distributed intelligent systems · • ud dh dld i l l dunderstand how to model, design, control,...
TRANSCRIPT
![Page 1: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/1.jpg)
Distributed Intelligent Systems
EPFL WS 2013-2014EPFL, WS 2013-2014
http://disal.epfl.ch/teaching/distributed_intelligent_systems/
1
![Page 2: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/2.jpg)
Distributed Intelligent Systems – W1Part I: Course Organization,
Team and ContentTeam, and Content
2
![Page 3: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/3.jpg)
Access to e-material exercise andAccess to e material, exercise, and building room
• Web site moodle for the course: need to register also there (using the pass g ( g pcommunicated by head TA the first time); this will not be public access since a lot of pcopyrighted material
• For non-SIE students we need to request forFor non SIE students we need to request for you explicit access rights for the computer room and building: please by Fri at latest toroom and building: please by Fri at latest to IS-academia and moodle to get access
3
![Page 4: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/4.jpg)
R ti l f Thi CRationale for This Course
4
![Page 5: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/5.jpg)
Rationale• Well-balanced course: theory, algorithms, experimental
labs and projects• Understand quantitatively natural collective phenomena
(focus on biological societies) and how to combine bio-inspired principles with advanced engineering methodsU d d h d l d i l l d• Understand how to model, design, control, evaluate, and optimize distributed intelligent systems(f d f lt i ith j t) L i t k i• (for default version, with project) Learning to work in an interdisciplinary team under time pressure, cross-review and defend projects and dig out relevantreview and defend projects, and dig out relevant literature
5
![Page 6: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/6.jpg)
What is this Course aboutWhat is this Course about
• Distributed natural and artificial systems• Distributed natural and artificial systems • Coordination algorithms• Distributed sensing and action• Models, simulation tools, and machine-, ,
learning targeted to distributed intelligent systemsy
• Multi-robot systems, wireless sensor networks intelligent transportation systemsnetworks, intelligent transportation systems
6
![Page 7: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/7.jpg)
Course PrerequisitesC d M tl b k l d• C and Matlab knowledge
• Fundamentals of programmingF d l f b bili l l• Fundamentals of probability calculus
• Fundamentals of analysis (differential equations, ti d di t ti )continuous and discrete time)
• Fundamentals of linear algebraF d l i i l d• Fundamentals in signal and systems
For SIE students: BS introductive course on SignalsFor SIE students: BS introductive course on Signals, Instruments, and Systems highly recommended! 7
![Page 8: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/8.jpg)
Organization of the Course
8
![Page 9: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/9.jpg)
This ear again t o ersionsThis year again two versions• Default version (with course project) for all• Default version (with course project) for all
the students but SMT
• Light version (without course project) for all SMT students
• SMT students interested to take the default version have to enroll in a minor supportingversion have to enroll in a minor supporting such version (e.g., biocomputing, CSE)
9
![Page 10: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/10.jpg)
Credits and Workload –Credits and Workload With Project (Default)
• 6 ECTS• 1 ECTS = 30 h workload → 180 h workload1 ECTS 30 h workload → 180 h workload• Rough breakdown
66 h l t (i l di di d )– 66 h lecture (including reading and exam prep)– 54 h exercise (including preparation + reporting)
h j (i l di di– 60 h course project (including reading, implementation, defending, reviewing, and reporting)
10
![Page 11: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/11.jpg)
Grade - With Project (Default)Grade With Project (Default)
• Final written exam, winter session:,– 180 minutes; – open book with simple non-programmable calculator; – all topics covered in the lecture/exercise and selected
distributed reading material• 60% performance during semester 40%60% performance during semester, 40%
performance during the exam (compromise US/Europe style)
• During semester: 30% performance of lab+hwk(average 3 homework series); 30% performance project (report + presentation)project (report + presentation)
11
![Page 12: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/12.jpg)
Credits and Workload –Credits and Workload Without Project
• 4 ECTS• 1 ECTS = 30 h workload → 120 h workload1 ECTS 30 h workload → 120 h workload• Rough breakdown
66 h l t (i l di di d )– 66 h lecture (including reading and exam prep)– 54 h exercise (including preparation + reporting)
12
![Page 13: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/13.jpg)
Grade - Without ProjectGrade Without Project
• Final written exam, winter session:,– 180 minutes; – open book with simple, non-programmable calculator; – all topics covered in the lecture/exercise and selected
distributed reading material• 45% performance during semester 55%45% performance during semester, 55%
performance during the exam (compromise US/Europe style)
• During semester: 45% performance of lab+hwk(average 3 homework series);
13
![Page 14: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/14.jpg)
Lecture
• Start 8:30 because of family logistic, 10 min breaks, finish 10:10, start a bit later exercise , ,(10:20)
• 2 (default) or 3 h lecture (when2 (default) or 3 h lecture (when appropriate), today exceptionally 4 h
• 1 or 2 regular breaks (today 3 breaks)• 1 or 2 regular breaks (today 3 breaks)
14
![Page 15: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/15.jpg)
Lecture Notes & ReadingLecture Notes & Reading• Policy: master, research-oriented course → no manuscript!
slides + papers + web→ slides + papers + web• Roughly 60 pages/week to read; list subject to change
during the semester as a function of the lectureduring the semester, as a function of the lecture• Preliminary lecture slides in pdf format available for
download on the course web site possibly before eachdownload on the course web site possibly before each lecture (tue late evening), definitive ones after lecture by Friday at latest)y )
• Reading distributed 1 week before (also because useful for exercise preparation)
• List of additional reading for each topic provided in the lecture notes 15
![Page 16: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/16.jpg)
Suggestions for a Successful C M i l P iCourse Material Processing
From last years experience:• For high-gear courses such as this one with a lot of
i l d di i h kiraw material to process and digest is worth taking advantage of the lecture for having an idea about what’s important and what notwhat s important and what not
• Trained ability: reading what’s needed and quickly
16
![Page 17: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/17.jpg)
Labs and Hwk • Lab session: 3 h (default) or 2 h (when appropriate) after
the lecture until 13:00 on Wed, GR B0 01 and GR C0 02the lecture until 13:00 on Wed, GR B0 01 and GR C0 02• Mini-tutorial (< 10 min) by the main lab designer at the
beginning of the labg g• 9 lab sets total, not graded (solution distributed)• 3 hwk assignments, graded (personalized feedback)3 hwk assignments, graded (personalized feedback)• Download/upload via moodle server (see also student
area on the course web site). )• [email protected] for ANY issue!
