Post on 23-Aug-2020
Speaker Area
People explore gait dimensions, and reduce this explorationas they learn to walk with exoskeleton assistanceSabrina Abram1, Katherine Poggensee2, Steve Collins2, Max Donelan1
1Locomotion Lab, Department of Biomedical Physiology and Kinesiology, Simon Fraser University2Biomechatronics Lab, Department of Mechanical Engineering, Stanford University
IntroductionThe success of assistive devices relies on users learning to take advantage of the assistance [1]. In both everyday tasks and novel conditions, the nervous system is faced with the trade-off between exploiting, perhaps erroneously, previously learned strategies and exploring new, unknown strategies [2]. The goal of this study was to test how people balance this trade-off when learning to walk with ankle exoskeleton assistance. To accomplish this, we performed a post-hoc analysis of data from our previous study [3].
We hypothesized that 1) people explore many candidate gait dimensions as they identify which dimensions can take advantage of assistance, and 2) people reduce this exploration with experience as they learn to exploit new strategies that lower metabolic cost.
Experimental setupFigure 1. We randomly assigned 5 participants to a Static Group, and 5 to a Continued Optimization Group. Both experienced a predefined Generic Assistance (GA; blue) torque profile. Static repeatedly experienced it whereas Continued Optimization also experienced human-in-the-loop optimization (HILO; red). By design, both groups can influence the torque timing by varying their step frequency as well as the power and work applied to the ankle by varying their ankle kinematics. We are interested in how the nervous system learns to take advantage of this assistance.
Experimental protocolFigure 2. We analysed how all participants (n=10) learned to walk with repeated exposure to GA across multiple days. (A) All participants completed a training session on each day for a total of 6 days. (B) From day 2 onwards, participants first completed an adaptation trial where they either experienced GA repeatedly (Static Group) or periodically (Continued Optimization Group). (C) All training days ended with a validation trial that included two, 6-minute GA conditions.
1. Exploration �rst increases along many gait dimensions, and then decreases with increased experience.Figure 3. We observed higher variability along (A) step frequency, (B) ankle angle, (C) total soleus activity, and (D) total medial gastrocnemius activity at the beginning of the multi-day protocol compared to baseline levels in normal walking (NW) or zero torque (ZT), and lower variability on the last day compared to the first day.
3. Exploration results in adaptation along some gait dimensions.Figure 5. Participants learned to adapt their (A) step frequency, (C) total soleus activity and (D) total medial gastrocnemius activity with experience. We did not observe adaptation in (B) ankle angle range during stance when comparing the last day to the first day.
2. Exploration converges on baseline levels of exploration for some gait dimensions with increased experience.Figure 4. Variability decreased along all gait dimensions and converged on baseline levels for (A) step frequency and (C) total soleus activity, whereas it remained elevated for (B) ankle angle and (D) total medial gastrocnemius activity.
4. New strategies result in lower metabolic cost. Here we find that the nervous system
learns to reduce metabolic cost by first exploring along many gait dimensions, and then reducing this exploration with experience. However, this only results in adaptation along some dimensions, suggesting that the nervous system did not know a priori which dimensions to adapt.
Conclusions
[1] N. Sanchez, et al. (2019) J Physiol.[2] R.C. Wilson, et al. (2014) J Exp Psychol Gen.[3] K.L. Poggensee, et al. In Prep.
References
A.
B.
Day 1Training
Static Group
Continued Optimization Group
0 10 20 30 40 50 60 70 80Time in exoskeleton (mins)
Con
ditio
n
0 10 20 30 40 50 60 70 80
Con
ditio
n C.
Con
ditio
nC
ondi
tion
Normal walkingZero torque
GAHILO
A. B. C. D.
p=0.75p=4.5x10-7
p=0.52
τ=115.1 mins τ=207.2 mins τ=52.6 mins τ=6.0 mins
A. B. C. D. A. B. C. D.
1
2
3
1
2
3
4
5
6
1
2
3
1
2
3
estimateinstantaneous cost
human-in-the-loop optimization
Time (Stride)
Torq
ue
step frequency
genericassistance
Time (Stride)
Torq
ue
NW GADay 1
GADay 6
0
1
2
3
4
5
ZT GADay 1
GADay 6
0
1
2
3
4
5
NW GADay 1
GADay 6
0
0.02
0.04
0.06
NW GADay 1
GADay 6
0
0.02
0.04
0.06
Step
freq
uenc
y va
riabi
lity
(bpm
)
Ankl
e an
gle
varia
bilit
y (d
eg)
Tota
l sol
eus
varia
bilit
y (n
orm
. to
NW
pea
k ac
tivat
ion)
Tota
l med
ial g
astro
c. v
aria
bilit
y (n
orm
. to
NW
pea
k ac
tivat
ion)
Step
freq
uenc
y va
riabi
lity
(nor
mal
ized
to N
W)
Ankl
e an
gle
varia
bilit
y(n
orm
aliz
ed to
ZT)
Tota
l sol
eus
varia
bilit
y (n
orm
aliz
ed to
NW
)
Tota
l med
ial g
astro
c. v
aria
bilit
y (n
orm
aliz
ed to
NW
)
0 200 400 600Time in exoskeleton (mins)
0 200 400 600Time in exoskeleton (mins)
0 200 400 600Time in exoskeleton (mins)
0 200 400 600Time in exoskeleton (mins)
0.9
1
1.1
1.2
0.5
1
1.5
2
2.5
0.5
1
1.5
0.5
1
1.5
2
0 200 400 600Time in exoskeleton (mins)
0 200 400 600Time in exoskeleton (mins)
0 200 400 600Time in exoskeleton (mins)
0 200 400 600Time in exoskeleton (mins)
Step
freq
uenc
y(n
orm
aliz
ed to
NW
)
Ankl
e an
gle
rang
e (n
orm
aliz
ed to
ZT)
Tota
l sol
eus
activ
ity
(nor
mal
ized
to N
W)
Tota
l med
ial g
astro
c. a
ctiv
ity
(nor
mal
ized
to N
W)
commanded torque
metabolic cost
p=1.7x10-4
p=9.1x10-5
p=1.9x10-6
p=4.8x10-4
p=0.0016p=0.0096 p=4.4x10-5
p=0.0018 p=0.015 p=0.10 p=0.013 p=0.0016
Figure 6. The average time constant of exploration (τ=95.2±87.0 mins) is similar to that by which the nervous system reduces metabolic cost (τ=99.9 mins).
Day 2Training
Day 3Training
Day 4Training
Day 5Training
Day 6Training
0 6 12 24 30 3618
0 6 12 24 30 3618 42 48
p=1.9x10-4
τ=21.9 mins τ=172.0 mins τ=274.5 minsτ=837.4 mins
Time in exoskeleton (mins)
Continued Optimization Group
Static Group
0 200 400 600Time in exoskeleton (mins)
1
2
3
4
5
Net
met
abol
ic p
ower
(W/k
g)
τ=99.9 mins