distributed intelligent systems – w13: self-aggregation...
TRANSCRIPT
Distributed Intelligent Systems – W13:Self-Aggregation and Self-Assembly in Natural and Assembly in Natural and
Artificial Systemsy
1
OutlineOutline• Self-aggregation in artificial
systems• Collective decision mediatedCollective decision mediated
by self-aggregation (mixed societies)societies)
• Self-assembly General background– General background
– Artificial examples (sub-mm and cm scales)and cm scales)
2
Robot Self-Aggregation using Wireless Communication - Overview
[Correll & Martinoli, IJRR 2011][ , ]
• Eg. 2 clusters of 3 robots, 6 of 1 robot• Webots experiments using Alice II with
radio, 12 robots totalA di t 1
Real robotic node: Alice II with 802.15.4 com module
• Arena diameter: 1 m• Communication range: 7-12 cm• Simplified com model 4
Original Inspiration: Aggregation in Gregarious Arthropods
• Aggregation in Cockroaches• Individual Behavior is well
di dstudied• Probability to rest is a
function of the number offunction of the number of cockroaches nearby
• Probability to move ydecreases when in larger aggregates
© Sempo, Halloy, Detrain & Deneubourg
5
Individual Behavioral Parameters
pjoin pleave
D ll d bData collected by Jeanson et al. (Animal Behavior 2004)
# of cockroaches nearby
6
Experimental Setup• Submicroscopic model (Webots)• Estimate of neighborhood using local
communication used as ancommunication used as an “omnidirectional teammate sensor”
• Local communication: – presence or absence of a teammate in thepresence or absence of a teammate in the
communication range– use of static or from a large set dynamically
randomly chosen IDs for facilitation of the discriminationdiscrimination
– no additional explicit information shared (e.g., counting process involving multi-hop, etc.)
• Objective: relation between individual and collective dynamics rather than biological accuracy
i i ( h d f• Systematic experiments (thousands of runs) infeasible with biological agents
7
Individual Robot BehaviorIndividual Robot Behavior
pjoin(Environment)
Search Rest
pleave(Environment)
Note: avoidance behavior for walls implemented but neither represented here nor captured by the models (impact negligible) 8
AssumptionsAssumptions• A robot moves through the environmentA robot moves through the environment
(random walk) during which it encounters other robots with constant probabilityp y
• The probability to encounter one robot is pc, the probability to encounter and join a cluster p y jof n robots pjoin(n) ; the probability of leaving such cluster pleave(n)
Uniform distribution of objects in the environment and linear super-position of p pencountering probabilities
10
Probabilistic Finite State Machinef I di id l R bfor an Individual Robot
Robot joining/leaving aggregatesgg g
“Passive” state transitions
b bilit ti t t t th b t
transitions
pc = probability per time step to encounter another robotNx = fraction of time spent in state x 11
“Passive” State TransitionsPassive State Transitions
j j 1j j+1
Example: 4 robots change their state without actually moving
12
Probabilistic Finite State Machine f th Wh l S tfor the Whole System
Ns= average number of robots in searchNj= average number of robots in an aggregate of size j 13
Number of Agents in a j-size Cluster
• Flows from a single aggregate
jpcpjoin(j) jpcpjoin(j-1)
j-Aggregate
jpleave(j) jpleave(j+1)jpleave(j) jpleave(j+1)
15
Average Number of Agents in a j-size Cluster
j iNs(k)Nj(k)jpcpjoin(j) Ns(k)Nj-1(k) jpcpjoin(j-1)
j-Aggregate
Nj(k)jpleave(j)Nj+1(k)jpleave(j+1)
16
Temporal Evolution of the Self-A i P (12 b )Aggregation Process (12 robots)
Submicroscopic model (Webots)• 1500 runs