17
![Page 18: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/18.jpg)
Graded Homework • Lecture and exercise material over 3 weeks for
each hwk assignmentg• Deadline at noon, 2 weeks time to get write-
ups done; no late submission accepted unlessups done; no late submission accepted unless motivated by provable health problems
• Penalty for late submission• Penalty for late submission• Specific planifiable issues in attending a lab
i / b itti it h ld bsession/submitting a write-up should be mentioned in advance to the TAs
• Network/moodle server problems → TAs immediately notified 18
![Page 19: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/19.jpg)
Suggestions for a Successful Exercise SeriesExercise Series
From last years experience:d h l b i i d i hi ill• Read the lab assignments in advance, in this way you will
be more efficient when the TAs are around for helping you on the toughest questionson the toughest questions …
• Have an idea of the point distribution of any assignment: this roughly correspond to the breakdown in time youthis roughly correspond to the breakdown in time you should have; if your time is tight invest where it is worth!
• Take lab notes so that you will find them for the hwkTake lab notes so that you will find them for the hwkassignment
• Effort for the hwk is calculated based on acquired qknowledge during the labs; if you do not work during labs, it will take much longer to solve the hwk 19
![Page 20: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/20.jpg)
Course Project (1)• Course project list distributed in W6• 60 h effort, from W8 (final allocation) to W14
(report submission)• W10 practical kick-off tutorial W 0 p ac ca c o u o a• Project workplan (compulsory but not graded,
1 page max) and debriefing with responsible1 page max) and debriefing with responsible TA/staff member in W10Team of 3 st dents from at least 2 different• Team of 3 students from at least 2 different sections (default) or 2 students (if needed);
ill di ib h / h• Will distribute hw/sw at home20
![Page 21: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/21.jpg)
Course Project (2)• Final presentation in front of the class • Each of the project will have another team ofEach of the project will have another team of
students as reviewers• Each of the team members has to present• Each of the team members has to present• Project report to be submitted (max # of pages
d f bli h d) l k f hand format pre-established) last week of the semester
• Project defense (1 h presence in a specific topic session); possible dates: Dec 20, Jan 10 or 1 day (or 2 half days) in the exam session. Preferences? 21
![Page 22: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/22.jpg)
Suggestions for a Successful Course ProjectSuccessful Course Project
From last years experience:• take advantage of the first 7 weeks for asking• take advantage of the first 7 weeks for asking
question to TAs/staff about projects (not all topics will be covered by then), checking previous lecture y ), g pweb site for topics, and choosing project partners
• Contact ASAP the responsible supervisor when the projects will have been assigned for getting started
• Plan your effort (milestone, time, etc.), coordinate your team
• Arrive at the workplan milestone with a clear implementation plan, background reading and understanding of the objectives behind you 22
![Page 23: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/23.jpg)
Collaboration PolicyCollaboration Policy• Lecture and exam preparation: encouragedp p g• Project: encouraged (team up)• Lab: discussion encouraged but workLab: discussion encouraged but work
individually• Hwk: write-up submitted individually!Hwk: write up submitted individually!
Good training for final exam! We will be very strict in penalizing too similar write-y p gups copies because a substantial amount of the grade is acquired with hwk (n copies = #
f i / )of points/n)!23
![Page 24: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/24.jpg)
Team beyond this Coursehttp://disal.epfl.ch
• Instructor: Alcherio Martinoli• Guest lecturers: Ezequiel Di Mario, William C. Evans, Iñaki Navarro• Teaching assistants:
– José Nuno Pereira (head TA, PhD student)José Nuno Pereira (head TA, PhD student) – Bahar Haghighat (TA, PhD student)– Jorge Soares (TA, PhD student)– Miloš Vasić (TA PhD student)Miloš Vasić (TA, PhD student)
•Support staff:– Ezequiel Di Mario (PhD student)– William C Evans (PhD student)– William C. Evans (PhD student)– Ali Marjovi (postdoc)– Iñaki Navarro (postdoc)
Gillian Milani (SIE master student DIS alumnus)– Gillian Milani (SIE master student, DIS alumnus)– Lorenzo Sarti (SIN exchange master student, DIS alumnus)
24
![Page 25: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/25.jpg)
Course Syllabus and Course Syllabus and SummarySummary
25
![Page 26: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/26.jpg)
Goal
• Course overview• Course flavorCourse flavor• 4 main blocks
A f lid bl k• A few slides per block
26
![Page 27: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/27.jpg)
Block I – Swarm Intelligenceg
• Key Principles of Swarm Intelligence• Key Principles of Swarm Intelligence• Trail laying/following mechanisms
A t C l O ti i ti • Ant Colony Optimization as an example of a successful multi-agent
t h i timetaheuristic
27
![Page 28: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/28.jpg)
From Natural to Artificial Systems• Modeling to understand microscopic to
From Natural to Artificial Systemsg p
macroscopic transformation• Modeling as interface to artificial systems
