1

Active Tremor Compensation via Robotic Handheld Instrument and Wearable Orthosis

Wei Tech ANGAssistant Professor

School of Mechanical and Aerospace EngineeringNanyang Technological University, Singapore

wtang@ntu.edu.sg

2

Active Noise/Disturbance Compensation

Low signal-to-noise ratio applicationsPhysiological tremor – micromanipulation Pathological tremor – activities of daily living

External Disturbance

System

Intended Motion

Filtering

Disturbance + Intended motion

Motion Sensing

ControllerDisturbance

Manipulator

Disturbance Compensation

One sampling cycle

3

Active Pathological Tremor Compensation in Wearable Orthosis

Nanyang Technological UniversityDingguo ZHANG Nanyang Technological UniversityFerdinan WIDJAJANanyang Technological UniversityKalyana VELUVOLU Tan Tock Seng HospitalAdela TOWNational Neuroscience InstituteLouis TANNanyang Technological UniversityCheng Yap SHEE Technische Universitaet MuenchenMarkus RANK (Exch Student)LIRMM, U of Montpellier IIPhilippe POIGNETNational Neuroscience InstituteWing Lok AU

4

Active Pathological Tremor Compensation in Wearable Orthosis

5

Active Physiological Tremor Compensation in Handheld Microsurgical Instrument

Nanyang Technological UniversityMingli HAN

Nanyang Technological UniversityTun Latt WIN Nanyang Technological UniversityKalyana VELUVOLU Nanyang Technological UniversityU-Xuan TAN Nanyang Technological UniversityCheng Yap SHEE Carnegie Mellon UniversityCameron RIVIERE Singapore General HospitalYee Siang ONG National University HospitalThiam Chye LIM

Carnegie Mellon UniversityDavid CHOI

6

Microsurgery with Active Handheld Instrument

Visuomotor Control System

Noisy, Tremulous Motion

Motion Sensing

Visual Feedback

Micron

Microsurgery

Estimation of erroneous motion

Tip manipulation for active error compensation

Task Definition

Problem Analysis

Engineering Solutions

Technical Details

7

Active Physiological Tremor Compensation in Handheld Microsurgical Instrument

8

Involuntary Hand Movement for Healthy People

Physiological Tremor

Roughly sinusoidal, ≤50 µm rms, 8-12 Hz

Others: Myoclonicjerk, drift

9

Physiological Tremor and Microsurgery

Complicates microsurgical procedures and makes certain delicate interventions impossible

10

Vitreoretinal Microsurgery

Removal of membranes ≤ 20 µm thick from front or back of retina

11

Vitreoretinal Microsurgery

Injection of anticoagulant using intraocular cannulation to treat retinal vein (~∅100 µm) occlusion

12

Physiological Tremor and Microsurgery

13

Physiological Tremor and Microsurgery

Impact on microsurgeons2 of 10 surgeons become microsurgeons

Factors affecting tremorFatigue – strenuous exercise etc.Caffeine/alcohol consumption (withdrawal syndrome)Lack of practice – long vacation etc.Age – experience vs hand stability

Microsurgeons’ consensus:10 µm positioning accuracy

14

Microsurgery with Active Handheld Instrument

Visuomotor Control System

Noisy, Tremulous Motion

Motion Sensing

Visual Feedback

Micron

VitreoretinalMicrosurgery

Estimation of erroneous motion

Manipulation of tip for active error compensation

15

Comparison of Robotic Solutions

Telerobotics‘Steady Hand’ robot Active Handheld Instrument

CheapUnobtrusiveDexterityLimited workspaceNo motion scalingNo ‘third hand’

Obtrusive

Unobtrusive< US$15K

> US$150K> US$1M

Da Vinci Steady Hand robot, Johns Hopkins Univ,

16

Micron Current Prototype

Length: 150 mm long Diameter: Ø20(16) mmWeight <100 g6 DOF inertial at the back end3 DOF piezoelectric driven parallel manipulator at front end with disposable surgical needleSignal processing and control performed by PC via ADC & DAC

150 mm (w/o needle)