Macroscopic model
• 3 h (10800 s) per run • sample time 10 s•10 cm communication range 18
Modulating the Encountering ProbabilityModulating the Encountering Probability by the Communication Range (12 robots)
• 7cm communication range
• 1500 runs• 12 cm communication range
• 1500 runs 20
DiscussionDiscussion
• The “encountering probability” is indeed a function of the communication range, which leads to a bifurcation of the system dynamics (scattered vs. giant component)
• Aggregation performance of the biological algorithm very low
• Might be the sole choice for micro-robots
21
DiscussionDiscussion
• Modeling limitations:– Clusters are assumed to be circular– Only one individual joins/leaves at a time– Clusters do not split/merge
• Function of robot shape/sensorial characteristics
22
The LEURRE Project
• Leurre: European project focusing on mixed insect-robot
The LEURRE Project
Leurre: European project focusing on mixed insect robot societies (http://leurre.ulb.ac.be)
• Relevant Leurre partners for the presented work:
• ULB (coordinator modeling experimentalULB (coordinator, modeling, experimental work with insects/robots)
• Uni Rennes/CNRS (chemistry)EPFL ASL/LSRO ( bil b t d i• EPFL – ASL/LSRO (mobile robots design and development)
• EPFL – DISAL (vision-based tracking and d lli )modelling)
24
Collectively Selecting a ShelterCollectively Selecting a ShelterGoal of the project: quantitatively characterize the self-organized p j q y gcollective decision-making process of a cockroach society by unravelling and influencing the local interaction rules
• A simple decision-making scenario: 1 arena, 2 shelters
• Shelters of the same and different darkness
• Groups of pure cockroaches (16)Groups of pure cockroaches (16), mixed robot+cockroaches (12+4)
• Infiltration using chemical
[Halloy et al., Science, Nov. 2007]
camouflage and statistical behavioral model 25
Sample ResultsSample ResultsLegend: mixed, pure; shelter 1, shelter 2
A . Symmetric shelters (infiltration) B. Asymmetric shelters (active control)Experiment
Macroscopic Model
Experiment
Macroscopic Model
26
The EPFL Tool Contributions3. Multi-level modeling
( 1) ( ) ( ) ( 1) ( ) ( ) ( )join joinN k N k p N k j p j N k p j N k Macroscopic
1. Robots as a flexible, interactive, societal “microscope”
Ss Sa
1
1
( 1) ( ) ( ) ( 1) ( ) ( ) ( )
( ) ( ) ( 1) ( )j j r s j j
leave leavej j
N k N k p N k j p j N k p j N k
p j N k j p j N k j
Agent-based microscopic
Macroscopic model
Ss SaS S
Module-based
pmodel
Ss Sa
[Asadpour et al., ARS J., 2006]
microscopic model2. A marker-less vision-based tracking
tool: SwisTrack
Target system
[Correll et al., ALife J., in prep.][Correll et al., IROS 2006, Lochmatter et al., IROS 2008]; http://sourceforge.net/projects/swistrack 27
Wh t i S lf A bl ?What is Self-Assembly?Spontaneous organization of pre-existing units
into spatial structures or patterns without external guidance.
Self-Assembly at
29 orders of magnitude
Self-Assembly at All Scales
29 orders of magnitude, same underlying
mechanism!
Our galaxy, the Milky Way, about 1021 m
Carbon nanotube, about 10-8 m 29
Wh t i NOT S lf A bl ?What is NOT Self-Assembly?Planned and externally
Living beings, because they are not made of
guided constructions pre-existing unitsUnstructured aggregates
Gazprom building, St-Petersburg, 102 m
Boxelder bug aggregate, 10-3 m
Various animals
Sequential, limited scalability (from a few
illi t t f
Aggregation may serve a purpose (e.g., protection,
ll ti d i i
Self-assembly may have played a key role in the
i iti l f lifmillimeters up to a few kilometers)
collective decision-making), but the structure
itself does not.
initial appearance of life.