Modeling
Individual behaviors Global structuresIdeas forIndividual behaviors and local interactions
Global structuresand collective
decisionsartificialsystems
28
![Page 29: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/29.jpg)
Biological Mechanisms and ModelsBiological Mechanisms and Models
Choice occurs randomly
(Deneubourg et al., 1990)
29
![Page 30: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/30.jpg)
The Traveling Salesman Problem10 7
8
9
3
10 7
8
9
3
10 7
8
9
10 7
8
9
g
1
24 6
3
1
24 6
3
1
24 6
3
1
24 6
3
Graph (N,E)N : set of cities (nodes)
551
51
5
E : set of connecting roads (links)dij : distance between city i and j
Problem: Find the shortest path which allow the salesman to visit once and only once each city in the graphDifficulty: NP hard problem; time for computing the shortest routeDifficulty: NP-hard problem; time for computing the shortest route grows in a nonpolinomial way with the number of cities in the network -> metaheuristics provide near-optimal solutions! 30
![Page 31: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/31.jpg)
Block II – Individual Nodes
• Introduction to mobile robotics• Robotic tools (simulation and real HW)• Basic control architectures• Machine-learning for control shaping
(single-robot examples)(single robot examples)
31
![Page 32: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/32.jpg)
Real and Simulated e-puck• For course project and labs• Multi-robot operation ok
• Webots realistic robotic simulator
• No manipulation, no highly accurate odometry
• Discrete sensor and actuators• Single and multi-robot simulator
32
![Page 33: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/33.jpg)
Perception-to-Action LoopR i ( li• Reactive (e.g., nonlinear transform, single loop)
• Reactive + memory (e.g. filter, state variable, multi-loops)
• sensors • actuators
n
, p )• Deliberative (e.g. planning,
multi-loops)
Computation
Perc
eptio
n
Act
ion
P
Environment
33
![Page 34: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/34.jpg)
Shaping Robot Controllers
Note:Controller architecture can be of any type but worth usingof any type but worth using GA/PSO if the number of parameters to be tuned is importantp
34
![Page 35: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/35.jpg)
Robot Localization ?• Key task for:• Path planning
?• Mapping• Referencing
di i• Coordination
• Type of localizationN 46° 31’ 13’’E 6 ° 34’ 04’’• Type of localization
• Absolute coordinates• Local coordinatesLocal coordinates• Topological information
35
![Page 36: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/36.jpg)
Block III – Distributed Systems
• Collective movements (flocking, formation traffic)formation, traffic)
• Machine-learning for multi-unit tsystems
• Multi-level modelingDi i i f l b d t k ll ti• Division of labor and task allocation
36
![Page 37: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/37.jpg)
Ex. of Collective MovementsEx. of Collective Movements
37
![Page 38: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/38.jpg)
Reynolds’Rules for Flocking1. Separation: avoid collisions with nearby flockmates
y g
2. Alignment: attempt to match velocity (speed and direction) with nearby flockmates
3 C h i tt t t t l t b fl k t
Position control Position controlVelocity control
3. Cohesion: attempt to stay close to nearby flockmates
separation alignment cohesion38
![Page 39: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/39.jpg)
Implementation of Flocking Rules inImplementation of Flocking Rules in Artificial Embedded Agents
Realistic simulator (Webots)Real robots39
![Page 40: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/40.jpg)
Distributed Intelligent Transportation SystemsDistributed Intelligent Transportation Systems
• Multi-vehicle dynamic car simulator (Webots)• Noise-resistant automatic design methods • Driving assistance systems• Possible applications: intelligent transportation systems• [Zhang, 2001 – 2006, Caltech; Gowal 2008 –, Vasic
2012 –, EPFL]
40
![Page 41: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/41.jpg)
Co-Learning Collaborative BehaviorgThree orthogonal axes to consider (extremities or balanced solutions are possible):
• Individual and group fitnessPri ate (non sharing of parameters) and p blic (parameter sharing) policies• Private (non-sharing of parameters) and public (parameter sharing) policies
• Homogeneous vs. heterogeneous systems
Example with binary encoding of candidateencoding of candidate solutions
41
![Page 42: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/42.jpg)
Multi-Level Modeling Methodology
n
g gy
n n
nnn tNtnnWtNtnnW
dttdN )(),|()(),|()(
S S rics
trac
tion
Ss SaS S
Ss Sa
Microscopic : multi-agent models,
Macroscopic: rate equations, mean field approach, whole swarm
mon
met
r
Abs
t
me
Ss SaSs Sa only relevant robot feature captured, 1 agent = 1 robot
Com
m
ntal
timSubmicroscopic : intra-robot (e.g.,
S&A, transceiver) and environment (e.g., physics) details reproduced f ithf ll
peri
men
Target system (physical reality):
faithfully
Exp
info on controller, S&A, comunication, morphology and environmental features 42
![Page 43: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/43.jpg)
Market-Based CoordinationMarket Based CoordinationRobots simulate a market economy: Tasks, resources are commodities of measurable worth. ,
When robot performs task: gets paid for service it provided (+ $)gets paid for service it provided (+ $)pays for resources it consumed (- $)
Robots trade tasks and resources to maximize profit
Idea! pursuit of individual profit leads to efficient team p psolutions.
