_l * f_v
|
||||
```
|
||||
|
||||
### Motor unit recruitment (size principle)
|
||||
```python
|
||||
def recruit(excitation, n_units=100, threshold_max=1.0):
|
||||
thresholds = np.linspace(0.05, threshold_max, n_units)
|
||||
return np.where(excitation > thresholds,
|
||||
(excitation - thresholds) / (1 - thresholds), 0)
|
||||
```
|
||||
|
||||
### EMG → activation mapping
|
||||
```python
|
||||
from scipy.signal import butter, filtfilt
|
||||
|
||||
def emg_to_activation(emg_raw, fs=1000):
|
||||
b, a = butter(4, [20, 450], btype="band", fs=fs)
|
||||
emg = filtfilt(b, a, emg_raw)
|
||||
emg = np.abs(emg)
|
||||
b, a = butter(4, 6, btype="low", fs=fs)
|
||||
env = filtfilt(b, a, emg)
|
||||
return env / env.max()
|
||||
```
|
||||
|
||||
### Inverse dynamics (joint torque)
|
||||
```python
|
||||
def inverse_dynamics(theta, theta_dot, theta_ddot, m=5, l=0.4, g=9.81):
|
||||
I = m * l**2 / 3
|
||||
return I * theta_ddot + 0.5 * m * g * l * np.sin(theta)
|
||||
```
|
||||
|
||||
### CPG (Matsuoka oscillator)
|
||||
```python
|
||||
def matsuoka_step(x1, x2, v1, v2, u=1.0, beta=2.5, tau=0.1, dt=0.001):
|
||||
y1, y2 = max(0, x1), max(0, x2)
|
||||
dx1 = (-x1 - beta*v1 - 2.0*y2 + u) / tau
|
||||
dx2 = (-x2 - beta*v2 - 2.0*y1 + u) / tau
|
||||
dv1 = (-v1 + y1) / (tau * 12)
|
||||
dv2 = (-v2 + y2) / (tau * 12)
|
||||
return x1 + dx1*dt, x2 + dx2*dt, v1 + dv1*dt, v2 + dv2*dt
|
||||
```
|
||||
|
||||
### EMG-driven prosthetic control
|
||||
```python
|
||||
class MyoelectricController:
|
||||
def __init__(self, n_channels=8):
|
||||
self.classifier = train_lda()
|
||||
def predict(self, emg_window):
|
||||
feat = extract_features(emg_window) # MAV, ZC, WL, AR4
|
||||
return self.classifier.predict(feat)
|
||||
```
|
||||
|
||||
## 매 결정 기준
|
||||
| 상황 | Approach |
|
||||
|---|---|
|
||||
| Whole-body simulation | OpenSim / MyoSuite |
|
||||
| Single joint, real-time | Hill-type + LDA EMG |
|
||||
| Locomotion robot | CPG + reflexes |
|
||||
| Pathology study | Inverse dynamics + EMG |
|
||||
|
||||
**기본값**: Hill-type muscle + size-principle recruitment + EMG envelope.
|
||||
|
||||
## 🔗 Graph
|
||||
- 변형: [[Perceptual-Motor-Skills]]
|
||||
- Adjacent: [[Reinforcement Learning]] · [[BCI]]
|
||||
|
||||
## 🤖 LLM 활용
|
||||
**언제**: simulation parameter tuning, EMG feature engineering, paper synthesis.
|
||||
**언제 X**: clinical motor diagnosis — neurologist 필수.
|
||||
|
||||
## ❌ 안티패턴
|
||||
- **Linear EMG-force assumption**: 매 force 는 nonlinear (length × velocity × activation).
|
||||
- **Ignoring co-contraction**: stiffness control 무시 → unstable model.
|
||||
- **Pure feedback control**: feed-forward (internal model) 누락 → laggy.
|
||||
|
||||
## 🧪 검증 / 중복
|
||||
- Verified (Zajac 1989, Delp OpenSim 2018, Henneman 1965).
|
||||
- 신뢰도 A.
|
||||
|
||||
## 🕓 Changelog
|
||||
| 날짜 | 변경 |
|
||||
|---|---|
|
||||
| 2026-05-08 | Phase 1 |
|
||||
| 2026-05-10 | Manual cleanup — Hill model + EMG + CPG 패턴 |
|
||||
Reference in New Issue
Block a user