Robotic Assisted Hand for Learning

a Timing-based Task by the Elderly

Preliminary Results

Amy E. Bouchard

and Marie-Helene Milot

Bishop’s University, 2600 College St., Sherbrooke, J1M 1Z7 Québec, Canada

Université de Sherbrooke, Faculté de médecine et des sciences de la santé, École de réadaptation,

2500, boulevard de l'Université Sherbrooke, J1K 2R1 Québec, Canada

1 INTRODUCTION

Timing of movement is crucial in the performance

of daily tasks, like playing tennis (Marchal-Crespo

et al., 2013). With age, the need to learn new timing

tasks persists (e.g. learning to drive a powered

wheelchair). However, significant impairments in

timing have been noted, like longer execution timing

of movements (Seidler et al., 2010) and slower

reaction times (Marchal-Crespo et al., 2010).

To improve motor learning, two types of robotic

training have been studied: haptic guidance (HG)

and error amplification (EA). HG suggests that the

learning of a motor task can be enhanced by

showing the correct movement in order to teach the

motor system how to imitate it (Patton and Mussa-

Ivaldi, 2004). EA is based on the idea that error

drives learning; by artificially increasing error, a

faster and more complete learning can be achieved

(Emken and Reinkensmeyer, 2005).

Both types of training have significantly

improved the temporal aspect of movement in young

healthy people (Luttgen and Heuer, 2013, Marchal-

Crespo et al., 2013, Milot et al., 2010). However,

few studies have used HG or EA to try to improve

movement timing in the elderly.

Up till now, only one study has used HG training

to improve seniors’ timing. Results showed an

improvement in timing when they had to straighten a

wheel immediately after turning it (Marchal-Crespo

et al., 2010). It seems that no study has directly

evaluated and compared the impact of HG and EA

on the improvement of timing errors for the elderly.

2 OBJECTIVES

The objective of the current project is to evaluate

and compare the impact of HG and EA robotic

training types on the immediate improvement in

timing error for elders. This project will aid in the

understanding of the efficacy of robotic therapy to

improve timing for seniors, and help gather

reference values for a future study on chronic stroke

survivors.

3 METHODS

Subjects had to meet the following criteria: 1) be

aged ≥60 years; 2) be able to painlessly flex their

right wrist 10

; 3) be right-handed. The exclusion

criteria included: 1) having a cognitive impairment

(score 25/30 on the MoCA exam); 2) having an

active neurological or orthopaedic problem of the

right upper limb; 3) having a vision problem which

would inhibit the proper viewing of the game’s

computer screen.

3.1 Timing Exerciser Orthosis (TEO)

TEO (Figure 1) is modified from TAPPER, a robot

used in one of our previous studies (Milot et al.,

2010). TEO is a one-degree-of-freedom robot that is

mechanically actuated by a Dynamixel MX-106

actuator (Robotis inc, USA), mounted on an

aluminium frame and connected to an articulated

hand allowing flexion/extension of the right or left

hand. A forearm brace is placed on the frame to

ensure the proper stabilization of the subjects. All

the apparatuses are connected to a USB-6008 data

acquisition card (National Instruments, USA) and

sampled at 5000 Hz. A button is also attached to the

frame to ensure sensory feedback, since the subjects’

fingers touch this button at each movement.

3.2 Pinball Simulator

The pinball simulator was designed with

Bouchard A. and Milot M..

Robotic Assisted Hand for Learning a Timing-based Task by the Elderly - Preliminary Results.

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

LabVIEW 2013. The goal of the task was to hit as

many targets as possible by triggering a wrist

movement at the proper timing, to activate TEO

(torque ≥ 0.5Nm). When activated, TEO caused the

flipper to rotate on the computer screen and lead the

falling ball towards a randomly positioned target.

Subjects were successful when the ball hit the target

at a timing accuracy of 4 ms. Visual feedback was

provided to the subjects on each trial (e.g. “Wow!

Just on time!” and “Too early! Hit later!”).

Figure 1: TEO and the computerized pinball-like game.

3.3 Haptic Guidance and Error

Amplification Algorithms

To decrease subjects’ timing errors during HG, we

delayed or sped up the start of the robot when the

subjects initiated wrist movement too early or too

late, respectively. The exact opposite was done to

increase errors during EA. The algorithms were

based on our previous study. In sum, t = 0 was

defined as the time the ball began falling toward the

flipper, and Tbp was defined as the time in which

TEO moved. Now:

Tbp = Tip+Dc (1)

where Tip is the time the motor sensors detected the

initiation of a wrist flexion by the subject and Dc

was a programmed delay from when the subject

initiated movement to when TEO was commanded

to move. For each target, the values that ensured

success were defined as Tbd, Tid and so

Tbd = Tid+Dcd (2)

where Tbd is the anticipated time in which TEO

must move in order to successfully hit the target, Tid

represents the desired time when the subject should

initiate a wrist movement and Dcd is a constant

(0.5s).

