Dimitar Stefanov

1
Lecture 15

• Dimitar Stefanov

2
Multi multifunction control scheme for powered
upper-limb prostheses
Some examples
Pattern-coded systems which use a M- vectors of
unique values and a pattern classifier to produce
M-prosthetic functions.
3
Prosthetic hand (NTU-Hand) National Taiwan
University (NTU), Dept. of ME, Robotic Lab
http//robot0.me.ntu.edu.tw/english/E_Researches/m
edical/medical_res.htm http//robot0.me.ntu.edu.tw
/english/E_Researches/medical/prosthesis.htm

• Developed by Dr. Li-ren Lin and Ji-Da Wu
• A modular robotic hand
• The degrees of freedom of the hand range from
  five to eleven
• The weight of the NTU-Hand II - 1300 g

4
EMG signal processing
http//robot0.me.ntu.edu.tw/english/E_Researches/m
edical/EMG.htm
The EMG controller uses two pairs surface
electrodes to acquire the EMG signal from the
flexor digitorum superficialis muscle and the
extensor pollicis brevis muscle.
Eight types of hand movements (Three-jaw chuck,
lateral hand, hook grasp, power grasp,
cylindrical grasp, centralized grip, flattened
hand and finger flexion)
5
Used parameters variance, zero-crossings,
autoregressive model and spectral estimation. The
control method combined the pattern recognition
technique and a pulse-coding analysis. An error
backpropagation neural network and a
k-nearest-neighbor rule are applied to
discriminate among the feature sets. Experiments
PC digital signal processor (DSP) 3-D graphic
interface program.
6
17 degree of free in the five fingers force
sensors 8051 graphic shows that the hand grasps
the egg.
http//robotweb.me.ntu.edu.tw/English/E_Researches
/robotman/Image/5Finger.htm
7
Rutgers university (NJ)
http//uc.rutgers.edu/news/science/arthand.html
http//opl.rutgers.edu/opl.html http//opl.rutger
s.edu/opl.html
Rutgers University, Department of biomedical
engineering in Piscataway, N.J.
William Craelius, an associate professor of
biomedical engineering, PhD student Rochel Lieber
Abboudi.
Multi-finger control and proportional control of
force and velocity.
8
Three sensors inside the sleeve pick up natural
motions of tendons and transmit them to a desktop
computer which then control the fingers of the
hand.
lack of fingers

• Tendon Activated Pneumatic (TAP) control.
• Based on sensing the command signals in the
  forearm by pressure sensors that are located in
  the limb socket.
• The system detect specific finger movement
  requests and send them directly to small
  actuators that move fingers.
• Biomimetic approach the natural motor control
  system is used to activate the fingers (minimal
  learning time).
• the TAP system is not appropriate for everyone.
• tested on 12 people, 9 of whom were able to use
  it successfully.
• several operable fingers with controlled grasping
  force.

9
Universidad National Argentina
10
"A Microprocessor-Based Multifunction Myoelectric
Control System"by Kevin Englehart, Bernard
Hudgins, Philip Parker and Robert N. Scott,
Institute of Biomedical Engineering, University
of New Brunswick. 
23rd Canadian Medical and Biological Engineering
Society Conference "A Microprocessor-Based
Multifunction Myoelectric Control System,
Toronto, Canada, May 1997
11
"A Microprocessor-Based Multifunction Myoelectric
Control

• The control scheme uses the myoelectric signals
  produced in the first two hundred milliseconds
  following a contraction in the muscles. 
• This information is used to train a pattern
  classifier to recognize the specific pattern
  unique to the amputee, and to determine the
  intent of the amputee. 
• The pattern classifier matches the pattern to
  select the device that is controlled, such as the
  hand, elbow, or wrist. 
• Once a pattern is recognized, the actuator is
  activated until the input myoelectric signal goes
  below a certain threshold.

12

• The controller operates in two modes
• a training mode
• a normal mode.

The training mode or PC interface mode involves
the use of a host PC for offline processing.

• Steps of the training mode
• Specify myoelectric control parameters.
• Collect the MES for the user.
• Extract certain features from the MES, and store
  them for training the artificial neural network.
• Train the pattern classifier to recognize the
  input MES. This requires a host computer for
  off-line processing.
• Train the artificial neural network to obtain the
  weights. These weights are used to determine a
  match.
• Store these weights in the device in non-volatile
  RAM.