Ø20 mm

Ø16 mm

Manipulator System

Sensing System

Disposable surgical needle

17

Microsurgery with Active Handheld Instrument

Visuomotor Control System

Noisy, Tremulous Motion

Motion Sensing

Visual Feedback

Micron

VitreoretinalMicrosurgery

Estimation of erroneous motion

Manipulation of tip for active error compensation

• Resolution

• Accuracy

18

On-board Sensing System

All-accelerometer inertial measurement unit (IMU):

3 dual-axis miniature MEMS accelerometers Analog Devices ADXL-203: 5mm x 5mm x 2mm, < 1g

Housed in 2 locations

Back Sensor Suite

Sensing direction

Front Sensor Suite

ZB

XB

YB

Dual-axis accelerometer

Dual-axis accelerometers

Manipulator System

19

Differential Sensing KinematicsBody acceleration sensed by accelerometer at location i:

Differential Sensing

Solve system of nonlinear equations for Ω = [ωx ωy ωz]T by Gauss-Newton or Levenberg-Marquart method

onsAccelerati inducedRotation−

×+×Ω×Ω++= BiBiCGi PPgAA α

Xw

P23

P13

P12

P3

P2P1

Yw

Zw

W

B 2

3

1

Y2

Z2

( )

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡••

=⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

•

•=

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

••=

=×+×Ω×Ω=−=

z

y

x

ijijij

aAaA

aA

jiPAAA

13

122323

13

13 ,,

3 2, 1,,,][]][[ α

20

Sensing Kinematics

Updating quaternions:

Directional Cosines matrix

Gravity Removal: WAE = WCBBA – Wg

Tip Displacement:

⎥⎦

⎤⎢⎣

⎡Ω−

Ω×Ω=ΩΩ=

×

××

0][

21~ ),()(~)(

31

1333Ttqttq

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

+−−+−−−+−++−−−+

=23

22

21

2010322031

103223

22

21

203021

2031302123

22

21

20

)(2)(2)(2)(2)(2)(2

qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq

CBW

tipBB

BW

t

TtE

Wtip

Wtip

W PtCddATtPtP ])[()()()( ×Ω++−= ∫ ∫−

τττ

21

Sensing Resolution (Error Variance) Analysis

Sensing resolution dependent on sensor noise floorAngular Sensing

Sensing equation:Aij = f(Ω)= ([Ω×] [Ω×] + [α×])Pij

Covariance:C(Aij) = C(Ω) Pij

Pij↑, C(Ω)↓

Pij

σAx2

σAy2

σAy2

σAx2

σωAx2 orσωAy

2

Back sensor suite

Front sensor suite

22

Proposed All-Accelerometer vs Conventional Inertial Measurement Unit

All-accelerometer IMUMaximized Pij, with physical constraint of a slender handheld instrument

Conventional IMU (3A-3G)Tokin CG-L43D rate gyros x 3

96.9% / 32x4.42 × 10-21.41ωz

99.3% / 130x1.08 × 10-21.41ωx & ωy

Noise reduction / resolution

improvement

6A Error std. dev.

(deg/s)

3G-3AError std. dev.

(deg/s)

23

Angular Sensing Resolution Comparison

Small angular velocity & sensor noise floor

0 50 100 150 200 250 300 350 400 450 500-2

0

2

4

6

8

0 50 100 150 200 250 300 350 400 450 500-2

0

2

4

6

8

0 50 100 150 200 250 300 350 400 450 500-2

0

2

4

6

8

ωz

(deg

/s)

ωx, ω

y(d

eg/s

)

ωG

(deg

/s)

Time (ms)Time (ms)

Tokin Gyroscope

All-accelerometer IMU

24

Sensing Resolution (Error Variance) Analysis

Translational Sensing2 accelerometers in each sensing direction:

Sensing resolution improves by a factor of

Better orientation estimation → more complete removal of gravity → better translation estimation

2111222

AiA

AjAiA

σ=σ→

σ+

σ=

σ2

25

Integration Drift of Inertial Sensors

Integration driftErroneous DC Offset

Ramp QuadraticError accumulates and grows unbounded over time

Poor sensing accuracy

∫∫0 0.2 0.4 0.6 0.8 1

-1

-0.5

0

0.5

1Acc

eler

atio

n (m

m/s

2 )

0 0.2 0.4 0.6 0.8 10

0.02

0.04

0.06

Vel

ocity

(mm

/s)