Robust fast handle complex tasks Robust, fast, handle complex tasks Can take advantage of centralized planning
43
![Page 44: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/44.jpg)
SBlock IV – Distributed Sensing and Actionand Action
S t k b ti • Sensor networks, robotic sensor networksC ll ti d i i d • Collective decisions and consensus
• Assembly and disassembly, aggregation and segregation
• Self-aggregation and self-assembling
44
![Page 45: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/45.jpg)
MICAz
• Atmel ATmega128L– 8 bit microprocessor, ~8MHz
k k– 128kB program memory, 4kB SRAM– 512kB external flash (data logger)
• Chipcon CC2420• Chipcon CC2420– 802.15.4 (Zigbee)
• 2 AA batteries– about 5 days active (15-20 mA)– about 20 years sleeping (15-20 µA)
• TinyOS45
![Page 46: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/46.jpg)
Wireless Sensor NetworksFeatures:- Very low sampling frequency < 1Hz- Very low power consumption: 25mW- Solar panel- Radio communicationRadio communication
Sensors:Air Temperature and Humidity- Air Temperature and Humidity
- Infrared Surface Temperature- Anemometer- Solar Radiation- Pluviometer- Soil moisture http://sensorscope.epfl.chSo o s u e- Soil pressure
At DISAL: [Bahr, Evans, 2009 -] 46
![Page 47: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/47.jpg)
Mixed Mobile and Static Nodes
• 100 e-pucks 60 mica-Z; audition100 e pucks, 60 mica Z; audition • Power-aware resource allocation • Distributed attention algorithms • Self-deployment, self-reconfiguration, self-gathering, interpolationp y , g , g g, p• [Cianci, 2004 – 2009; Bahr, Evans 2010 -]
47
![Page 48: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/48.jpg)
OpenSenseAir Pollution Monitoring
SENSING SYSTEMFrom many wireless, mobile,h li bl
INFORMATION SYSTEM… to reliable, understandable and
b ibl l iANO TER
Air Pollution Monitoring
heterogeneous, unreliable rawmeasurements …
Web‐accessible real‐time informationN
ARA
i i dinterpretation andpresentation of data
wirelessfixed nodes
mobile nodes
Internet
GPRSGPS
At DISAL [Arfire, 2010 -]electric vehicle(iON) 48
![Page 49: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/49.jpg)
Distributed Boundary CoverageDistributed Boundary Coverage• Case study: turbine inspection• Goal: complete sensor coverage• Goal: complete sensor coverage
of the turbine/compressor blades
• Technical challenges limit• Technical challenges limit possible designs of robotic sensors
• Test-bed: 40 Alice II• Test-bed: 40 Alice II • Could pave the way for similar
applications in coverage/inspection ofcoverage/inspection of engineered or natural, regularstructures with heavily constraints on robotic equipment
• [Correll, 2003-2007]49
![Page 50: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/50.jpg)
Distributed Odor Source Localization• Bio-inspired and information theoretical algorithms• Distributed control and sensing• Integration of anemometry, olfaction, and inter-robot g y, ,
localization capabilities• Wind tunnel and simulation experiments• Possible applications: environmental pollution,
search and rescue operations, humanitarian demining• [Lochmatter, 2005 - 2010]
50
![Page 51: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/51.jpg)
Corpse Aggregation in the Ant Messor Sancta
Reduction of the spread of infection? Chretien (1996)51
![Page 52: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/52.jpg)
Aggregation/segregation using RobotsAggregation/segregation using RobotsBecker et al, 1994 Holland & Melhuish, 1998
52
![Page 53: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/53.jpg)
Aggregation and Collective Decision in Mixed SocietiesCollective Decision in Mixed-Societies
• Given an delimited area containing two similar or dissimilar (i.e. one bright and one dark) shelters infiltrate using ( g ) gmechatronic lures the insect society and participate to the collective decision, voting for a natural or artificial solution
• Applications:Applications:– Low-stress animal management– Alternative pest control
R b t d d ith• Robot endowed with:– 1D Vision– Chemical camouflage– IR-based navigation and com modules
[Halloy et al, Science 2007]
53
![Page 54: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/54.jpg)
Self-AssemblingAlice mobile robot SU-8 microfabricated parts
2 300
Size: 2 centimeters Size: 50 to 500 μm
2cm 300µm
Typical swarm size: a few dozen units Sensing, computation, communication Controllable (but noisy) self‐
Typical swarm size: 103 to 106 units No sensing, no computation, no communication, but local interactions
locomoted units No self‐locomotion, stochastic motion
[Mermoud, 2006 - 2012] 54
![Page 55: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/55.jpg)
Conclusion
55
![Page 56: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/56.jpg)
Take Home MessagesTake Home Messages1. Course is rich and intensive; consider your overall ; y
semester load before enrolling2. Balanced theoretical contents and hands-on
experience; first two lab sessions give an idea of the workload
3. Course projects are aligned with course content; have a look to past course project on the web
4. Check previous editions, discuss with TAs if appropriate, to decide whether to definitively enroll in the coursein the course
56
![Page 57: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/57.jpg)
Distributed Intelligent Systems – W1Part II: An Introduction to
Swarm Intelligence Swarm Intelligence, Foraging Strategies in Ant Societies, and Ant-Inspired
MetaheuristicsMetaheuristics
57
![Page 58: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/58.jpg)
Outline• Swarm Intelligence
– A possible paradigm and motivationK i i l– Key principles
• Foraging StrategiesRecruitment based mechanisms– Recruitment-based mechanisms
– Inaccuracies of chemical communication
• Bridges experiments in the lab• Bridges experiments in the lab• Open space and multi-source experiments• Ant networksAnt networks• The Traveling Salesman Problem (TSP)• An Ant System (AS) for the TSPAn Ant System (AS) for the TSP
58
![Page 59: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/59.jpg)
An Introduction to Swarm I t lli M ti ti Intelligence – Motivation,
Definitions, and Key Definitions, and Key Principles
59
![Page 60: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/60.jpg)
Some natural collective h i l i l phenomena implying a close
interconnection among individuals g…
60
![Page 61: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/61.jpg)
All rights reserved61
![Page 62: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/62.jpg)
All rights reserved62
![Page 63: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/63.jpg)
All rights reserved 63
![Page 64: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/64.jpg)
© Guy Theraulaz, UPS, 1999
64
![Page 65: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/65.jpg)
Collective PhenomenaCollective Phenomena• Limited local information
Each individual in the group has access only to limited local information and has no global knowledge of the structure which it is engaged in constructing together with the other members of the groupg g p
• A set of simple individual rulesEach individual obeys a collection of a few simple behavioural rules. This rule set permits the group collectively to coordinate its activities and to build a global structure or configuration.
