Why it is worth attention

It provides an open‑source, pretrained model that achieves strong fingertip force prediction from surface EMG, with a simple calibration procedure that reduces error roughly 5× for robot demonstration transfer.

Who it is for

Robotics researchers working on dexterous manipulation
Human‑computer interaction engineers developing wearable input devices
Prosthetics developers seeking non‑invasive force estimation
Academics studying EMG‑based force inference

Use cases

Transferring human hand force skills to robot hands via calibration
Estimating grip forces for prosthetic or exoskeleton control
Analyzing fine motor skills or rehabilitation progress
Enabling force‑aware interactive systems with low‑cost hardware

Strengths

Open‑source pretrained checkpoint and validation data released under Apache‑2.0
Calibration on a small demo set cuts force prediction error 5× (MAE from 46.0 to 9.5)
Strong thumb/index/middle force tracking (R² 0.68/0.65/0.50) on held‑out sessions
Low‑cost hardware design with publicly available BOM and EMG hardware survey

Considerations

Full pretraining dataset and DIY hardware tutorial are not yet released
Ring and pinky finger forces are noisier and carry less force
Calibration is required when transferring to a new setup (different sensor placement or environment)

README quick start

ForceBand: Learning Forceful Manipulation with sEMG

Official EMG2Force model code for "ForceBand: Learning Forceful Manipulation with sEMG" (arXiv:2606.26093 · project page).

ForceBand is a low‑cost wrist‑worn sEMG + IMU band; its EMG2Force model predicts per‑finger fingertip forces (thumb, index, middle, ring, pinky) from the 8‑channel sEMG + IMU stream — a DINOv2 ViT‑S spectrogram encoder + Transformer (~28.5 M params) that conditions on the previous window's force for continuity.

EMG2Force on a held‑out validation session — predicted (color) vs. ground‑truth (grey) per‑finger force. -->

What ships

Artifact	Status	Where
✅ Pretrained checkpoint (DINOv2, Apache‑2.0)	released	`checkpoints/`
✅ Validation set — held‑out pretraining chunks, ~5‑min quick checkpoint test	released	`data/val/` · `data/splits/val.txt`
✅ Deploy‑demo set — calibration + held‑out validation	released	`data/demos/` · `data/splits/{calib,demo_val}.txt`
⏳ Full pretraining dataset	under preparation	—
➕ RGB overlay videos (optional, ~1.3 GB)	Google Drive	`scripts/download_videos.py`

Roadmap

Before paper acceptance

Hardware Bill of Materials (BOM)
Hardware DIY tutorial (build your own ForceBand)
EMG hardware survey — sEMG front‑ends / electrodes / acquisition boards compared
Full pretraining dataset
Robot policy training pipeline
DINOv3 backbone variant — its license needs further validation on our side before we redistribute, so for now we ship DINOv2; planned as an opt‑in download‑DINOv3‑yourself recipe (the two perform on par — see Backbones)

Install

conda env create -f environment.yml && conda activate emg2force
pip install -e .                          # or: pip install -r requirements.txt

GPU recommended. **No bac

Related repositories

Similar projects matched by category, topics, and programming language.

lopopolo

Featured

harness-engineering

Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.

AI & Machine LearningAI Agents

2,390

slvDev

Featured

esp32-ai

A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.

AI & Machine LearningLarge Language Models

1,960

littledivy

Featured

mimic

mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.

AI & Machine Learning

1,482