0 0.2 0.4 0.6 0.8 1

0

0.01

0.02

0.03

Time (s)D

ispl

acem

ent (

mm

)

Mean = 0.05 mm/s2

26

Measurement Model

Measurement model allows error analysis and compensationMeasurement Model = Physical (Deterministic) Model + Stochastic Model

Measurement Error

Measurement Model• Physical

• Stochastic

Compensation

27

Experimental Observations

-1 -0.5 0 0.5 11.5

2

2.5

3

3.5

Acceleration (g)

Acc

eler

omet

er O

utpu

t (V

)

A

B+

+

−

−

CW: A+ → B-

CCW: B+ → A-

g

-1 -0.5 0 0.5 11.5

2

2.5

3

3.5

Acceleration (g)

Acc

eler

omet

er O

utpu

t (V

)

α±150

α±90

g

α = 90°, β = -180° to 180°

α±30

α = 30° & 150°, β = -180° to 180°

28

Phenomenological ModelingBias, Bx(Vy, Vz) = Bx+gx(Vy)+hx(Vz)Scale Factor,SFx(Vz) = rx2Vz

2 + rx1Vz + rx0

Model

Ax = (Vx− Bx(Vy, Vz)) / SFx(Vz)

-1 -0.5 0 0.5 11.06

1.07

1.08

1.09

1.1

1.11

z-Acceleration (g)

Sca

le F

acto

r (V

/g)

1.5 2 2.5 3 3.5 4-0.03

-0.02

-0.01

0

0.01

0.02

0.03

0.04

y-Accelerometer Output, Vy (V)

Res

idue

Rxy (V

)

-1 -0.5 0 0.5 1-5

0

5

10

15x 10-3

z-Acceleration (g)

Res

idue

Rxz (V

)

gx(Vy) = px2Vy2 + px1Vy + px0

hx(Vz) = qx2Vz2 + qx1Vz + qx0SFx(Vz) = rx2Vz

2 + rx1Vz + rx0

In plane cross- axis effect

Out of plane cross- axis effectScale Factor

290 0.2 0.4 0.6 0.8 1-500

-400

-300

-200

-100

0

100

200

Time (s)

Acc

eler

atio

n (m

m/s

2 )Sensing Results - Translation

0 0.2 0.4 0.6 0.8 1-150

-100

-50

0

50

100

150

200

Time (s)

Acc

eler

atio

n (m

m/s

2 )

--89.7Error Reduction (%)

<1<531*Proposed Physical Model

6272300Linear Model

Scale Factor (mm/s2)Bias (mm/s2)Rmse (mm/s2)

Linear Model Proposed Physical Model

* ADXL-203 rated rms noise = 22.1 mm/s2

Interferometer

Accelerometer

30

Residual Zero Offset - Integration Drift

0 0.2 0.4 0.6 0.8 1

-0.1

-0.05

0

0.05

0.1

0.15

Time (Sec)

0 0.2 0.4 0.6 0.8 1-2

0

2

4

6

8

10

12

14x 10-3

Time (Sec)

0 0.2 0.4 0.6 0.8 10

1

2

3

4

5

6

7x 10-3

Time (Sec)

Acceleration

Position

Velocity

Numerical Integration

Zero Offset

31

Analytical Integration

0( )

0 00

sin(2 ( ) ) cos(2 ( ) )L f f G

k r rr

r ry a f k b f kG G

π π= −

=

= + + +∑

0 02 20 0 0

sin(2 ( ) ) cos(2 ( ) )(2 ( )) (2 ( ))

Lr r

k r rr G G

a br ry f k f kf G f G

π ππ π=

⎡ ⎤= − + + +⎢ ⎥+ +⎣ ⎦

∫ ∫ ∑

Acceleration:

Velocity:

Position:

0 00 0 0

cos(2 ( ) ) sin(2 ( ) )(2 ( )) (2 ( ))

Lr r

k r rr G G

a br ry f k f kf G f G

π ππ π=

⎡ ⎤= − + − +⎢ ⎥+ +⎣ ⎦

∫ ∑

Real-time modeling of Tremor

32

Analytical Integration

0 0.2 0.4 0.6 0.8 1-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

Time (Sec)0 0.2 0.4 0.6 0.8 1

-0.1

-0.05

0

0.05

0.1

0.15

Time (Sec)Acceleration

0 0.2 0.4 0.6 0.8 1-8

-6

-4

-2

0

2

4

6x 10-5

Time (Sec)

