Material(integrated,Intelligence,...
Transcript of Material(integrated,Intelligence,...
Material(integrated,Intelligence,for,Robot,Autonomy
Nikolaus)Correll
University)of)Colorado)at)Boulder
PhD)EPFL)2007)@)SWIS)(DISAL)
Components,of,Robotic,Systems
1.)Perception 2.)Planning)and)Control3.)Tight)integration)of)sensing,
Computation)and)actuation
Robotics Robotic)Materials
@correlllab @correlllab
Robotic,Materials
M.)McEvoy and)N.)Correll.)Materials)that)couple)Sensing,) Actuation,)Computation) and)Communication.)Science,)2015.@correlllab @correlllab
Tight)integration)of)sensing,)actuation)and)computation)
inside(the)material
Advantages,of,distributed,computation
• High,bandwidth( sensing:)soft)robotic)skin
• High,speed(Feedback(control:)object)manipulation
• Distributed(computation:(optimization)
and)classification
@correlllab @correlllab
Example,I:,Soft,Sensing,SkinRubbed)
Texture
Signal Microphone
FFT
Ambient)Signal
Signal)FFTMeasured)Signal Transient)Signal
Ambien
t*Up
date
1000)Hz
@correlllab @correlllab
Discrete,Computation:,DFT,and,filtering
@correlllab @correlllab
ADC)
@1kHz
Continuous
Signal256)bytes)
FIFOS(f,t)=
f
256ms 15.9ms)on)Atmel)Xmega 128A3
0.6ms
Material)
Physics
D. Hughes, N. Correll (2015): Texture Recognition and
Localization in Amorphous Robotic Skin. In: Bioinspiration &Biomimetics, 2015.
Discrete,Computation:,Classification
@correlllab @correlllab
Human)Skin Brillo
DFT)Spectrum
…Linear)regression
15*128)=)1920)weights
0.6ms
71.7%)accuracy
100)samples
10cfold)crosscvalidation
Example,II:,Shape,change,by,Variable,Stiffness• Octopus:)No)skeleton)c>)variable)stiffness)using)muscular)hydrostatsKier,)1985
• Arm)can)make)autonomous)
movementsGraziadei,)1965
• Redundant)mechanisms)pose)great)
load)on)motor)control)systemGutfreund,)2006
• 2/3rds)of)the)animal’s)neurons)in)
arms (peripheral)nerve)cord)Hochner,)2013
@correlllab @correlllab
Hochner,) 2013
Variable,Stiffness,beam,with,embedded,computation
@correlllab @correlllab
M.)A.)McEvoy,)N.)Correll)(2014):) "Thermoplastic)variable)stiffness)composites)with)embedded,)
networked)sensing,) actuation,)and)control.)In:)Journal)of)Composite)Materials.
Original)State Tune)specific)stiffness Apply)external)moment
Fix)beam)length)/)melt Freeze Repeat)for)negative)curvature
Inverse,Kinematics
• Large)N)(DoF))require)iterative)methods
• Inverse(Jacobian Method• Calculate)small)changes)to)input)
that)reduce)error
• Move)into)that)direction)using)
gain)α• Repeat
• J)is)a)3xN)matrix,)pseudoc
inverse)is)NxN c>)O(N3)
g
s(κ)
@correlllab @correlllab
Distributed,Computation,of,Matrix,Inverse
• Distributed)algorithm)requires)
O(log)N))time
• Problem:)requires)O(N4))
computers
• Idea:)reduce)arm)to)multiple)
M<<N)DoF systems)and)solve)
sequentially
3)DoF 3)DoF
3)DoF
@correlllab @correlllab
M. A. McEvoy, N. Correll (2016): Distributed Inverse Kinematics for Shape-Changing Robotic Materials . In:3rd International Conference on System-integrated Intelligence: New Challenges for Product and Production Engineering , Paderborn, Germany. (Best Paper)
Convergence,properties
@correlllab @correlllab
Likelihood) to)find)a)solution Average)error Required)iterations
Scalability)improves) from)O(N3))to)O(M3N))
Example,III:,Manipulation,and,Feedback,Control
@correlllab @correlllab
17
@correlllab @correlllab
N.)Farrow,)N.)Correll)(2015):)A)SelfcContained)Soft)Pneumatic)Actuator)that)can)sense)touch)
and)grasp.)Int.)Conf.)on)Intelligent)Robots)and)Systems)(IROS).)
Active,complianceInitial)grasp
(bad)force)closure)
After)wiggling
(good)force)closure)
Nicholas) Farrow,)Yang)Li)and)Nikolaus) Correll.)“Morphological) and)Embedded)Computation) in)a)A)Selfc
contained)Soft)Robotic)Hand”.)arXiv:1605.00354
Combined,distance,and,force,sensing
R.)Patel)and)N.)Correll.)Integrated)force)and)distance)sensing) for)robotic)manipulation) using)elastomerc
embedded) commodity)proximity) sensors.)``Robotics:)Science)and)Systems'')(RSS),)2016.
Tactile,and,distance,sensing,during,grasping
1.)Approach
2.)Align
3.)Contact
4.)Loading
5.)Disturbance
(wrench)
6.)Disturbance
(tapping)
7.)Replacement
8.)Release
R.)Patel)and)N.)Correll.)Improving)grasp)performance)using)inchand)proximity)and)
force)sensing.)Int.)Symposium) on)Experimental)Robotics)(ISER),)2016.
Whole(body,sensing:,social,touch,and,collision,avoidance
D.)Hughes,)N.)Farrow,)H.)Profita,)N.)Correll (2015): Detecting)and)Identifying)
Tactile)Gestures)using)Deep)Autoencoders,)Geometric)Moments)and)Gesture)
Level)Features. ACM)International)Conference)on)Multimodal) Interaction,)
Recognition)of)Social)Touch)Gestures)Challenge, Seattle,
Multi(modal,manufacturing
• Combine)additive)and)
subtractive)manufacturing)
techniques
• Use)robots)for)precise)placement)of)components)
and)interconnects
• Use)robots)to)provide)the)“glue”)between)different)
manufacturing)steps
Conclusion• Distributed solutions)for)classification,)optimization)
and)control)problems
• Scalable*with)the)number)of)elements
• Robust*against)failures)of)individual)elements
• Distributed,)Scalable,)Robust:)Autonomous)Material
• Challenges• What’s)the)underlying)science)of)autonomous)materials
• What)kind)of)autonomy)would)otherwise)not)possible
• What)are)the)lowchanging)fruits)relevant)to)industry)now
@correlllab @correlllab