metadata
license: cc-by-4.0
task_categories:
- tabular-classification
- robotics
tags:
- imu
- biomechanics
- motion-capture
- physical-ai
- prosthetics
- robotics
- sensor-certification
- emg
- ninapro
- gesture-recognition
pretty_name: NinaPro DB5 — S2S Physics Certified
size_categories:
- 1K<n<10K
NinaPro DB5 — S2S Physics Certified (v1.7.0)
Physics-certified windows from NinaPro DB5 forearm EMG+IMU dataset. Each window validated against 8 biomechanical laws using S2S.
Bad training data costs you months. S2S finds it in milliseconds.
What this adds
| Column | Description |
|---|---|
tier |
GOLD / SILVER / BRONZE / REJECTED |
score |
0–100 physics compliance score |
laws_passed |
Which of 8 laws passed |
verdict |
Human-readable quality statement |
recommendation |
Actionable engineering guidance |
wavelet_signal_type |
biological / mechanical_synthetic / random_noise |
wavelet_cv |
Energy drift CV (real human: 0.15–1.5) |
wavelet_entropy |
Spectral entropy (biological: 0.75–0.96) |
issues |
Specific failures with hardware fix suggestions |
Statistics
| Metric | Value |
|---|---|
| Windows | 1,500 |
| Subjects | 10 |
| Sample rate | 2000Hz |
| Window size | 500 samples (250ms) |
| GOLD | 109 (7.3%) |
| SILVER | 1,388 (92.5%) |
| BRONZE | 3 (0.2%) |
| REJECTED | 0 |
Quick start
import pandas as pd
df = pd.read_csv("ninapro_db5_certified.csv")
# High quality windows only
train = df[df['tier'].isin(['GOLD', 'SILVER'])]
# Confirmed biological signal
bio = df[df['wavelet_signal_type'] == 'biological']
# Score threshold
high = df[df['score'] >= 70]
The 8 Physics Laws
| Law | What it catches |
|---|---|
| Newton F=ma | EMG-acceleration timing mismatch |
| Segment Resonance | Non-physiological tremor frequency |
| Rigid Body Kinematics | Decoupled accelerometer/gyroscope |
| Ballistocardiography | Missing heartbeat signal in IMU |
| Joule Heating | EMG-thermal mismatch |
| Motor Control Jerk | Superhuman motion (>500 m/s³) |
| IMU Consistency | Independent signal generators |
| Inter-window Continuity | Teleportation / data splices |
Source dataset
Original NinaPro DB5: https://ninapro.hevs.ch/instructions/DB5.html
10 subjects, forearm EMG (16ch) + accelerometer (3ch) at 2000Hz.
Certification engine
pip install s2s-certify
from s2s_standard_v1_3.s2s_physics_v1_3 import PhysicsEngine, audit_report
engine = PhysicsEngine()
result = engine.certify(imu_raw=window, segment='forearm')
report = audit_report(result)
print(report['verdict'])
github.com/timbo4u1/S2S | DOI: 10.5281/zenodo.18878307