Muscle-Driven Robotic Control for Human-Robot Interaction

MSc assignment

Background

Human-robot interaction (HRI) is essential for assistive robotics, rehabilitation, and interactive systems. Traditional approaches often rely on vision-based tracking to detect user intent, but these methods can be limited by occlusions, lighting conditions, and privacy concerns. Wearable sensors (like surface electromyography (sEMG) offers an alternative by leveraging muscle activity to infer user intentions, enabling intuitive, non-vision-based robotic control.

This thesis focuses on developing a muscle-driven robotic control system that enables seamless interaction between a human and a robotic arm. As a test case, the system will be designed to execute a dynamic interaction task (e.g., a high-five, fist bump) based solely on sEMG signals. A vision-based method will also be implemented as a benchmark for comparison, helping to evaluate the effectiveness of the sEMG approach.

Objective

The goal of this thesis is to develop a muscle-driven robotic control system for intuitive human-robot interaction. Using sEMG signals, the system should detect user intent, classify movements, and control a robotic arm in real-time. A vision-based method will serve as a benchmark for evaluating system performance.

Key Tasks

  1. Literature Review
    • Analyze current vision-based approaches as a benchmark.
    • Study existing sEMG-based gesture recognition techniques.
    • Review robotic control strategies for human-robot interaction.
  2. sEMG Signal Processing & Classification
    • Acquire and preprocess sEMG data to detect movement patterns.
    • Develop deep learning models for gesture classification and recognition and motion trajectory estimation.
  3. Robotic Arm Control & Interaction Design
    • Integrate sEMG-based intent recognition with robotic arm control.
    • Implement real-time response mechanisms for interactive tasks, deployed AI network.
  4. Benchmarking with Vision-Based Approach
    • Implement a vision-based tracking system (e.g., OpenPose, MediaPipe) for intent recognition.
    • Compare the accuracy, latency, and robustness of the sEMG and vision-based methods.
  5. System Testing & Evaluation
    • Conduct real-world experiments with users interacting with the robotic arm.
    • Evaluate system performance in terms of accuracy, responsiveness, and user experience.
  6. Documentation
    • Provide a detailed thesis covering methodology, implementation, experimental results, and insights.

Expected Outcomes

  • A working muscle-driven robotic control system enabling real-time human-robot interaction.
  • A comparative study of sEMG-based vs. vision-based intent recognition.
  • Insights into the feasibility and limitations of muscle-driven robotic control in interactive applications.

Requirements

  • Background in sEMG signal processing, robotics, and machine learning.
  • Experience with robot arm control (e.g., ROS, Python, C++).
  • Interest in human-robot interaction and real-time control systems.

This project offers a unique opportunity to explore biologically inspired robotic control, advancing the future of intuitive, camera-free human-robot collaboration.