designing exoskeletons and prosthetic limbs that enhance

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 1/34

Steve CollinsAssociate ProfessorMechanical EngineeringStanford University

Designing exoskeletons and prosthetic limbs

that enhance human performance

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 2/34

First: A difficult design problem.

Then: Tools that help.

Finally: Energy-efficient embodiments.

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 3/34

Goal: Restore and enhance mobility

[Daily Herald (2013)] [Duncan et al. (2007)] [V. Lockett (2016)] [UNICEF (2016)]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 4/34

[Yagn (1890); Seireg (1971); Zoss et al. (2006) ; van Dijk et al. (2011)]

Challenge: Improving performance is hard

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 5/34

Best results: Modest improvements

[MIT] [Harvard] [Ghent] [NCSU/CMU]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 6/34

Problem: What should the device do?Intuition (only) inspiresPresent models don’t predict

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 7/34

Solutions:

Universal Device EmulatorsExperiments on humans, fast.

Human-in-the-Loop OptimizationDiscover, individualize and co-adapt.

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 8/34

prosthesis

motor

Emulators: Fast, cheap and in control

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 9/34

…

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 10/34

End effectors: Easy to get good performance

[Caputo & Collins (2014) JBME; Collins et al. (2015) ICRA; Witte et al. (2017) ICORR]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 11/34

HLO: Discover. Customize. Adapt.

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 12/34

Big idea: Human-In-the-Loop Optimization

[Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 13/34

Specific method: Minimize metabolic rate

[Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 14/34

Application: Ankle torque pattern

[Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 15/34

Application: Exoskeleton on one ankle

[Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 16/34

Metabolic cost: Exceptionally large reduction

[Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 17/34

Optimized settings: Substantial variation

[Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 18/34

Static comparison: Suggests three contributors

[Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 19/34

Static comparison: Facilitating motor learning

[Jackson & Collins (2015) J. Appl. Physiol.; Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 20/34

19.3%reduced

*

[Jackson & Collins (2015) J. Appl. Physiol.; Zhang et al. (2017) Science]

Static comparison: Facilitating motor learning

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 21/34

Video here

[Zhang et al. in press]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 22/34

Single-subject studies: Test generality

[Zhang et al. (2017) Science]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 23/34

[Zhang et al. (2017) Science]

Single-subject studies: Test generality

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 24/34

Implications: Great tools for design.

Future work: Faster learning + optimization

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 25/34

Questions?

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 26/34

Going mobile: Efficient assistive robotics

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 27/34

Bioinspiration: Ankle catapult mechanism

Ultrasound imaging: Calf muscles nearly static

[Ishikawa et. al. (2005) J Appl Physiol][Farris et. al. (2013) J Appl Physiol]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 28/34

Device: Unpowered ankle exoskeleton

[Collins et al. (2015) Nature]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 29/34

Results: Reduced force more efficient walking

[Collins et al. (2015) Nature]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 30/34

Limitation: Controllability

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 31/34

Actuator design: Electroadhesive clutch

[Diller et al. (2016) ICRA]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 32/34

Results: Electroadhesive clutch characteristics

One clutch pair:Peak force: 200 NMass: 1.5 gPower used: 0.3 mWApplied Voltage: 300 VSwitching time: 0.01 s

Demonstration system:Mass: 26 gEfficiency: 95%Fatigue life: > 2,000,000

[Diller et al. (2016) ICRA]

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 33/34

Future work: Energy-recycling actuators

High-bandwidth, discretely-variable force control

Steve Collins Stanford University biomechatronics.cit.cmu.edu @StevenHCollins stevecollins@stanford.edu 34/34

Thank you:

Sponsors:

PieterFiers

RachelJackson

KatherinePoggensee

KirbyWitte

JuanjuanZhang

ChristopherAtkeson

StuartDiller

GregSawicki