The subject’s timing error in initiating a

movement was defined as Ep, so:

Ep = Tip-Tid (3)

Next, TEO timing error was defined as Eb:

Eb = Tbp-Tbd = Ep+Dc-Dcd (4)

We wanted Eb to be proportional to Ep:

Eb = kEp (5)

where k is the error-amplification gain. Substituting

equations 4 into equation 5 and solving for Dc, the

programmed delay gave:

Dc = Dcd+Ep(k-1) (6)

As we wanted each subject to experience a 30%

rate of success, we adjusted the k value during a 39-

trial adjustment phase. Since Eb’s maximum value

was 4 ms (the upper limit of timing accuracy to be

successful), the k value was calculated accordingly

using equation 5. Therefore:

k = 4 /Ep (7)

To do so, we classified each subject’s timing errors

in an ascending order and took the 12

Ep value to

calculate the final k value. The k value was then

increased or decreased by 90% during EA and HG

training, respectively, to increase or decrease each

subject’s timing error.

3.4 Study Timeline

Each subject received the HG and EA trainings in a

random order. First, a baseline condition (B1) was

played at the adjusted game difficulty for 40 trials.

B1 was followed by either HG or EA each having 75

trials. A retention condition (RC), identical to B1,

followed each training condition. The absolute and

relative timing error values at B1 and RC were

retained. T-tests were used to evaluate the difference

in timing error between B1 and RC, for each training

condition, and for the difference in the change in

timing error obtained between both training

conditions. The p value was set at 0.05.

4 RESULTS

Eleven subjects (mean age 684 years) took part in

the study. When comparing the first and last 10 trials

of B1, to evaluate the presence of a learning plateau,

no change in the subjects’ timing errors were noted

(127 vs 11 4 ms; p=0.2). This means that they had

reached a learning plateau before being introduced

to the training conditions.

A significant difference in timing error was

found when comparing the last 10 trials of B1 with

the first 10 trials of HG (125 vs 10.8 ms; p0.05)

and EA (114 vs 226 ms; p0.05). This means that

introducing subjects to HG and EA significantly

decreased and increased their timing errors,

respectively.

4.1 Impact of HG/EA on Timing Error

When comparing the absolute timing error during

RC to that of B1, no improvement in timing error

was noted (p0.14), regardless of the training

condition (t(10)=-1.2, p=0.13) (Figure 2 A).

Figure 2: Comparison of subjects’ A) absolute and B)

relative timing error between the baseline condition and

retention condition following HG and EA robotic training.

However, when analyzing the relative timing

error, where a negative value indicated that the

subjects initiated movement too early, a trend

towards an improvement in timing error was noted

when comparing B1 to RC following HG training (-

412 vs 0.0110 ms; p=0.09), paralleled by a trend

towards a decrease in the variability of the relative

timing error (SD) (1711 vs 126 ms; p=0.08). This

means that subjects learned to initiate movement

later to more successfully hit the targets, and were

more homogenous in doing so. No difference was

noted when comparing B1 to RC following EA

training (-0.97 vs -412 ms; p=0.2) (between

conditions, t(10)=1.3, p=0.11) (Figure 2B).

5 DISCUSSION

These preliminary results suggest that as age

increases, learning can still occur since the subjects’

relative timing error decreased after HG training.

This also supports the results of previous studies on

the elderly’s ability to learn new tasks (Marchal-

Crespo et al., 2010).

Moreover, it appears that a robotic assisted hand

could be an effective approach in improving elders’

timing errors; however, only HG appears to benefit

them. This supports the results of our previous

study, which was conducted on young healthy

individuals (Milot et al., 2010); here, less-skilled

subjects did not benefit from EA in the timing-based

task (k value  0.1). It is plausible that for this sub-

group of subjects, EA training was too challenging

since the motor system was overwhelmed with too

much information, preventing any improvement in

performance. This could be the case in this current

study, since the seniors’ mean k value is 0.07 (range:

0.02; 0.1), falling into the less-skilled sub-group

category.

This current study is part of an ongoing project,

so more subjects are needed in order to validate the

preliminary results and to assess the long-term

benefits of HG and EA on improving movement

timing. If HG and EA trainings are proven to

effectively do so, they could potentially help

improve the movement timings of neurologically

impaired individuals like chronic stroke survivors.

REFERENCES

EMKEN, J. L. et al. 2005. Robot-enhanced motor

learning: accelerating internal model formation during

locomotion by transient dynamic amplification. IEEE

Trans Neural Syst Rehabil Eng, 13, 33-9.

LUTTGEN, J. et al. 2013. The influence of robotic

guidance on different types of motor timing. J Mot

Behav, 45, 249-58.

MARCHAL-CRESPO, L. et al. 2010. The effect of haptic

guidance, aging, and initial skill level on motor

learning of a steering task. Exp Brain Res, 201, 209-

20.

MARCHAL-CRESPO, L. et al. 2013. The effect of haptic

guidance and visual feedback on learning a complex

tennis task. Exp Brain Res, 231, 277-91.

MILOT, M. H. et al. 2010. Comparison of error-

amplification and haptic-guidance training techniques

for learning of a timing-based motor task by healthy

individuals. Exp Brain Res, 201, 119-31.

PATTON, J. L. et al. 2004. Robot-assisted adaptive

training: custom force fields for teaching movement

patterns. IEEE Trans Biomed Eng, 51, 636-46.

SEIDLER, R. D. et al. 2010. Motor control and aging:

links to age-related brain structural, functional, and

biochemical effects. Neurosci Biobehav Rev, 34, 721-

33.

-20

-15

-10

-5

Baseline

Retention

Baseline

Retention

Timing error (ms)

Timing error

Baseline

Retention

Baseline

Retention