Estimation with BMFLC

Analytical Integration

Velocity

Position without drifting

Zero Offset

33

Vision Aided Inertial Sensing

Sensor Fusion (drift

correction)

High speed camera

Microscope

Micron

Inertial Sensors

Pose Estimate

34

Vision Aided Inertial Sensing

Instrument tip

Pan angle

Tilt angle

Gravity sensed by the accelerometers is used to obtain tilt angle of the instrument

Microscope with Camera

8x Magnification:• Tip Position Error = 2.13 µm rms• Tip Pan Angle Error= 0.2° rms.

35

Sensor Fusion

Bias

+

_Scale Factor

Local Gravity

Misalign-ment

Correction

Location & Orientation

Error

Augmented State Kalman

Filtering

Acc. Trend

Acceleration Random

Walk1/s

Velocity Random Walk

+

+

Acc.

Cam.

Noise removal

Physical Model

Edge Detection

Hough Transform

Tilt

KalmanFiltering

(Quaternion)

Kinematics

Tool tip displacement

Analytical Integration

Dilation +

Erosion

MotionDispl.

Pan

Displ.

36

Microsurgery with Active Handheld Instrument

Visuomotor Control System

Noisy, Tremulous Motion

Motion Sensing

Visual Feedback

MICRON

VitreoretinalMicrosurgery

Estimation of erroneous motion

Manipulation of tip for active error compensation

• Real-time

• Phase lag

37

Zero Phase Filtering

Phase lag = time delaySeparation of tremor from the intended motion without introducing phase lag

Prediction/projection capabilityAdaptive

Non-linear phase response of IIR filter, i.e. phase characteristic changes with frequency

38

Fourier Linear Combiner (FLC)

Truncated Fourier series to adaptively estimate amplitude and phase of periodic signal with known frequency (ω0)

39

Weighted-frequency Fourier Linear Combiner (WFLC)

Extends FLC to also adaptively estimate the frequency using another LMS algorithmBand-pass filter to select the band of interest

Assumption: rate of change of the dominant input signal frequency is slow

Zero-phase notch (band-stop) filter effect

FLC

WFLC

Signal input

ω0

Tremor

BandpassFilter

40

Weighted-frequency Fourier Linear Combiner (WFLC)

41

Weighted-frequency Fourier Linear Combiner (WFLC) Experiment

1 DOF motion canceling experimentAve. rms tremor amplitude reduced 69%

42

Tremor Recordings

0 5 10 15 20 25 30 35 400

0.5

1x 10-5 Spectrum of the Signal

0 5 10 15 20 25 30 35 400

0.5

1 x 10-5

0 5 10 15 20 25 30 35 400

0.5

1 x 10-5

Frequency

Band of frequencies

Band of frequencies

Band of frequencies

43

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 55

6

7

8

9

10

11

12

Time (sec)

Fre

quen

cy (

Hz)

Frequency adaptation in WFLC

The frequency adaptation becomes unstable due to the presence of two frequencies 8 & 8.6 Hz

44

Bandlimited Multiple FLC

To estimate the tremor signal within a band of frequencies or comprising of multiple frequency components (modulated signals)

Corresponding weight will adapt to the corresponding frequency of the input signalCan deal with input signals of multiple frequency components unlike WFLC

45

0 1 2 3 4 5−3

−2

−1

0

1

2

3

4

Time (sec)

WFLC Vs. BMFLC

BMFLCWFLC

Comparative Performance : Estimation Errors

Presence of two frequencies 8 and 8.6 degrades the performance of WFLC

46

Performance of BMFLC with Real-Tremor

0 5 10 15 20 25-1

0

1x 10-4 Tremor signal and compensated signal in x-axis

0 5 10 15 20 25-1

0

1x 10-4 Tremor signal and compensated signal in y-axis

0 5 10 15 20 25-1

0

1x 10-4 Tremor signal and compensated signal in z-axis

Red line shows the compensated motionBand of BMFLC: 3- 12 Hz.

