R. C. SîmpetruContact

n-squared lab · FAU · Erlangen

Neural interfaces for hand control.

Clean-line comic illustration of Raul C. Sîmpetru presenting a live demo in the n-squared lab in Erlangen: he raises his hand toward a projection of motor-unit spike trains and a pixel hand with one finger lit, beside a recording rigClose-up drawing of his raised right hand with the bracelet on his wrist, the ring finger bent like the lit finger of the hand on the screen

Dr.-Ing. defence expected December 2026

I build neural interfaces that turn muscle activity into control of virtual hands, orthoses and prostheses.

Selected publications Download CV

TNSRE · 2023

The whole hand, in real time

More than 20 degrees of freedom decoded live from forearm EMG, with under 50 ms latency.

TNSRE · 2024

Control after spinal cord injury

98.3% mean accuracy across 8 attempted movements in 8 tetraplegic participants.

Science Advances · 2025

Open software for assistive devices

With MyoGestic, 7 participants controlled a virtual hand, an orthosis, a prosthesis and a cursor.

Featured work

Research areas.

Three landmarks
Clean-line comic desert with three landmarks joined by a ribbon cable: a stone hand wearing an electrode band, a robotic hand on a pedestal steered by a small traveller, and a tree shaped like a neuron
Decoding

Decoding hand movement from forearm EMG

We use forearm muscle recordings to estimate hand movement in real time, then investigate what the network learns.

From spinal motor neurons to a 22-degree-of-freedom hand model
TNSRE 2023TBME 2024J. Neural Eng. 2024
MyoGestic

Decoding spared muscle activity for assistive control

We train decoders with participants to turn spared muscle activity into control of assistive devices.

A participant wearing the EMG bracelet controls a virtual hand, an orthosis, a prosthesis and a cursor
TNSRE 2024Science Advances 2025
MyoGen

Simulating neural drive, muscle and EMG

MyoGen links models of neural drive, muscle and EMG, letting us test decoders against known simulated signals.

Cortical neurons, spinal motor neurons, afferents, muscle and the resulting EMG, simulated
bioRxiv 2026
Papers

Selected publications.

Click a paper to open it
Clean-line comic illustration of a pastel desert: Raul, seen from behind in a linen shirt, scarf, satchel and desert boots with a bamboo staff, stands on a dune ridge looking at a giant segmented robotic hand half-buried in the sand, a ribbon cable snaking toward him
EMBC
2022
TNSRE
2023
JNETBMETNSRE
2024
Sci. Adv.
2025
bioRxiv
2026

Decoding hand kinematics from EMG

  1. Camera setup, raw HD-sEMG, the 3D convolutional network and predicted joint angles

    14 jointspredicted from forearm EMG, 2.78° mean absolute error

    Paper ↗

    Accurate continuous prediction of 14 degrees of freedom of the hand from myoelectrical signals through convolutive deep learning. Raul C. Sîmpetru, Marius Osswald, Dominik I. Braun, Daniela S. Oliveira, Andre L. Cakici, Alessandro Del Vecchio. 44th IEEE EMBC · Glasgow, July 2022.

  2. Neural-to-EMG pathway and real-time hand-kinematics estimates compared with camera-based estimates

    >20 DoFdecoded live, under 50 ms latency, 32 predictions a second

    Paper ↗

    Proportional and simultaneous real-time control of the full human hand from high-density electromyography. Raul C. Sîmpetru, Michael März, Alessandro Del Vecchio. IEEE Trans. Neural Syst. Rehabil. Eng. 31 · July 2023.

Filtering and learned representations

  1. Neural information, movements and EMG, then filtering, a neural network and a comparison of filters

    320electrodes; best with all of them and no low-pass filtering

    Paper · Open access ↗Preprint ↗

    Influence of spatio-temporal filtering on hand kinematics estimation from high-density EMG signals. Raul C. Sîmpetru, Vlad Cnejevici, Dario Farina, Alessandro Del Vecchio. J. Neural Eng. 21 · March 2024 · Open access.

