Bridging Neural Intent and Rehabilitation Robotics

This project develops a comprehensive EMG-controlled rehabilitation system that translates muscle electrical activity into precise control of both a robotic hand and wheelchair robot. Moving beyond traditional machine learning approaches, the system implements a robust threshold-based classification with IMU integration for intuitive, context-aware control essential for rehabilitation applications.

The Rehabilitation Technology Challenge

EMG-based control systems for rehabilitation face unique constraints—they must be reliable, fast-responding, and intuitive for patients with varying abilities. Traditional machine learning approaches, while sophisticated, often struggle with:

Inconsistent response times (3-10 seconds) unsuitable for real-time control
Generalization problems across different users and sessions
Complex calibration requiring extensive training data
Limited contextual awareness for intuitive operation

Technical Innovation: Threshold-Based Classification with IMU Context

System Architecture Redesign

From ML to Robust Control: After extensive collaborative exploration of machine learning approaches (Random Forest achieving 84% accuracy but with significant latency), I redesigned the system using a threshold-based classification approach that prioritizes reliability and real-time performance.

Core Implementation in DataManager.py:

def process_emg_with_imu(emg_data, imu_data):
    # Dynamic amplitude thresholding
    threshold = baseline_amplitude * threshold_multiplier
    
    # Gesture detection with state management
    if signal_amplitude > threshold and not in_cooldown:
        gesture = classify_gesture(emg_pattern)
        orientation = get_palm_orientation(imu_data)
        
        # Context-aware command generation
        command = interpret_gesture_with_context(gesture, orientation)
        return command

Advanced Signal Processing Pipeline

Adaptive Filtering System:

Butterworth Bandpass Filter: 10-150 Hz preserving EMG signal content
Notch Filter: 50 Hz power line interference removal
Dynamic Thresholding: Configurable multiplier parameters for individual adaptation
State Management: Cooldown periods preventing gesture repetition artifacts

IMU-Enhanced Context Awareness: The breakthrough innovation integrates built-in accelerometer data from Delsys sensors to detect palm orientation, enabling contextual gesture interpretation:

Fist Gesture + Palm Down = Wheelchair Forward
Fist Gesture + Palm Up = Wheelchair Backward
Same gesture, different context = Intuitive bidirectional control

Dual-Protocol Communication Architecture

Serial Control for InMoov Robotic Hand

Protocol Implementation (serial.ino):

// Command syntax: <finger>:<angle>
void processCommand(String command) {
    int delimiterIndex = command.indexOf(':');
    String finger = command.substring(0, delimiterIndex);
    int angle = command.substring(delimiterIndex + 1).toInt();
    
    moveServo(finger, angle);
    sendAcknowledgment();
}

Performance Characteristics:

Response Time: <1 second for finger movements
Error Detection: Built-in acknowledgment and retry mechanisms
Precision Control: Individual finger angle control with servo feedback

WiFi Control for Wheelchair Robot

RESTful API Implementation (wifi-car.ino):

server.on("/move", HTTP_POST, []() {
    String direction = server.arg("direction");
    String context = server.arg("palm_orientation");
    
    executeMovement(direction, context);
    server.send(200, "application/json", "{\"status\":\"success\"}");
});

Advanced Features:

Endpoint-Based Control: Dedicated REST endpoints for each movement type
Context Integration: IMU orientation data influences movement interpretation
Error Recovery: Automatic reconnection and command queuing
Safety Mechanisms: Emergency stop and timeout protections

User Interface and Experience Design

Automated Calibration System

Guided Setup Process (CollectDataWindow.py):

Gesture Sequence: User performs standardized movements
Automatic Threshold Calculation: System computes optimal detection parameters
IMU Reference Mapping: Establishes palm orientation baselines
Validation Testing: Real-time verification of calibration accuracy

Impact on Usability:

Setup Time: Reduced from hours to <30 minutes
User Confidence: Visual feedback during calibration process
Adaptation: Automatic adjustment for individual muscle response patterns
Accessibility: Simplified interface suitable for rehabilitation patients

Seamless Mode Switching

Unified Control Interface: The system provides intuitive switching between controlling the InMoov hand via serial and the wheelchair via WiFi through a single interface, automatically adapting gesture interpretations based on the selected mode and palm orientation context.