47

0 5 10 15 20 25 30 35 400

0.5

1x 10-5 Spectrum of the Signal

0 5 10 15 20 25 30 35 400

0.5

1x 10-5

0 5 10 15 20 25 30 35 400

0.5

1x 10-5

Frequency

Spectrum of the compensated motion.

Performance of BMFLC with Real-Tremor

48

Analytical Integration via BMFLC

0( )

0 00

sin(2 ( ) ) cos(2 ( ) )L f f G

k r rr

r ry a f k b f kG G

π π= −

=

= + + +∑

0 02 20 0 0

sin(2 ( ) ) cos(2 ( ) )(2 ( )) (2 ( ))

Lr r

k r rr G G

a br ry f k f kf G f G

π ππ π=

⎡ ⎤= − + + +⎢ ⎥+ +⎣ ⎦

∫ ∫ ∑

Acceleration:

Velocity:

Position:

0 00 0 0

cos(2 ( ) ) sin(2 ( ) )(2 ( )) (2 ( ))

Lr r

k r rr G G

a br ry f k f kf G f G

π ππ π=

⎡ ⎤= − + − +⎢ ⎥+ +⎣ ⎦

∫ ∑

• Frequency components remain constant in BMFLC

• Once the weights adapt, the weights can also be assumed to be constant

By Performing analytical integration:

49

Microsurgery with Active Handheld Instrument

Visuomotor Control System

Noisy, Tremulous Motion

Motion Sensing

Visual Feedback

MICRON

VitreoretinalMicrosurgery

Estimation of erroneous motion

Manipulation of tip for active error compensation

•High precision tracking control

• Mechanism design

50

Manipulator Design

3 DOF piezoelectric-driven parallel manipulator

1 actuator per axis, max effective stroke = 12.5 µmMotion amplification = 9.4x, total stroke > 100 µm

Tool tip approximated as a point, hence only 3 DOF manipulationParallel manipulator design because

Rigidity, compactness, and design simplicity

51

Design of Parallel Mechanism

Flexure Based MechanismMonolith design using Stereolithography(SLA)∅22 x 58 mmIEEE EMBS 2005 (Shanghai) Best Student Design Competition winner

David Choi (CMU) et al.

52

25 Hz

15 Hz

5 Hz

Piezoelectric Actuator Hysteresis

Pros : High bandwidthFast responseHigh output force

Cons : Hysteresis

~15% of max. displacementHysteresis is rate-dependent

0 20 40 60 80 1000

2

4

6

8

10

12

14

Voltage (V)

Dis

plac

emen

t (µm

)

53

0 20 40 60 80 10010

20

30

40

50

60

70

80

90

0 0.2 0.4 0.6 0.8 1-20

0

20

40

60

80

100

Commercial Piezo-System with Feedback Controller

Piezo-driven 3 axis micro-positionerPolytec-PI, Germany, NanoCubeTM P-611>$ 10,000Feedback sensors: strain gagesTracking a 10 Hz, 100 µm p-p sinusiod

Hysteresis still presentLow-pass filtered behavior

Time (s)

Dis

plac

emen

t (µm

)

Measured Desired

Dis

plac

emen

t (µm

)

Voltage (V)

54

Feedforward Controller with Inverse Hysteresis Model

Develop an invertible mathematical model that closely describes the hysteretic behavior of a piezoelectric actuatorPrandtl-Ishlinskii Model

0 2 4 6 8 10 12 140

20

40

60

80

100

0 20 40 60 80 1000

2

4

6

8

10

12

14

0 2 4 6 8 10 12 140

2

4

6

8

10

12

14

Feedforward controller Piezoelectric

Desired Displacement (µm)

Actuator Response (µm)

Voltage(V)

Displacement (µm)

Dis

plac

emen

t (µm

)

Voltage (V)

Vol

tage

(V)

Actuator Response (µm)

Displacement (µm)

Dis

plac

emen

t (µm

)

Desired Displacement

(µm)

55

Prandtl-Ishlinskii (PI) Operator

Rate independent backlash operator:Hr = maxx(t) – r, minx(t) + r, y0Linearly weighted superposition of backlash operators:

r1

-r1

y

x

wh1

ri

y

x

Whi = Σwhi

[ ] )(,)( 0 tyxHwty rTh=

56

Inverse PI Hysteresis Model

Hysteresis Model

Inverse Model

Reflection about Y = X (45° line)

57

Tremor Tracking Results

Tracking recordings of real tremor using 1 piezoelectric stackRmse = 0.64% of max ampl.; Max error = 2.4% of max ampl.