30
Self-Assembly at All Scales
D SA f 1 i d h l l t i
A.Crystal structure of a ribosome [Yusupov et al 2001]
D. SA of 1-mm-sized hexagonal plates using “capillary bond” [Whitesides, Harvard]
E. SA of electrical networks [Whitesides, Harvard]al., 2001]
B.Scaffolded DNA origami of 100 nm in diameter [Rothemund, Caltech]
C.Capillary SA of 10-μm-sized hexagonal
]F. Electromechanical programmable parts
specifically designed for SA [Klavins, University of Washington]
G A swarm of mobile robots capable of SAplates into large crystal structures [Whitesides, Harvard]
G. A swarm of mobile robots capable of SA [Dorigo, UniversitéLibre de Bruxelles]
31
Ingredients and Features Ingredients of Self-Assembly:
1. Positive feedback: binding force, structural reinforcement2 N ti f db k l i f ( th i l2. Negative feedback: repulsive force (otherwise, merely
aggregation)3. Multiple reversible interactions4. Randomness
Features of Self-Assembly: Robustness: self-assembly deals with component defects,
environmental noise and external perturbations Parallelism: self-assembly occurs wherever there are interacting y g
components Scalability: self-assembly is a strategy applicable at all scales,
and even across scales Ubiquity: structures form spontaneously at all scales in Nature,
and most of them exhibit similarities across scales 32
M ti ti f St d i S lf A bl (1)Motivation for Studying Self-Assembly (1)Scientific understanding of a broad variety of
phenomena:phenomena:
Water ripples~10-3 m
Seashell~10-2 m
School of fish~101m
DNA origami~10-7m
Tornado~102 m
Cyclone~106 m~10 7m ~102 m ~106 m
33
Motivation for Studying Self-Assembly (2)Motivation for Studying Self-Assembly (2)Manufacturing and coordination of ultra-small
distributed intelligent systems:distributed intelligent systems:
Left: A microrobot (10 μm) is inoculating a red blood cellagainst someLeft: A microrobot (10 μm) is inoculating a red blood cellagainst some kind of disease. Right: This nanoscaled device (1 nm) enables a nanorobot to catch HIV individuals and destroy them. 34
The quest for miniaturizationHow to go from the Alice robot (2cm in size)…
...to fully-fledged microrobots?35
Limitations of the Top-Down Approach
HOW TO ASSEMBLE
INTO THESE USING THESE?THESE
ANSWER: LET THEM ASSEMBLE THEMSELVES!36
Self-Assembly of Micro-Systems
bi di fBinding surface
Non-binding surface
Self-AssemblySelf-Assembly Hydrophobic interaction and
capillary forces (in fluid)
V d W l f (i d Van der Waals forces (in dry state)
Experimental challenges: imaging (3D high speedimaging (3D, high speed camera, microscope) and controllability
37
Intelligent… and less intelligent agentsAli bil b t MEMSAlice mobile robot MEMS
100µm2cm
Typical size: 2 centimeters Typical size: 50 to 500 μm
100µm2cm
Typical swarm size:< 100 units Sensing, computation,
communication (only local)
Typical swarm size:> 104 unitsNo sensing, computation or
communication, but localcommunication (only local)Controllable self-locomoted
units (but quite noisy)
communication, but local interactions only Stochastic motion only 38
Multi-Level Modeling MethodologyMacroscopic 1: rate equations mean field
stra
ctio
n
met
rics
Macroscopic 2: Chemical Reaction Network, t h ti i l ti
Macroscopic 1: rate equations, mean field approach, whole population
Microscopic 1: Monte Carlo model, 1 robot = 1 agent non-spatial
Abs
Com
mon
stochastic simulations
agent, non spatial
Microscopic 2: Agent-Based model, 1 robot = 1 agent spatial
ime
agent, spatial
Realistic: faithful representation of the robots (e g geometry S&A) and the environment (e g
rim
enta
l ti(e.g., geometry, S&A) and the environment (e.g.,
friction, gravity, inertia)
Physical reality: real experiments, imaging with overhead cameras, tracking software E
xpe
39
Multi-Level Modeling MethodologyMacroscopic 1: rate equations mean field
stra
ctio
n
met
rics
Macroscopic 2: Chemical Reaction Network, t h ti i l ti
Macroscopic 1: rate equations, mean field approach, whole population
Microscopic 1: Monte Carlo model, 1 robot = 1 agent non-spatial
Abs
Com
mon
stochastic simulations
agent, non spatial
Microscopic 2: Agent-Based model, 1 robot = 1 agent spatial
ime
agent, spatial
rim
enta
l ti
Underlying physical reality (and, therefore
submicroscopic models) may
Exp
esubmicroscopic models) may vary dramatically!
