DIY Electromyography with MyoWare Sensor and Arduino

Biomechatronic systems integrate humans with machines including exoskeletons, prostheses, rehabilitation robots, and vehicles. The research uses linear graph theory and symbolic computing (Maple) to create integrated models across biological, mechanical, and electrical domains. Three approaches model human motor control: feedback/feedforward controllers, optimal control theory assuming cost function minimization, and Non-linear Model Predictive Control (NMPC). NMPC is particularly valuable for real-time applications as it captures human adaptation to changing tasks. For athletic performance, cost functions like 'faster, higher, stronger' successfully predicted optimal cycling techniques. For daily activities, cost functions include muscle effort, metabolic energy, and jerk minimization. The FDA has endorsed computer simulation to replace some clinical trials for device approval. In stroke rehabilitation, NMPC successfully predicted human arm adaptation to shifting targets. Force field control approaches include assistive fields (guiding users along desired trajectories) and resistive fields (pushing users off trajectories). Simulation results show patients who struggle against resistive fields perform better in subsequent trials, consistent with literature showing resistive training benefits motor recovery after stroke. Exoskeletons can provide active balance control during quiet standing without requiring crutches. For spinal cord injury subjects with limited torque capabilities, NMPC-based controllers successfully maintain balance when subjected to perturbations. Terrain recognition uses deep convolutional neural networks trained on chest-mounted camera images at five frames per second. With extensive data collection (over 900,000 images across 12 terrain classes), CNNs achieve high classification accuracy for automatic mode selection. For upper limb exoskeletons, EMG signals predict human intent because they fire before movement. Neural networks (CNNs, recurrent CNNs) map EMG to joint angles and torques, with CNNs performing better with raw data versus filtered data. The slowest network achieved less than 5% error in 1.4 milliseconds, sufficient for real-time control.

Welfare robots are categorized into public service, medical, and welfare robots for elderly and disabled individuals, with objectives including nursing care and lifestyle support through communication and daily activity support. Hand finger movement patterns necessary for daily life include grasping, pinching, power grip, finger opposition, and finger extension. The two main approaches to EMG signal processing for lower limb control are: (1) Supervised EMG control, which uses EMG signals to classify locomotion modes and switches to pre-programmed low-level autonomous control, reducing user cognition and muscular workload; (2) Direct EMG control, which uses EMG signals from antagonist muscle pairs to continuously and directly control prosthetic joint dynamics, mimicking biological neuromuscle control without predefined modes. Muscles work in pairs: agonist contracts to create movement while antagonist relaxes to allow it. Muscles can only produce contraction force, requiring both muscle pairs for bidirectional joint movement. Exoskeleton control systems require feedback mechanisms to detect and amplify user movements using sensors like accelerometers, gyro sensors, and EMG signals. The control system must be fast enough to respond to the frequency of human movements, especially for activities like running. Pneumatic actuators may have slower response times compared to electromagnetic actuators. Different exoskeleton models are designed for specific applications: the Forge for military and first responders, the Ascend for people with knee issues, and the Elevate for skiers.

Myoelectric prosthetic devices use electromyography (EMG) signals to detect muscle activity and control prosthetic movement; the process involves recording electrical signals from muscles using surface electrodes, applying signal processing techniques including filtration, rectification, and amplification to extract meaningful data, determining a threshold value to distinguish between muscle states (such as hand open versus closed), and using this processed signal to control servo motors that drive the prosthetic mechanism, demonstrating how mechatronic systems integrate biomechanical sensing with electronic control for functional prosthetic devices.

Building a functional exoskeleton requires integrating multiple subsystems: pneumatic muscles, valves, controllers, and structural frames. The frame uses mild steel for simplicity and ease of machining, joined by TIG welding with bushings for smooth articulation. 3D printed bracers with steel inlays balance comfort with strength. Three-eighths inch polyurethane tubing maximizes airflow efficiency. Control systems use microcontrollers driving solenoid valves (18.7 GPM flow rate), with pressure sensors and load cells providing closed-loop feedback. This integration demonstrates how mechanical, electrical, and software systems must work together for complex mechatronic devices.