Real-Time Feedback System:

Visual Indicators: Clear display of current control mode and orientation
Status Monitoring: Connection health and system responsiveness
Error Reporting: User-friendly error messages and recovery suggestions
Performance Metrics: Real-time latency and accuracy monitoring

Technical Performance Achievements

Classification Accuracy and Speed

Gesture Recognition Performance:

5 Distinct Gestures: Reliably detected with contextual interpretation
False Positive Rate: <5% through dynamic thresholding
Response Time: <500ms for gesture classification
Consistency: Maintained accuracy across 30+ minute sessions

Context-Aware Control:

Palm Orientation Detection: >95% accuracy using IMU integration
Bidirectional Control: Single gesture enables forward/backward movement
Intuitive Operation: Natural hand movements translate to expected robot actions
Reduced Cognitive Load: Simplified control scheme for rehabilitation patients

Communication System Reliability

Performance Metrics:

InMoov Hand Control: <1 second latency for finger movements
Wheelchair Control: <5 second latency for navigation commands
Connection Stability: >99% uptime during testing sessions
Error Recovery: Automatic reconnection within 2 seconds

Robustness Features:

Protocol Redundancy: Fallback mechanisms for communication failures
Command Validation: Input sanitization and range checking
Safety Interlocks: Emergency stop capabilities across both platforms
Session Management: Persistent connections with automatic recovery

Collaborative Development and Project Management

Technical Collaboration

Machine Learning Investigation: Worked extensively with team member Danisha to evaluate Random Forest and other ML approaches, providing crucial performance data that informed the decision to pivot to threshold-based classification.

Hardware Integration: Collaborated with Gan on wheelchair control system integration, ensuring directional commands mapped correctly to physical movements with appropriate contextual responses.

Mechanical Optimization: Partnered with Phoebe to address InMoov arm mechanical issues, adjusting tendon tensions and calibrating servo ranges for accurate finger replication.

Project Infrastructure

Development Environment:

GitLab Repository: Structured branch management for feature development
Continuous Integration: Automated testing for signal processing algorithms
Version Control: Performance metric tracking across implementations
Code Review: Collaborative quality assurance processes

Documentation and Standards:

Technical Specifications: Comprehensive API and protocol documentation
User Guides: Step-by-step setup and operation instructions
Performance Benchmarks: Standardized testing protocols and metrics
Safety Protocols: Emergency procedures and risk mitigation strategies

Clinical and Rehabilitation Impact

Real-World Application Validation

Rehabilitation Requirements:

Response Time: Critical for maintaining patient engagement and natural interaction
Reliability: Essential for building user confidence and therapeutic effectiveness
Simplicity: Reduced cognitive load important for patients with varying abilities
Safety: Emergency controls and fail-safe mechanisms protect users

User Experience Outcomes:

Setup Efficiency: 30-minute calibration enables practical clinical deployment
Learning Curve: Intuitive control scheme reduces training requirements
Session Duration: System maintains accuracy over extended rehabilitation sessions
Patient Feedback: High acceptance rates for natural control paradigm

Technology Transfer Potential

Clinical Integration:

EMG Therapy: Biofeedback applications for muscle retraining
Prosthetic Control: Principles applicable to commercial prosthetic devices
Assistive Technology: Wheelchair and mobility aid integration
Rehabilitation Monitoring: Objective measurement of motor function progress

Scalability Considerations:

Multi-User Support: System architecture supports multiple patient profiles
Remote Monitoring: Data collection enables progress tracking and analysis
Customization: Adaptable parameters for different rehabilitation protocols
Cost Effectiveness: Open-source approach reduces deployment barriers

Future Development Pathways

Technical Enhancements

Advanced Sensor Fusion: Integration of additional biomedical signals (EEG, EMG arrays)
Machine Learning Augmentation: Hybrid approaches combining threshold and ML methods
Wireless Integration: Elimination of remaining wired connections
Cloud Analytics: Population-wide learning and optimization

Clinical Applications

Stroke Rehabilitation: Specialized protocols for motor function recovery
Spinal Cord Injury: Assistive technology for independence enhancement
Neuromuscular Disorders: Adaptive control for progressive conditions
Pediatric Applications: Age-appropriate interfaces for developmental therapy

Technical Skills and Innovation

This project demonstrates expertise in:

Biomedical Signal Processing: Real-time EMG analysis and noise reduction
Embedded Systems: Arduino programming and hardware integration
Human-Computer Interaction: Intuitive interface design for medical applications
Systems Architecture: Multi-protocol communication and integration
Clinical Engineering: Medical device development and safety protocols
Project Management: Collaborative development and version control

The successful pivot from machine learning to threshold-based classification, enhanced with IMU context awareness, represents innovative problem-solving that prioritizes practical deployment over theoretical sophistication.

This EMG-controlled rehabilitation system demonstrates how engineering innovation can directly impact patient care, providing reliable, intuitive technology that supports therapeutic goals while maintaining the safety and usability standards essential for clinical applications.

EMG-Controlled Rehabilitation System with Dual-Robot Integration

Tech Stack