InMoov Prosthetic Control – EEG-Driven Adaptive Learning System

This project integrates EEG-based movement classification, real-time prosthetic control, continuous learning through feedback, and vision-based grasp validation using ROS2.

---

Table of Contents

1. Overview
2. Project Structure
3. Getting Started
   • Install Dependencies
   • Train the Classifier
   • Real-Time EEG Prediction
   • Update Classifier with Feedback
4. Model Highlights
5. Integrated Movements
6. Vision Modules Explained
7. ROS2 Integration
8. Contributing
9. License

---

Project Structure

robot_classifier/
│
├── data/
│   └── motor_imagery/
│       └── raw/
│       └── preprocessed/
│
├── neuroscience/
│   ├── eeg_model/
│   │   ├── model.py
│   │   ├── train_classifier.py
│   │   ├── online_learning.py
│   │
│   ├── live_prediction/
│   │   └── predictor.py
│   │
│   ├── feedback/
│   │   └── correction_logger.py
│   │
│   ├── utils/
│   │   └── signal_processing.py
│   │
│   └── setup.sh
│
├── computer_vision/
│   ├── main.py
│   ├── grasp_validation.py
│   ├── grasp_identifier.py
│   ├── object_detection.py
│   ├── hand_landmarks.py
│   └── CONST.py
│
├── robotics/
│   ├── ROS2_node.py
│   └── arduino_servo_control.ino

---

Getting Started

1. Install Dependencies

cd neuroscience
bash setup.sh

2. Train the Classifier

Place your EEG training data in data/motor_imagery/preprocessed/ as X.npy and y.npy, then run:

python eeg_model/train_classifier.py

3. Predict Movements in Real-Time

python live_prediction/predictor.py

When a prediction is wrong, the system will ask for a correction, which is saved for later.

4. Improve the Classifier from Feedback

python -c "
import numpy as np
from feedback.correction_logger import CORRECTION_FILE
from eeg_model.online_learning import update_model_with_feedback
data = np.load(CORRECTION_FILE)
update_model_with_feedback(data['X'], data['y'])
"

---

Model Highlights

• EEGNetAdvanced
  • Depthwise separable convolutions
  • Spatial attention mechanism
  • Dropout, batch normalization
• Online Learning
  • Learns from feedback over time
  • Adapts to new user-specific EEG signals

---

Integrated Movements

• rest
• open_hand
• close_hand
• pinch_grip
• thumbs_up
• wave

These correspond to movements executed by ROS2 and the prosthetic arm.

---

Vision Modules Explained

The computer_vision/ folder includes:

• Real-time object and hand detection with MediaPipe and YOLO
• Grasp classification based on geometry + ML
• Stability and landmark-based grasp checking

---

ROS2 Integration

The robotics/ROS2_node.py reads EEG predictions, translates them into motion commands and sends them to:

• Arduino via serial (servo control)
• Other ROS2 components

---

Contributing

Feel free to fork and make improvements – especially if you’re improving classifier accuracy or adding new gesture support.

---

License

MIT License. See LICENSE file for details.