• The global structures which emerge accomplish some function• The global structures which emerge accomplish some functionThese structures often allow the group to solve problems. They are flexible(adapting easily to a novel environment), and they are robust, (if one or several individuals fail in their behaviour or make a simple mistake, the structures spontaneously re-form).
65
![Page 66: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/66.jpg)
F N t l t A tifi i l From Natural to Artificial Systems and more …y
66
![Page 67: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/67.jpg)
Collective/Swarm Intelligence?Some questions ariseSome questions arise ...
• How do animal societies manage to perform difficult tasks, indynamic and varied environments, without any externalguidance or control, and without central coordination?
• How can a large number of entities with only partialinformation about their environment solve problems?
• How can collective cognitive capacities emerge fromindividuals with limited cognitive capacities?g p
67
![Page 68: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/68.jpg)
From Natural to Artificial Systems• Modeling to understand microscopic to
From Natural to Artificial Systemsg p
macroscopic transformation• Modeling as interface to artificial systems
Modeling
Individual behaviors Global structuresIdeas forIndividual behaviors and local interactions
Global structuresand collective
decisionsartificialsystems
68
![Page 69: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/69.jpg)
Computational Swarm-Intelligent Systems
• In a virtual world, most of the mechanisms shown by natural SI can be easily y yreproduced
• Some of the mechanisms are intentionallySome of the mechanisms are intentionally modified and further are added in order to improve the performance of a givenimprove the performance of a given algorithm
69
![Page 70: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/70.jpg)
Physical Swarm-Intelligent Systemsy g y• Bio-inspiration
– social insect societies– flocking, shoaling in vertebrates
Beyond bio-inspiration: combine natural principles with engineering knowledge and technologies
• Unit coordination– fully distributed control (+ env. template)– individual autonomy
lf i i
g
(extend definition)– self-organization• Communication
– explicit/implicit local communication indirect communication through signs in the– indirect communication through signs in the environment (stigmergy)
• Scalability• RobustnessRobustness vs efficiency trade offRobustness
– redundancy– balance exploitation/exploration– individual simplicity
Robustness vs. efficiency trade-off
• System cost effectiveness– individual simplicity– mass production 70
![Page 71: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/71.jpg)
Some Definitions of Swarm IntelligenceSome Definitions of Swarm Intelligence• Beni and Wang (1989):
– Used the term in the context of cellular automata (based on cellular robots concept of Fukuda)
– Decentralized control, lack of synchronicity, simple and (quasi) identical members, self-organization
• Bonabeau, Dorigo and Theraulaz (1999)– Any attempt to design algorithms or distributed solvingAny attempt to design algorithms or distributed solving
devices inspired by the collective behavior of social insect colonies and other animal societies
• Beni (2004)Beni (2004)– Intelligent swarm = a group of non-intelligent robots
(“machines”) capable of universal computationUsual difficulties in defining the “intelligence” concept (non– Usual difficulties in defining the intelligence concept (non predictable order from disorder, creativity)
71
![Page 72: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/72.jpg)
Key Mechanismsybehind Natural Swarm
I t lliIntelligence
72
![Page 73: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/73.jpg)
Two Key MechanismsTwo Key Mechanisms in Natural Swarm-Intelligent Systems
1 Self-Organization1. Self-Organization
2. Stigmergy
73
![Page 74: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/74.jpg)
Self-OrganizationSelf Organization• Set of dynamical mechanisms whereby structure appears
at the global level as the result of interactions amongat the global level as the result of interactions among lower-level components
• The rules specifying the interactions among the system's constituent units are executed on the basis of purely local information without reference to the global patterninformation, without reference to the global pattern, which is an emergent property of the system rather than a property imposed upon the system by an external ordering i flinfluence
74
![Page 75: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/75.jpg)
Characteristics of Natural Self-Organized Systems
• Creation of spatio-temporal structures– E.g., foraging trails, nest architectures, clusters of objects, ...
• Multistability(i.e., possible co-existence of several stable states)– E.g., ants exploit only one of two identical food sources, build a
cluster in one of the many possible locations, ...
• Existence of bifurcations when some parametersExistence of bifurcations when some parameters change– E.g., termites move from a non-coordinated to a coordinated phase
l if h i d i i hi h h h h ld lonly if their density is higher than a threshold value
75
![Page 76: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/76.jpg)
Basic Ingredients ofBasic Ingredients of Natural Self-Organized Systems
• Multiple interactions• Randomness• Positive feedback
– E.g., recruitment, reinforcement• Negative feedbackNegative feedback
– E.g., limited number of available foragers, pheromone evaporationp p
76
![Page 77: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/77.jpg)
StigmergyGrassé P. P., 1959
• “La coordination des taches, la regulation des constructions nedependent pas directement des oeuvriers, mais des constructionselles-memes. L’ouvrier ne dirige pas son travail, il est guidé parlui. C’est à cette stimulation d’un type particulier que nous donnonsle nom du STIGMERGIE (stigma, piqure; ergon, travail, oeuvre =oeuvre stimulante).”
• [“The coordination of tasks and the regulation of constructions does not depend directly on the workers, but on the constructions themselves. The worker does not direct his work, but is guided by it. It is to this special form of stimulation that we give the name STIGMERGY (stigma, sting; ergon, work, product of labor = stimulating product of labor).”]
77
![Page 78: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/78.jpg)
StigmergyDefinition
It defines a class of mechanisms exploited by social insects to coordinate and control their activity via indirect interactions.