  2. Neural pathway, the movement set, electrode grid placement and the 22-degree-of-freedom hand model

    22 DoFkinematics and force from 320 EMG sensors

    Article ↗Code ↗Paper · Open access ↗

    Learning a hand model from dynamic movements using high-density EMG and convolutional neural networks. Raul C. Sîmpetru, Andreas Arkudas, Dominik I. Braun, Marius Osswald, Daniela Souza de Oliveira, Bjoern Eskofier, Thomas M. Kinfe, Alessandro Del Vecchio. IEEE Trans. Biomed. Eng. 71 · 2024.

Assistive control after injury or amputation

  1. Uninjured and motor-complete SCI participants, EMG and kinematics feed a deep learning model whose latent space separates gestures and groups

    98.3%mean accuracy across 8 attempted movements, 8 tetraplegic participants

    Paper · Open access ↗

    Identification of spared and proportionally controllable hand motor dimensions in motor complete spinal cord injuries using latent manifold analysis. Raul C. Sîmpetru, Daniela Souza de Oliveira, Matthias Ponfick, Alessandro Del Vecchio. IEEE Trans. Neural Syst. Rehabil. Eng. 32 · October 2024 · Open access.

  2. A participant wearing the EMG bracelet controls a virtual hand, an orthosis, a prosthesis and a cursor

    7 participantscontrol demonstrated with a virtual hand, orthosis, prosthesis and cursor

    Paper ↗Code ↗Archive ↗

    MyoGestic: EMG interfacing framework for decoding multiple spared motor dimensions in individuals with neural lesions. Raul C. Sîmpetru†, Dominik I. Braun†, Arndt U. Simon, Michael März, Vlad Cnejevici, Daniela Souza de Oliveira, Nico Weber, Jonas Walter, Jörg Franke, Daniel Höglinger, Cosima Prahm, Matthias Ponfick, Alessandro Del Vecchio († equal contribution). Sci. Adv. 11, eads9150 · April 2025.

Biophysical simulation of neuromotor activity

  1. Cortical neurons, interneurons, motor neurons, afferents, a Hill-type muscle, volume conductor and resulting surface and intramuscular EMG

    Cortex to EMGsimulation of neural drive, muscle force, and surface and intramuscular EMG

    Preprint ↗Docs ↗Code ↗

    MyoGen: unified biophysical modeling of human neuromotor activity and resulting signals. Raul C. Sîmpetru*, Ricardo G. Molinari*, Devon R. Rohlf, Rebeka L. Batichotti, Renato N. Watanabe, Leonardo A. Elias, Alessandro Del Vecchio (* joint first authors). bioRxiv · January 2026.

Twelve more, as co-author
Software

Research software.

Open source
Clean-line comic workshop under a striped awning: a cloaked figure seen from behind repairs a segmented robotic hand at a workbench, a hover-glider parked beside it

Real-time myocontrol software with decoder calibration during participant sessions.

MyoGesticFramework · Python

Records, trains decoders and controls in real time.

Co-creator and core developer · v2.7.0

Docs ↗GitHub ↗Zenodo ↗

MyoGestic-VHI3D hand · Godot/.NET

Real-time 3D hand visualisation for HD-sEMG, driven over LSL and gRPC.

Author and maintainer

Docs ↗GitHub ↗

Biophysical simulation from neural drive to EMG.

Author and maintainer · v0.13.0

$ pip install myogen
Docs ↗GitHub ↗
Tools

Smaller tools.

github.com/RaulSimpetru

Recording and tracking

Rust · CLI

LSL Recording Toolbox

Records several Lab Streaming Layer streams in sync to Zarr, with tools to manage and analyse the recordings.

GitHub ↗Docs ↗
Rust

Muovi LSL interface

Streams the OT Bioelettronica Muovi EMG device into Lab Streaming Layer in real time.

