Breaking It Down Is Better Haptic Decomposition of Complex Movements Aids in Robot- Assisted Motor Learning J. Klein, S. Spencer, & D. Reinkensmeyer IEEE.

Breaking It Down Is Better

Haptic Decomposition of Complex Movements Aids in Robot-Assisted Motor Learning

J. Klein, S. Spencer, & D. Reinkensmeyer

IEEE Transactions on Neural Systems and Rehabilitation Engineering

May 2012

Motivation – Klein et al.

• sports training• rehab from neurologic injury

– stroke– spinal-cord injury

Motivation – CHARM

• stroke rehab (Michele, Mike)• needle-steering training (Ann)• robot-assisted surgery (Ilana)

Background

HAPTIC GUIDANCE• of questionable effectiveness• to date, been used almost

exclusively to demonstrate entire movements

“PART-WHOLE” TRANSFER• sequential vs. continuous tasks• effectiveness correlated with

coordination requirements

shoulder rotation

elbow flexion/extension

desired trajectory

Hypothesis

“Decomposing” a movement into fewer-DOF components is more effective than training the movement as a whole. Moreover, the degree of effectiveness is dependent on the form of the decomposition.

Methods

4 DOFs1. shoulder abduction/adduction2. should flexion/extension3. shoulder int./ext. rotation4. elbow flexion/extension

EULER

Methods

2 Motions1. “main” (θ) ≈ tennis backhand2. transfer (θ’) ≈ front crawl swim

Reflection…

What if the task had been framed as a goal-directed movement?

That is, what if subjects had been told: “pretend

you are swinging a tennis racket”?

Methods

Experimental Protocol1. Baseline2. Baseline Transfer3. Training4. Assessment5. Assessment Transfer6. Retention7. Retention Transfer

4 Training Groups1. “Whole” (control)2. Euler3. Anatomical4. Visual

Reflection…

Klein et al. randomized the presentation order

of the joint components during Training. How necessary was this

control? Is component presentation order an area for further study?

rand{}

Methods

Assessment

“GLOBAL”

lower score = better learning

“LOCAL”

* accounting for time delay d

Reflection…

Why is it necessary to fit the shift parameter d? Shouldn’t d simply be the time between the

assessment start signal and movement onset? ...

Reflection…

… Are we assuming that subjects are “playing catch up” with the virtual arm? What if, instead of being

shown a virtual arm, subjects were given binary feedback (e.g., GREEN vs.

RED) based on their current performance?

Results“MAIN” TRANSFER

Results

• no Baseline differences between decomposition groups

• all groups significantly improved with Training• Anatomical decomposition exhibited greatest

improvement during all assessments– Training – no significant difference compared to “Whole”

– Short-Term – significant compared to all other groups

– Long-Term – only (weakly) significant compared to Visual

Reflection…

Why would there be greater learning at long-

term retention (vs. short-term retention) for

“Whole”, Euler, and Visual decompositions*?

Given enough time, would all groups

equalize?

* Anatomical was approximately the same

Results

• “global” performance correlated with

1. “local” performance2. proximity of training joint positions to those

required by whole motion

• no improvement for transfer motion haptic guidance training IS task specific

Discussion

“Part-Whole” Learning

• counterintuitive• success of Anatomical (vs.

Visual) decomposition suggests that spatiotemporal summation mechanism operates in joint (vs. visual coordinates)

* see (Kakei et al., 2001)

• NOT generalizable• more spatial than temporal

Reflection…

I am unconvinced by Klein et al.’s theory for why

(anatomical) movement decomposition aids in

learning a target motion more than practicing the motion itself. What other

explanations can we propose?

“One possibility is that the motor system has trouble determining where the problems lie in making, accurate, complex movements; breaking the movements down may allow better identification and then more focused practice on key problems.”

Discussion

“Part-Whole” Learning

• counterintuitive• success of Anatomical (vs.

Visual) decomposition suggests that spatiotemporal summation mechanism operates in joint (vs. visual coordinates)

* see (Kakei et al., 2001)

• NOT generalizable• more spatial than temporal

Robotic Movement Training

• opens the door for simpler robotic devices in rehab

• BUT, benefits of movement decomposition might only exist with haptic guidance

Reflection…

What does this mean for stroke rehabilitation? Is

there a “whole-part” mechanism that could

help decouple patients’ abnormal muscle

synergies?

shoulder rotation

elbow flexion/extension

desired trajectory

“Breakthrough” or just another addition to the pool of inconclusive

literature on haptic guidance?