Answer R1 R2 R3 R4 R5
Stimulus S1 S2 S3
time
S 4 S 5 Stop
Stigmergic mechanisms can be classified in two different categories: • quantitative (or continuous) stigmergy • qualitative (or discrete) stigmergy
[Theraulaz & Bonabeau., Alife J. 1999] 78
![Page 79: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/79.jpg)
F i St t i i A tForaging Strategies in Ants
79
![Page 80: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/80.jpg)
Diff t A t Diff t Different Ants, Different Strategies
80
![Page 81: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/81.jpg)
Not All Foraging Strategies are Collective and based on Stigmergy …
l l hi d• Example: Cataglyphis desert ant• Excellent study by Prof. R. Wehner
(University of Zuerich Emeritus)(University of Zuerich, Emeritus)• Individual foraging strategy• Underlying mechanismsy g
– Internal compass (polarization of sun light) – Dead-reckoning (path integration on neural
chains for leg control)chains for leg control)– Local search (around 1-2 m from the nest)
• Extremely accurate navigation: d f f faveraged error of a few tens of cm over
500 m path!81
![Page 82: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/82.jpg)
More individual Foraging StrategiesIndividual navigation + learning capabilities for memorizing the foraging zone
82
![Page 83: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/83.jpg)
Tandem Recruitment Strategies- Mediated by thropallaxis, antennal contact- Based on food chemical signatures on the ant body
83
![Page 84: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/84.jpg)
Group Recruitment Strategies
84
![Page 85: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/85.jpg)
Mass Recruitment Strategies
85
![Page 86: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/86.jpg)
Mass Recruitment
86
![Page 87: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/87.jpg)
Behavior of Individual Ants
Sequence of actions performed by an ant communicating the discoveryof a food source
Food source Foraging area Nest
Stimulatingnest matesLaying a chemical trail
Picking upfood
D i i fDeposition offood
87
![Page 88: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/88.jpg)
F ti f R it t T il i A tFormation of Recruitment Trails in Ants
88
![Page 89: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/89.jpg)
Number of Ants at the Food Source vs. Time
Growing phase
S t ti
(positive feedback)
Saturation phase (negative feedback)feedback)
89
![Page 90: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/90.jpg)
Stochastic Individual Behavior Combined with the Amplification of Information can lead to Collective
DecisionsDecisions
90
![Page 91: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/91.jpg)
The Role of Randomnessin the Organization of Foraging
How does individual behavior with a strong stochasticcomponent lead to statistically predictable behavior at thel l f th l d ll ti d i i ?level of the colony and collective decisions?
© Guy Theraulaz91
![Page 92: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/92.jpg)
Experimental Strategy
• Most of the studies to assess quantitatively the role of randomness have been carried out in the lab because:– Controlled environmental conditions– Repeated runs for statistics
• Studies in the field can lead often only to qualitativeconclusion because they might be influenced by:conclusion because they might be influenced by:– Multiple food sources– Predators and competitorsp– Environmental changes (temperature, climate, etc.)
92
![Page 93: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/93.jpg)
E l ti Th IExploration: The Inaccuracyof Chemical Communication
93
![Page 94: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/94.jpg)
Termite Following a Pheromone TraceTermite Following a Pheromone Trace
Prof. J.-L. Deneubourg (ULB, Bruxelles) 94
![Page 95: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/95.jpg)
Ants can Reacquire a Trail by Local SearchAnts can Reacquire a Trail by Local Search
95
![Page 96: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/96.jpg)
Probability of Trail Losing
E l A f i f h fi i (V h h l 1980)
y gdepends on the Ant Species
Example: Accuracy of recruitment of the first recruit (Verhaeghe et al., 1980)
Tetramoriumimpurum
Tapinomaerraticum
Successfulrecruitments (%)
L th f t il
impurum erraticum
18 74
Length of trailfollowed (%) 17 68
Tapinoma follow trails much more reliably than Tetramorium → depends on the en ironment thedepends on the environment the species have evolved (food scattering, etc. ) 96
![Page 97: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/97.jpg)
Probability of Trail Losing i C Ti
• The longer the traveled
is Constant over TimeThe longer the traveled path and the smaller is the number of ants on th t ilthe trail
• Appears to be independent ofphenomena such as learning or sensory adaptive response (atadaptive response (at least under such short time scale)Log # of ants on the trail as a function
of the traveled path for a constant pheromone concentration 97
![Page 98: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/98.jpg)
Probability of Trail Losing dependsProbability of Trail Losing depends on Chemical Concentration
The higher is the gpheromone concentration and the more reliably can be followed a trail
Mean path length as aMean path length as a function of the pheromone concentration 98
![Page 99: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/99.jpg)
Biological Significance of the E ploitation E ploration Balance
Does the accuracy of the chemical communication channel used b i d h i ffi i ?
Exploitation-Exploration Balance
by ants increase or decrease their efficiency?
• Noise can have a certain flexible value for theorganization of the societyorganization of the society.
• The fact that a significant proportion of recruits get losten route can be of benefit when food is scattereden route can be of benefit when food is scatteredthroughout the environment or when several sources arepresent simultaneously.
• If too many ants get lost for a given food scattering theefficiency of recruitment also decreases.
