Thesis Presentation

							SDSU


       Real-Time Control of a Multi-Fingered
         Robot Hand Using EMG Signals

                     Master’s Thesis
                          By
                   Luenin Barrios

               Supervisor: Marko Vuskovic

             Department of Computer Science
               San Diego State University
                     June 29, 2010
SDSU
                       Outline
    Introduction to Research
    Multi-Fingered Robot Hands and Prostheses
    Measurement of EMG Signals
    Feature Extraction and Classification
    Synergy and Robot Control System
    Hardware Description
    Implementation
    Observations and Results
    Summary
SDSU
                                 Introduction
    Goal of Research:
     To implement a program that uses the EMG
     Classifier output to control the grasp motions of
     the SDSU robot hand in real-time

    Grasp modes:




       Chris Miller Master’s Thesis 2008.
SDSU
          Prosthetic Hands Overview
    Early Models
    Restrictions and Limitations
    Degrees of Freedom
    EMG Signal Control




Otto Bock Grasp Pincher   TAP Version 3 Prototype   SDSU Robot Hand
SDSU

                       Overall Schematic

                                              Signal
                                            Processing

                                            Time Sample
              EMG              A/D           Extraction                   Prosthetic
            Amplification    Converter                       Classifier     Hand
              Device                          Feature                     Controller
                                             Extraction
Saksit Siriprayoonsak 2005                                                 This Project
                                           Transformation



                                         Chris Miller 2008
SDSU
                  EMG Signals
    Electromyography
    EMG potentials: 50 μV and up to 20 to 30 mV




                              Source:www.univie.ac.at/cga/courses/be522/emg/fiber.gif
SDSU
           Forearm Muscle Anatomy




Chris Miller Master’s Thesis 2008
          EMG Amplifier Device
   Saksit Siriprayoonsak 2005
    4 Bipolar Channels
    1 Reference Channel
    Surface Electrodes
SDSU

   EMG Amplifier Device Con’td
SDSU
           EMG Classifier Program
             Signal Detection
    Bonato Method
    Onset of Movement
SDSU
                   Classifier
               Signal Processing
    Feature Extraction
     Methods:
        Waveform Length (Farry et al., 1996)
        Spectral Moments (Vuskovic et al., 2005)
SDSU
         EMG Signal Processing
       Feature Extraction Method 1
  Waveform Length
SDSU
           EMG Signal Processing
         Feature Extraction Method 2
    Spectral Moments




   I-coefficients
SDSU
            Feature Classification
      Mahalanobis Distance(Mahalanobis, 1936)


     Sample Feature Vector Space
SDSU
       Feature Log Transformation
    Box Cox Transformation (1964)
SDSU
       Robot Joint Control System
    PID Controller and Actuator
SDSU
            Joint Control System
   Acutuator Model

                        Variable            Description            Unit

                        Ra         Terminal or Armature        3.38 Ohm
                                        Resistance

                        Ka         Torque Constant             8.11 mNm/A

                        Jm         Rotor Inertia               1.27 gcm2

                        Kg         Gear Transmission Ratio -   1:26
                                        Thumb                  1:19
                                   Gear Transmission Ratio -
                                        Finger

                        Ga         Driver Gain                 1

                        Kb         Speed or Proportionality    1180 rpm/V
                                        Constant

                        V0         Nominal Voltage             12 Volt

                        ω0         No Load Speed               13900 rpm
SDSU
                 Joint Control System
     PID Controller




     e = qmd - qm; // Get controller error
     qmdot = (qm-qmold)/_Ts; // Get derivative of error
     ei = eiold + e * _Ts; // Get integral of error
     u = _Kp*e - _Kv*qmdot + _Ki*ei; // Control law
     qmold = qm;
     eiold = ei;
SDSU
                 Synergetic Motion
     Synergetic Mapping
      θj = fj (m, D) where j = 0, 1…5 and m = 1…4

     Approximation Function (Vuskovic and Marjanski)




         am,j = γm,j                      cm,j = αm,j
SDSU
           Synergetic Training
  Joint Angle


                       θ1 (cm,1 + D1) = am,1 bm,1 - am,1 D1
Object Shapes and Sizes
Spherical          Point




Cylindrical       Lateral
SDSU
            Calibration and Training
        Sample Training for Point Objects:




       Sample Positions for Lateral, Cylindrical and Spherical
SDSU
               Robot Hardware
   Servo To Go Board
   Signal Transition Box
SDSU

       Servo To Go Interface Board
   Encoder Input A/B signal
   Analog Input/Output
SDSU
          Signal Transition Box
 Central hub for signals/cables
 Relays information
                                   Example: Joint 0
                                   P3 DAC         2
                                       EnIn A 14
                                       EnIn B 17



                                   DB50
                                      EnOut A 35
                                      EnOut B 34

                                   DB25
                                      AnalogIn    2
SDSU
                    EMG Robot Hand
    User Interface/Motion Command Interpreter
    Client/Server
    TCP/IP
    Real-time EMG/User Commands
    Grasp modes: Cylindrical, Spherical, Point, Lateral
SDSU
        EMG Robot Hand Cont’d
 Examples:
        Command: g 0 45




        Command: o 3 9
SDSU

       Overall Runtime Flow Chart
SDSU
  Overall Runtime Flow Chart Cont’d
SDSU
                   System Execution
          Step 2                Step 3




       Step 1
                                Step 4
SDSU
                Summary
 Multi-fingered Robot Hands and EMG Signals
 Collection of EMG Signals
 Feature Extraction and Classification
 PID Controller
 Synergetic Motion
 Overall System Diagram and Transition Box
 Real-time control of Robot Hand using EMG
  Signals
       Conclusions/Future Work
   Feasibility of EMG Signal for Hand Control
   Synergetic Grasp Motions
   Classifier for real-time control
   Combine projects so they reside on same
    machine
   Improve arm/amplifier device contact
   Wireless electrodes/sensory network
   Improve time delays in Classifier
SDSU




       Questions/Comments?