Robotic exoskeletons for assisting disabled individuals use EMG signals for control. The system consists of four connections (two for shoulder, two for elbow) with 3D force sensors measuring wrist force. Eight electrodes measure EMG signals from triceps (elbow) and deltoid, pectoralis major, and trapezius (shoulder). Signals are amplified and processed to extract meaningful information. The exoskeleton uses cables and pulleys connected to motors, with potentiometers providing joint angle feedback. Movement ranges are set to 0-120° for elbow and 0-190° for shoulder based on physiological limits. The neuro-fuzzy control system consists of three stages: input signal selection (choosing between wrist force and EMG signals based on activity levels), posture position selection (determining control parameters based on joint positions), and fuzzy control (applying rules to generate outputs). Membership functions convert inputs into fuzzy sets, and the system adapts control rules based on arm posture.

An EMG (Electromyography) controller translates electrical signals from muscle contractions into computer commands, enabling users to control video game actions like mining in Minecraft by contracting specific muscles; this technology demonstrates how electrodes detect the body's electrical signals and connects human physiology with computer systems through adjustable sensitivity settings.

Human-Computer Interaction research includes using muscle sensors to detect hand configurations. By placing electrodes on the forearm that sense muscle contractions, signal processing algorithms can determine the configuration of the hand from sensor signals. This enables applications like playing Guitar Hero with an air guitar, where the system interprets muscle movements as intended actions.

Electromyography (EMG) measures electrical activity generated during muscular contractions using surface electrodes. This physiological signal can be recorded, processed, and passed to machine learning algorithms which can discriminate between different muscle contractions. This technology effectively turns the human body into a controller where muscular contractions or gestures are associated with interactive commands, enabling hands-free interaction with computing systems.

This video demonstrates how muscle contractions can be detected and translated into digital control signals through electromyographic sensors, enabling users to interact with computer systems by intentionally contracting specific muscles to trigger actions such as game controls or device functions.

Myoelectric signal detection involves reading electrical signals from approximately 15,000 neurons that innervate about 14 muscles in the human hand. These signals can be detected from the surface of the skin and mapped to control technological devices. This technology allows users to control devices through subconscious muscle intentions rather than physical inputs like keyboards or switches, effectively shortcutting the learning process required for traditional human-computer interfaces.

EMG signals show distinct patterns: when muscles are relaxed, there is background noise; when muscles flex (like making a fist), alternating patterns emerge with higher amplitude. Different gestures activate different muscle groups, producing unique signal patterns across the sensor channels. This variation allows machine learning algorithms to classify different movements based on their characteristic signal signatures.

One approach to improving prosthetic control is using multiple EMG electrodes (4, 6, 8, up to 64 channels) to cover the entire forearm. This allows recognition of individual finger movements. However, this approach faces challenges with generalization, as different algorithms may be needed for each user.

This research demonstrates a supervised machine learning approach for hand gesture recognition using EMG signals, achieving 98.9% accuracy through a pipeline involving signal preprocessing (normalization, empirical mode decomposition), spectral domain feature extraction, and classification using a cubic Support Vector Machine (SVM) classifier trained on four hand gestures (resting hand, clenched fist, wrist flexion, wrist extension).

Advanced machine learning algorithms can classify multiple distinct hand movements from EMG signals recorded from muscle grafts. Using a hidden Markov model classifier, researchers achieved offline classification accuracy of 98-97% for 10 different movements including finger abduction, adduction, and wrist movements. Online performance during real-time prosthetic control reached approximately 92-86% accuracy. The latency for these classifications ranged from 160 milliseconds to 297 milliseconds depending on the participant.

The EMG (electromyography) wristband measures electrical signals sent from the brain through tendons to fingers. Each distinct hand gesture produces unique electrical impulse patterns that can be detected by sensors in the textile weave. The onboard machine learning computer processes these signals and maps them to control inputs, achieving approximately 80% accuracy in prototype versions with haptic feedback confirmation.