Sacrifice a little bit efficiency in order to be robust at facing environmental unpredictability 99
![Page 100: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/100.jpg)
B id E i t Bridge Experiments: Selecting the Shortest PathSe ect g t e S o test at
100
![Page 101: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/101.jpg)
The Suspended, Symmetric B id E iBridge Experiment
Food source
Two branches (A and B) of the same length connect
Food source
the same length connect nest and food source
© J.-L. Deneubourgg
Nest 101
![Page 102: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/102.jpg)
Bridge with two Branchesf th S L thExperimental Resultsof the Same LengthExperimental Results
102
![Page 103: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/103.jpg)
Microscopic Model(Deneubourg 1990)
Probabilistic choice of an=
( k + A i ) nPA
( k + A i ) n + ( k + B i ) n= 1 - PB
Probabilistic choice of an agent at the bridge’s bifurcation points
A : number of ants having chosen branch A
PA and PB : probability for the ant i+1 to pick up the branch A or B respectively
A i : number of ants having chosen branch A
B i : number of ants having chosen branch B
n (model parameter): degree of nonlinearityk ( d l ) d f i f k d b hk (model parameter): degree of attraction of a unmarked branch
Aii PifB
PifBB
1[1
Aii PifA
PifAA
1[1
Ai PifB [
Ai PifA
iBA ii δ = uniform random variable on [0,1] 103
![Page 104: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/104.jpg)
Parameters of the Choice Function1
k=1 n=2
k=1 n=1
Parameters of the Choice Function 1
k=10 n=2
k=10 n=11.0 1.0
0,5
0,75 k=1 n=5
PP AAPP AA
0,5
0,75 k=10 n=5
PP AAPP AA0.5
0.75
0.5
0.75
0,25
,PPPP
0,25
k hi hk l0.25 0.25
01 10 100
01 10 100
# of ants on branch A
k highk low
# of ants on branch A100101 100101
• The higher is n and the faster is the selection of one of the branches (sharper curve); n high corresponds to high exploitationTh t k th hi h th tt ti it f k d b h d th f th hi h i th
# of ants on branch A # of ants on branch A
• The greater k, the higher the attractivity of a unmarked branch and therefore the higher is the probability of agents of making random choices (i.e. not based on pheromones concentration deposited by other ants); k high corresponds to high exploration 104
![Page 105: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/105.jpg)
Bridge with two Branchesf th S L th
Model vs. Experimentsof the Same Length
p100100
experiment Parameters that fit
80
90
80
90 model
on th
e nc
h
Parameters that fit experimental data:n = 2
60
70
60
70
pass
ages
na
nt b
ran k = 20
50
60
50
60
of a
nt p
dom
in
400 500 1000 1500 2000
400 500 1000 1500 2000
Total number of ants having
%
Total number of ants having traversed the bridge
105
![Page 106: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/106.jpg)
The Suspended, Asymmetric B id E iBridge Experiment
Food source• Two branches (A and B)
differing in their length (length ratio r) connect(length ratio r) connect nest and food source
• Test for the optimization biliti f tcapabilities of ants
© J.-L. Deneubourg
Nest106
![Page 107: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/107.jpg)
All Bridge ExperimentsAll Bridge Experiments4 different experimental scenarios
Sh t t b hShortest branchadded later
107
![Page 108: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/108.jpg)
Selection of the Shortest BranchRepeated experiments with different ant colonies of the same ant species (Linepithema Humile) – finite experimental time window
erim
ents
rim
ents
% o
f exp
e
% o
f exp
er
% of traffic on a given branch % of traffic on shorter branch
tsof
exp
erim
ent
expe
rim
ents
% of traffic on shorter branch
% o
% of traffic on shorter branch
% o
f
108
![Page 109: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/109.jpg)
Asymmetric Bridge – Ant S i DiffSpecies Differences
What happens if the shorter branch is presented after 30 minutes?
• Argentine Ants (Linepithema Humile) get stuck on the longer branch (mainly pheromone-based navigation), see previous slide.
• Lasius Niger ants find the shorter branch because they integrate other• Lasius Niger ants find the shorter branch because they integrate other navigation modalities (compass, dead-reckoning) with pheromone navigation -> U-Turns (different from random walk)!
• Pharaoh ants recognize the right way to go from geometry of trails (trails geometry provide polarity information!), again dead-reckoning/compass capabilities!
109
![Page 110: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/110.jpg)
Asymmetric Bridge –
Th i d l d t
Asymmetric Bridge Microscopic Modeling
• The previous model does not work any more: distance/traveling time has to be considered in order tobe considered in order to incorporate the geometry of the bridge.
• Multi-agent simulationMulti agent simulation incorporating pheromone deposition, avoidance rules, … point simulator (take into account trajectories but no body) by Prof. M. Dorigo (ULB Bruxelles).
© Marco Dorigo 110
![Page 111: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/111.jpg)
Foraging in Free Space
111
![Page 112: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/112.jpg)
Selecting the Richest SourceThree different experimental scenarios:
Experiment N°1
Experiment N°3
Experiment N°2
Experiment N 3
112
![Page 113: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/113.jpg)
Selecting the Richest SourcegResults obtained with Lasius Niger ants:
The ants might get stuck within their trail system and therefore the colonyexploits primarily the first food source that has been discovered even if thismight lead to neglecting a richer source which just appeared at a later time.Lasius niger: exclusively uses pheromone-based recruitment mechanisms;probably since nest-sources path not so misaligned u-turn strategy does nothelp in this scenario! 113
![Page 114: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/114.jpg)
Selecting the Richest Source –Selecting the Richest Source Scenario 3Results obtained with Tetramorium caespitum, Myrmica sabuletti
• These two ant species exploit mixed recruitment strategies: mass• These two ant species exploit mixed recruitment strategies: mass (trail laying/following) and group (no stigmergy) and do not get stuck in their trail network 114
![Page 115: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/115.jpg)
Mitchel Resnick MITResnick, MIT, Media Lab
115
![Page 116: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/116.jpg)
An Example with ThreeAn Example with Three Different Food Sources
• Different richness• Different distances from
nest• Obstacle-free environment
116
![Page 117: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/117.jpg)
Ant Networks
117
![Page 118: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/118.jpg)
Ant Super-Colonies
The organization of inter-nest traffic in antsg
F t i l i t th f d t l l i l it i th l• For most social insects, the fundamental ecological unit is the colony.
• In a number of ant species, groups of workers, larvae, andreproductives can leave the nest and set up a new nest whilep pmaintaining close connections with the parent nest.