GitHub ↗Docs ↗
Rust

Sessantaquattro LSL interface

Streams the OT Bioelettronica Sessantaquattro into Lab Streaming Layer in real time.

GitHub ↗Docs ↗
Python · napari plugin

napari-cotrack

Label keypoints on a few frames; CoTracker3 tracks them through the rest of the video.

GitHub ↗

Research workflow

PythonMCP

Zotero Library MCP

Lets Claude, Codex and ChatGPT add papers and books to Zotero by DOI, arXiv ID or ISBN, and manage collections and tags.

GitHub ↗
TypeScript

Synology Site Hub

Live previews of every git branch of your websites, on your own Synology NAS behind its login.

GitHub ↗Demo ↗
Python · local LLM

GrantBridge

Watches German and EU funding calls, ranks them against a researcher’s confirmed capabilities and drafts project concepts.

GitHub ↗
PythonMCP

Seatsurfing MCP

Book desks and rooms in Seatsurfing from Claude Code.

GitHub ↗
Objective-C · macOS

Netron Quick Look

Preview machine-learning model files in Finder with Netron: select one and press Space.

GitHub ↗
Python

PDF Form Filler

Fills PDF forms with text and checkboxes so they render correctly in any viewer.

GitHub ↗
Experience

Research, teaching, talks.

The full CV (PDF) ↗
Clean-line comic amphitheatre carved into terracotta rock: a cloaked speaker stands in front of a large stone screen while an audience in cloaks, seen from behind, listens
Three panels: the delegation from Shanghai General Hospital with members of the lab in front of the FAU building at Nürnberger Straße 74, the projected hand and muscle-signal grid of the lab demo, and Raul C. Sîmpetru presenting the demo at the recording rig
Photo: Raul C. Sîmpetru at a lectern at Georgian Technical University in Tbilisi, presenting MyoGestic
Photo: Raul C. Sîmpetru at the lectern at ISEK 2026 in Jyväskylä, presenting a slide titled The neural manifold
On screen01 / 03

Erlangen, Germany · 4 November 2026Visit of a delegation from Shanghai General HospitalTbilisi, Georgia · 2026Public lecture at Georgian Technical UniversityJyväskylä, Finland · June 2026Invited symposium talk at ISEK 2026

Lab news ↗Congress page ↗
Now

Doctoral researcher

n-squared lab, FAU Erlangen-Nürnberg

Dr.-Ing. defence expected December 2026

Now

Invited professor

Kutaisi International University

Teaching neurocomputing, robotics, deep learning and DSP

Talks and recognition

TalkRecognition
2023
Recognition

Young Talent Award, 3rd prize

DMEA

Talk

Selected oral talk

The Physiological Society

2024
Recognition

Featured Article

IEEE TBME

Talk

Oral talk

ISEK

2025
Talk

Rapid-fire talk

International Motoneuron Society

2026
Talk · June

Invited symposium talk

ISEK, Jyväskylä ↗

Recognition

Best Student Paper, co-author

IEEE CBS

Public lecture

Muscles, Machines, and Movement

Georgian Technical University, Tbilisi ↗

Before
  • Research and teaching assistantFAU Erlangen-Nürnberg
  • Research assistantFraunhofer IIS
Education
  • Dr.-Ing. candidateFAU Erlangen-Nürnberg
  • MSc and BSc Medical EngineeringFAU Erlangen-Nürnberg
Teaching
  • Neurocomputing, robotics, deep learning, DSPKutaisi International University
  • Seminars and thesis supervisionFAU Erlangen-Nürnberg

How this drawing was made

The original photo: Raul C. Sîmpetru presenting the live demo in the lab, his hand raised toward the projection of motor-unit spike trains and a pixel hand
Model
OpenAI Codex (gpt-6-astra) with its built-in image_gen tool
Inputs
This photo, plus two of the site's own illustrations as style references
Passes
A full redraw, then one correction
Close-up
The hand zoom was drawn the same way, from a crop of this photo