0 0.2 0.4 0.6 0.8 2-2

0

2

4

6

8

10

12

14DesiredObtainedError

0 0.2 0.4 0.6 0.8 1-2

0

2

4

6

8

10

12

14DesiredObtainedError

Without Model Rate-Dependent Controller

Time (s) Time (s)

Dis

plac

emen

t (µm

)

Dis

plac

emen

t (µm

)

58

Adaptive Feedforward Controller

Eliminate parameter identificationWeight adapting mechanism: Recursive Least Square

[ ][ ] )(,)( 0 tyxHzwSw rThd

Ts p

)())(( tzbatzw iihi +=)(tzThw

Thw'

59

Experimental Results

1.30max error / actuator’s stroke length (%)

0.3899 ± 0.0291max error ± σ (µm)

0.31rmse / actuator’s stroke length (%)

0.0943 ± 0.0159rmse ± σ (µm)

Adaptive Rate-dependent Controller

0 5 10 15 200

5

10

15

20

25

30

Time(s)

Dis

plac

emen

t(um

)

measureddesirederror

5 5.5 6 6.5 7 7.5 8

0

5

10

15

20

25

30

Time(s)

Dis

plac

emen

t(um

)

measureddesirederror

60

Real-time Active Compensation – 1 DOF Disturbance, 1 DOF Compensation

61

Real-time Active Compensation – 1 DOF Disturbance, 3 DOF Compensation

62

Real-time Active Compensation –Handheld

C. Riviere (2006), Carnegie Mellon Univ.5 DOF sensing by 2 orthogonal position sensitive detectors No inertial sensingNon-surgical scenario

63

Active Pathological Tremor Compensation in Wearable Orthosis

64

Pathological Tremor

65

Pathological Tremor: Causes & Symptoms

3-12 HzFrom < 10 mm (fingers) to > 100 mm (arm)Common medical conditions

Essential tremor (postural tremor)Parkinson’s disease (resting tremor)Cerebellar dysfunction (intention tremor)

e.g. stroke, multiple sclerosis,Wilson’s disease

66

Impact of Pathological Tremor

buttoning writing Inserting a keytyping

Affects 5-9% of the population age > 40Activities of daily living become challenging or impossible

Social embarrassment and isolationLifetime cost per patient > US$1M

67

Treatment Options

Drug therapy> 50% are not responsive to drug

Stereotactic surgeryCost, psychological barrier, chances of complication

Assistive technologyActive tremor compensation via wearable orthosis

A 20-100 ms electromechanical time delay between Electromyograph (EMG) signals and muscle actuations

68

Active Tremor Compensation in Wearable Orthosis

69

Tremor Filter EMG Filter

Musculo-skeletal system

SensorFusion

Parameters Identification

Tremormodel

SignalProcessing

Inv. ArmModel

FES-Controller

Tremor EMG

Intended EMG

EMG

Acc

Tremor motion

Intended motion / position info

Model information

Camera systemSystem Overview

Modeling

Sensing

Filtering

Compensation70

Feedback Control System

Manipulator

Musculoskeletal System

AmplifiersStimulatorsController

Sensors

e.g. encoder

Sensors

e.g. EMG

71

Electrical Motor Muscle

Robot Arm vs Human Arm- Actuation system

(Hill-Type)

72

Robot Arm vs Human Arm- Links

VS.

Kinematics & Dynamics are solved in similar ways

Manipulator

Skeleton

73

Musculoskeletal Model of Upper Limb

Well studied and established

74

Key Challenges

SensingHow can we know what upper limb movement (tremulous + voluntary) will occur from the sensed SEMG of the muscles?

?

75

Key Challenges

FilteringHow can we differentiate between tremulous & voluntary SEMG of the muscles?

Tremor

Intended Movement

76

Key Challenges

Functional Electrical StimulationHow can we use FES to generate (anti-)tremulous movement in the upper limbs?