The PC interface mode allows the control of a
prosthesis by controlling a three-dimensional
"virtual arm" on the host PC.
13

• The normal mode
• MES is collected both from the biceps and from
  the triceps.
• Extraction of features from the signals
• Artificial neural network whose weightings were
  previously determined in the offline mode to
  identify the closest pattern match to drive the
  correct device.

• The system have to be fast enough to respond
  within 300 milliseconds. A system response longer
  than this would cause the user would become
  frustrated, and to try other motions.
• This response time is limiting because the system
  requires approximately 250 milliseconds to
  capture enough data in the MES for accurate
  pattern recognition. This leaves only 50
  milliseconds for the processing and activation of
  the prosthetic device.
• Even with these limitations, the current system
  has an over 90 accuracy in determining four
  types of prosthetic device motions.

The current prototype has an approximate size of
1.5" x 2.5" x 0.5" and operates on a 6V NiCad
battery.
14
Multifunctional prosthetic control using the
myoelectric power spectral density spectrum
Dr. Philip Parker, Jillian Mallory
The University of New Brunswick, Canada.
The system inputs the myoelectric signal, x(n),
at a sampling frequency F, then extracts a
feature vector of length N.
15
Multifunctional prosthetic control using the
myoelectric power spectral density spectrum
The system inputs the myoelectric signal, x(n),
at a sampling frequency F, then extracts a
feature vector of length N.
The myoelectric signal power spectral density
spectrum changes with variations in muscle
contraction patterns. The design and the
implementation of a control input to a
myoelectric control system can be based on the
classification of these power spectrum patterns.
A feature vector corresponding to the changes in
spectrum is extracted by the segmentation of the
spectrum. This vector is classified using a
pattern classifier and its output is used for the
prosthesis control.
16
Prosthetic Control by an EEG-based Brain-Computer
Interface
Prosthetic Control by an EEG-based Brain-Computer
Interface (BCI), Christoph Guger, Werner Harkam,
Carin Hertnaes, Gert Pfurtscheller, Uniniversity
of Technology Graz, AAATE99 http//www.fernuni-hag
en.de/FTB/aaate99/paper/99_90/99_90.htm

• It was shown recently that hand movement imagery
  results in EEG signals close to primary motor
  areas. An array of electrodes overlying motor and
  somatosensory areas (electrode positions C3 and
  C4).
• Oscillatory EEG components are used for BCI
• On-line analysis of EEG signals is required.

17
Test

1. Fixation cross was shown in the center of a
   monitor.
2. After two seconds a warning "beep" stimulus
3. From second 3 until 4.25 an arrow (cue stimulus),
   pointing to the left or right, was shown on the
   screen. The subject was instructed to imagine a
   left or right hand movement, depending on the
   direction of the arrow.
4. Between second 4.25 and 8 the EEG was classified
   on-line and the classification result was used to
   control the prosthesis.

If the person imagined a left movement, then the
prosthesis was closed a little bit more and vice
versa (correct classification assumed). One
session consisted of 160 trials. Three sessions
were made with subject i6.
18
BCI system allows to control a hand prosthesis by
imagination of left and right hand movement.
A practical EMG-based human-computer interface
for users with motor disabilities Armando B.
Barreto, PhD Scott D. Scargle, MSEE Malek
Adjouadi, PhD
Journal of Rehabilitation Research and
Development, Vol. 37 No. 1, January/February 2000

• Computer interaction (UP, DOWN, LEFT, RIGHT,
  left-click) from monitoring the activity of
  several pericranial muscles.
• Four electrodes are used

19
Intelligent prosthetic controller
DAISUKE NISHIKAWA, WENWEI YU, HIROSHI
YOKOI, YUKINORI KAKAZU Lab. of Autonomous Systems
Eng., Research Group of Complex Systems
Eng., Graduate School of Eng., Hokkaido
University.
The analyzing unit is based on the Wavelet
transform using Gabor mother wavelet function
into the analysis unit.
20

1. Pronation
2. Supination
3. Flexion
4. Extension
5. Grasp
6. Open

21
Development of Prosthetic Hand Using Adaptable
Control Method for Human Characteristics, Sadao
FUJII, Daisuke NISHIKAWA, Hiroshi YOKOI
Adaptation with Visual Feedback
22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
Adaptation
Two Functions
26
Three Functions