40
Ex.1: Self-Assembly of Alice RobotsG lGoal Formation of chains
Ph i l t tb d f d l t Physical testbed for models at multiple abstraction levels
Self assemblySelf-assembly “soft” binding force (implemented through local IR ( p gcommunication)
AdvantagesControllable, but still miniaturized 19 robots, accelerated 16x
Experimental ease: imaging, controllability, flexibility [Evans et al., ICRA 2010] 41
Single Robot AlgorithmTumble
Timer expiresTimer expires
OR
Start Wander
Unaligned object detected
Timer expiresAligned object
detected
AvoidBondNo recent communication
ORNo recently aligned object
Deterministic vs. stochastic variant42
Deterministic ControllerBonded Alices communicate with each other to determine the length of their chaindetermine the length of their chain
msg = 4
msg = msg + 1msg = 2
msg = 1
msg = 4
43
Deterministic Controller -Deterministic Controller Sample Results with 19 robots
Submicroscopic Macroscopic (Webots) (stochastic CRN)
44
Stochastic ControllerStochastic Controller
Every second, an Alice leaves its current bond depending on some
b bilitprobability.
45
Stochastic Controller - SampleStochastic Controller Sample Results with 19 robots
Submicroscopic MacroscopicSubmicroscopic (Webots)
Macroscopic (stochastic CRN)
46
Multi-Level ModelingMacro deterministic: rate equations
trac
tion
met
rics
Macro-stochastic: Chemical Reaction
Macro-deterministic: rate equations, mean field approach, whole population
Abs
tC
omm
on Network, stochastic simulations
Microscopic (non-spatial): Monte Carlo model 1 particle = 1 agent C
Microscopic (spatial): Agent-Based model 1 particle = 1 agent
model, 1 particle = 1 agentAUTOMATIONAUTOMATION
l tim
e
model, 1 particle = 1 agent
Underlying physical platforms
erim
enta
ly g p y pand submicroscopic models may
vary dramatically.