Stimulator

Controller

77

Musculoskeletal Modeling of Tremor in Upper Limbs

To understand the roles and characteristics of the skeletal muscles responsible for each type of tremorTo study SEMG-movement relationship

78

Tremor Study

120 patients with movement disorderParkinson Diseases – resting tremor (>30)Essential Tremor – postural tremor (>30)Multiple Sclerosis, Stoke, etc. – intention tremor (>30)Others (~10)

30 healthy peopleAge 16-85No personal & family medical history of tremor

National Neuroscience Institute, Singapore

79

Pathological Tremor Study18 control subjects, 5 Parkinson’s Disease patients, 6 Essential Tremor patients, 1 Psychogenic tremor patient, and 1 Holmes’ tremor patient

80

Data Collection

Record sensor data of patients performingStandard diagnostics: finger to noise, finger to finger, stretched out, drawing spirals, etc.

SensorsSEMG: 16 channels (8 muscle groups, mostly agonist & antagonist pairs) per limbAccelerometers: 3 x tri-axial per limbGoniometers: 1 per limbPosition and Orientation sensor: Vicon optical sensing system (4 cameras)

81

Sensing

SEMG

Accelerometers

Orthosis

SEMG

Accelerometers

Goniometer

Optical Sensors

Model

Sensor

Fusion

Muscle Activity & Physical Movement

82

CE : Contractile Element

SE : Series Elastic Element

PE : Passive Element (Parallel Elastic Element)

Muscle mechanical model (Hill-type)

Muscle Contraction Property

83

ϕ

211

2

211

2 ωωω

++ DsK

The spring damper systemis tuned to the actual tremorfrequency

Phenomenological Modeling of Upper Limb Tremor from EMG

84

Sensor Fusion – Kalman Filtering

EMG-derived joint angle as predictorACC-derived joint angle as corrector

Estimate of joint angle

( ) (1) ( ) (2)( ) (1) ( ) (2)

EMG EMG EMG

ACC ACC ACC

k c EMG k ck c ACC k c

θθ

= += +

2

2 2( ) ( ) ( ( ) ( ))EMG

EMG ACC EMGEMG ACC

k k k kσ

θ θ θ θσ σ

= + −+

85

Results

σ2true = 0.1045

σ2error = 0.022 (78.76% reduction)

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

time (s)

degr

eeKalman filter for elbow flextion-extension of PD patient (sitting, arm resting on thigh)

estimatedtrueerror

86

Results

0 5 10 15 20 250

0.2

0.4

0.6

0.8

1

X: 5.054Y: 0.902

Frequency content of the compensated tremor

frequency (Hz)0 5 10 15 20 25

0

2

4

6

8

X: 5.029Y: 7.315

Frequency content of original tremor

frequency (Hz)

Power of original tremor = 7.315 Power of compensated tremor = 0.902

The power of the tremor is reduced by 87.67%

87

Target circle

••• NN outputRaw trajectory

∆ Start position End position* Target circle

Tremor Filtering via ANN

Cascade Correlation Neural Networks with extended KalmanFiltering Experiments with 11 multiple sclerosis patients (intention tremor)Smoother trajectoryReach and dwell in target circle 31.8% faster

Mean over 29 tests

Output

Hidden

Input

88

Functional Electrical Stimulations

Controlling electrical pulses to stimulate the intact peripheral nerve to actuate muscles

Usually used to restore the motor functions for the paralyzed patients

Prochazka et al. (1992) demonstrated the effectiveness of FES for tremor attenuation

Offline trial & error tuning of intensity and phase

89

Fatigue

Muscle Activation

90

Stimulation Artifact

Blocking windowTurn off EMG channels when stimulation is on

91

FES Controller Design

92

0 2 4 6 8 10-800

-600

-400

-200

0

200

400

600

time [sec]

angu

lar v

eloc

ity [d

eg/s

ec]

Simulation Result of Tremor Suppression

No ethical clearance on patient trial yet

93

Wearable Orthosis

Wearable bands thin enough to be worn under sleevesBattery poweredMicrocontroller basedBeyond the first step

False alarmEEG

94

Questions & Comments

Wei Tech ANGSchool of Mechanical & Aerospace Engineering

Nanyang Technological UniversitySingapore

wtang@ntu.edu.sghttp://www.ntu.edu.sg/mae/centres/rrc/biorobotics/biorobotics.htm