Prompt · redraw

Use case: style-transfer with identity and composition preservation. Make ONE finished image, lab-screen, widest landscape available, exact 16:9 aspect ratio. Inputs: raul_1.jpeg is the edit target and absolute composition/identity reference. style-1.png and style-2.png are art-style references only. Completely redraw the laboratory photograph as a clean-line European science-fiction comic illustration in the Moebius / Sable visual style shown in the two references. Thin uniform black ink contours, sophisticated flat cel colour, utterly clean smooth fills, no texture or grain, no hatching, no photographic shading, no 3D rendering. Pastel sand, cream, ochre, terracotta, dusty teal, sage and lavender, with charcoal clothing. Keep the original photograph's exact perspective, camera angle, cropping, relative object positions and scale, and 16:9 framing. Do not zoom in or rearrange the composition. A man stands left of centre in the lower part of frame, roughly x=27%-39%, beside a lab demo rig. Keep him recognisable from the photo: short dark-brown hair, rectangular dark glasses, short dark beard, same slim build; right hand raised in the exact explanatory gesture, left arm down, head and gaze angled up/right toward the projection. Same charcoal grey blazer, dark navy shirt, grey trousers, white blank badge on dark lanyard, dark wristband. Preserve the distinctive face in flat drawn form. Two tall dark-framed windows at left, showing simplified pale buildings as in the photo. Ceiling-mounted projector partly cropped at upper left. A dark monitor beside the presenter, aluminium-frame recording rig with purple brackets and looping black cables on the desk at lower centre, matching the photograph's size and placement. Grey poster board in the right foreground, cropped as in photo, with only abstract unreadable colour blocks for its content. Left foreground audience member is a single simple dark silhouette from behind with no face; preserve their original crop and scale. The lower-right equipment/table edge stays where it was. The large upper-right wall projection shows a grid of motor-unit spike raster vertical bars, pale teal and cream, and a white blocky pixel hand with one finger highlighted blue, matching the source's placement and overall appearance. Make the projection feel softly luminous through pale flat colours and at most a very subtle soft border; no text anywhere in it. Negative space for future website overlay: the wall between the right window and projection, ceiling and upper wall must stay plain, flat, calm and nearly empty. Simplify ceiling into a few broad planar shapes with no speckle, tile texture, busy lines or added vents. Maintain clear light areas. Absolutely NO lettering, numbers, glyphs, logos, signatures or watermark anywhere, including projection, badge, equipment and poster. No desert exterior in this first version. Render the entire image as original flat ink-and-colour illustration, not as a filter on the photograph.

Prompt · correction

Edit this illustration with only the following focused corrections. Preserve the laboratory, composition, all object positions and scale, presenter identity and pose, all clothing, the raster and pixel-hand projection, and the wide 16:9 framing exactly. 1. Replace the entire left foreground audience person with ONE uniform flat dark charcoal silhouette, seen from behind, with absolutely no face, no visible facial profile, no skin or ear colours, no interior detail. Preserve the overall cropped head-and-shoulders outline in the bottom-left. Remove the little adjacent visible face too, merging it into the single silhouette if necessary. 2. Make room walls and ceiling strictly FLAT solid colour areas with no gradients, grain, texture or photographic light falloff. Use warm pale cream for wall planes and a calm pale sage-grey ceiling. Keep the wall between the right window and projection and the upper room calm and nearly empty, retaining original perspective. Keep thin uniform black ink lines and flat cel colours throughout in the supplied Moebius/Sable European comic art style. The projection alone may have a very faint luminous edge, with a flat pale dusty teal field. 3. No text, logos, numbers, glyphs or watermark anywhere. Deliver one landscape PNG with exact 16:9 pixel dimensions, preferably 1920x1080 or the widest supported exact 16:9 size. Do not crop or shift the image contents.