Exp
e
47
Alice Self-AssemblyReal System
(tracking)
M3 FrameworkM3 Framework
✗
1010 runs452 species
1474 reactions
[Mermoud, PhD EPFL 2012, STAR Volume Nr. 93] 48
Ex. 2: Self-Assembly of Lilies
40 runs6 species
12 reactions
mean
1.96·σ
[Di Mario et al., IROS 2011][Mermoud, PhD EPFL 2012]
49
Automated Model-Based Design
ModelingAUTOMATIONAUTOMATION
Model-basedDesign & Optimization
[Mermoud et al, ICRA 2012] 50
Model-Based Control of Self-Assembly Case study with four real Lily modules Two modes of agitation (bang-bang):
1 High agitation (exploration)1. High agitation (exploration)2. Low agitation (exploitation)
Simultaneous modeling & optimizationhapproach
Completely automated and real-time
Target 51
Quantitative results
ure
[s]
et st
ruct
uac
h ta
rge
me
to re
a
(low) (high)
Tim
40 runs of 30 minutes using 4 real Lily modules53
Ex 3 SelfSys: MEMS Self AssemblyEx. 3 SelfSys: MEMS Self-Assembly(a) (b)
100 µm
10001000 runs2 species
2 reactions
100 building blocks10 building blocks[Mermoud, PhD EPFL 2012][Jacot-Decombes et al, Soft Matter, 2013] 54
SelfSys: Experimental Set-Up
2 cm2 cm2 cm[Goldowsky et al., JMM 2013][Mastrangeli et al., ICRA 2014, submitted] 55
AiSelfSys: Modeling and Agitation
Air socketsCFD simulations:
• Laminar flow (Re ~ 100)2 h fl ( t / i )• 2-phase flow (water/air)
• Acoustic actuation (45 KHz)
water5 mm
water
[Goldowsky et al., JMM 2013]57
SelfSys: Agitation Control Modes
10 Control space
y g
10 mm Control space
1st: 42 4 KHz
FEM Eigenfrequencies
1st: 42.4 KHz
[Goldowsky et al., JMM 2013] 58
SelfSys: Multi Modal Control of theSelfSys: Multi-Modal Control of the Agitation in the Microchamberg
[Goldowsky et al., JMM 2013] 59
Take Home MessagesTake Home Messages• Self-aggregation can be modeled with the usual multi-level
modeling techniques; the resulting models at the macroscopic g q g plevel are very similar to those of object aggregation
• Self-aggregation can facilitate collective decisions, especially in systems endowed with low-range communicationin systems endowed with low-range communication
• Self-assembly implies the constitution of structures and ordered spatial pattern (rather than just aggregate) and
i i l h i ll h i dconstituting elements that incorporate all the required assembly mechanisms (although they are not necessarily endowed with self-locomotion capabilities)
• Self-assembly can happen at all scales and is a powerful, parallel, potentially cheap coordination principle for a number of manufacturing applications in which traditionalnumber of manufacturing applications in which traditional techniques cannot be used or have an excessive cost
62
Additional Literature – Week 13Papers• S. Garnier, C. Jost, R. Jeanson, J. Gautrais, A. Grimal, M. Asadpour, G. Caprari,
and G. Theraulaz. “The Embodiment of Cockroach Aggregation Behavior in a Group of Micro-Robots”. Artificial Life, 14: 387-408, 2008.
• R. Gross, M. Bonani, F. Mondada, and M. Dorigo. Autonomous self-assembly in swarm-bots. IEEE Trans. on Robotics, 22(6):1115–1130, 2006.
• Gross R and Dorigo M "Self-assembly at the macroscopic scale " Proceedings of• Gross R. and Dorigo M. Self-assembly at the macroscopic scale, Proceedings of the IEEE, 96(9):1490-1508, 2008
• Goldowsky J., Mastrangeli M., Jacot-Descombes L., Gullo M. R., Mermoud G., Brugger J., Martinoli A., Nelson B. J., Knapp H. F., “Acousto-Fluidic System A i ti I Li id S lf A bl f Mi t ” J l fAssisting In-Liquid Self-Assembly of Microcomponents”. Journal of Micromechanics and Microengineering, 23: 125026 (11pp), 2013.
• Jacot-Descombes L., Martin-Olmos C., Gullo M. R., Cadarso V. J., Mermoud G., Villanueva L. G., Mastrangeli M., Martinoli A., and Brugger J., “Fluid-Mediated Parallel Self-Assembly of Polymeric Micro-Capsules for Liquid Encapsulation and Release”. Soft Matter, 9: 9931-9938, 2013.
• Mastrangeli M., Mermoud G., and Martinoli A., “Modeling Self-Assembly Across Scales: The Unifying Perspective of Smart Minimal Particles”. Special issue on y g p pSelf-Assembly, Böhringer K. F. and Elwenspoek M. editors, Micromachines, 2(2): 82-115, 2011.
• Klavins E., “Programmable self-assembly”. Control Systems, 27(4):43–56,2007. 63