• The collection of nests, or sub-colonies, forms what is called a super-lcolony.
118
![Page 119: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/119.jpg)
Super-colony of Formica Lugubris (Switzerland)Prof. D. Cherix (Uni Lausanne)
119
![Page 120: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/120.jpg)
The Organisation of Inter-Nest
R l f i l
Traffic in AntsResults for a triangular network (3 nest super-colony)with Linepithema y) phumile (Argentine ants)
a b c = % of traffic on brancha,b,c = % of traffic on branch a, b, or c
n = absolute number of
[Aron Deneubourg Goss Pasteels 1991]
passages
[Aron, Deneubourg, Goss, Pasteels, 1991]
120
![Page 121: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/121.jpg)
The Organisation of
R l f d l
Inter-Nest Traffic in AntsResults for a quadrangularnetwork (4 nest super-colony) with Linepithema humilep(Argentine ants)
a b c d = % of traffic ona,b,c,d = % of traffic on branch a, b, or c
n = absolute number of
[Aron Deneubourg Goss Pasteels 1991]
passages
[Aron, Deneubourg, Goss, Pasteels, 1991]
121
![Page 122: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/122.jpg)
Ants are Able to Optimize!Ants are Able to Optimize!• All the nests are connected either directly or
i di lindirectly• Ants are able to find the minimal spanning tree
connecting all the nests (probable ecological reasons: cost building and maintaining redundant spanning tree higher + extend predator exposure)
• This is similar to the Traveling Salesman Problem g(TSP)
• Can artificial ants solve the TSP?Can artificial ants solve the TSP? 122
![Page 123: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/123.jpg)
An introduction to Multi-Agent Systems based on Ant
Trail Laying/FollowingTrail Laying/FollowingMechanisms
123
![Page 124: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/124.jpg)
Motivation
• Ant Colony Optimization (ACO) algorithms as an example of successful transportation p pof ideas from natural systems to computational artificial systems (software p y (multi-agent systems)
• ACO algorithms as example of exploitationACO algorithms as example of exploitation of swarm intelligence principles as a particular form/instance of distributedparticular form/instance of distributed intelligence
124
![Page 125: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/125.jpg)
Th T li S l The Traveling Salesman Problemob e
125
![Page 126: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/126.jpg)
The Traveling Salesman Problem10 7
8
9
3
10 7
8
9
3
10 7
8
9
10 7
8
9
1
24 6
3
1
24 6
3
1
24 6
3
1
24 6
3
Graph (N,E)N : set of cities (nodes)E : set of connecting roads (links)
551
51
5
E : set of connecting roads (links)dij : distance between city i and j
Problem: Find the shortest path which allow the salesman to visitProblem: Find the shortest path which allow the salesman to visit once and only once each city in the graphDifficulty: NP-hard problem; time for computing the shortest route
i li i l ith th b f iti i thgrows in a nonpolinomial way with the number of cities in the network -> metaheuristics/machine-learning class (e.g., ACO, GA) provide near-optimal solutions! 126
![Page 127: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/127.jpg)
How Hard are NP-Hard Problems?TSP – Brute force
• A 30 city tour would have to measure the total distance of b 2 65 X 1032 diff t t A i t illibe 2.65 X 1032 different tours. Assuming a trillion additions per second, this would take 252,333,390,232,297 years.
• Adding one more city would cause the time to increase by a factor of 31.
QAP – exact algorithms (e.g. Bixius & Anstreicher 2001)• around 30+ max instances
d ( i i li i ) h• ex. 36 nodes (wiring application): 180h CPU on a 800 MHz Pentium III PC
• Same problem with ACO: 10 s on the same machineSame problem with ACO: 10 s on the same machine
127
![Page 128: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/128.jpg)
Artificial Ants andArtificial Ants and the Shortest Path Problem
Pheromone traildepositingdepositing
?
Probabilistic rule toh th th
Source
choose the path
DestinationDestination
128
![Page 129: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/129.jpg)
Problem!The extension of the real ant behavior (forward/backward trail deposit and slow pheromone decay rate) to artificial ants moving on
h d ’t ka graph doesn’t work: problem of self-reinforcing loops
Probabilistic solution generation plus pheromone update g p p p-> self-reinforcing loops
Source
Example of possibleExample of possibleself-reinforcing loop
Destination
129
![Page 130: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/130.jpg)
S l i !Solution!
Pheromone traildepositing Memorydepositing
?
Probabilistic rule toh th th
Source
choose the path
DestinationDestination
130
![Page 131: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/131.jpg)
Th Fi t ACO Al ithThe First ACO Algorithm:The Ant System (AS)e t Syste ( S)
131
![Page 132: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/132.jpg)
Design Choices for AS(Dorigo, Colorni, Maniezzo, 1991)
• Ants are given a memory of visited nodes• Ants build solutions probabilistically withoutAnts build solutions probabilistically without
updating pheromone trails (forwards ants)• Ants deterministically backward retrace the forward y
path to update pheromone (backwards ants)• Ants deposit a quantity of pheromone function of the p q y p
quality of the solution they generated• Pheromones evaporates much more quickly than in
nature132
![Page 133: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/133.jpg)
AS for TSP - OverviewAS for TSP Overview
b (t) (i = 1 n) : number of ants at the node i at the iteration tbi (t), (i 1 … n) : number of ants at the node i at the iteration t
i = 1i = 1
nnm = bi (t) = constant: total number of ants
133
![Page 134: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/134.jpg)
?AS for TSP- Individual Ant Behavior
am ?Memory of ant k: list
of visited nodesJi
k
b …i
dijdir
jrdik
kij
1
ikir
0 0,5
ij
The inverted value of the distance ηij = 1/dij between nodes i and j is called visibility; thisinformation (heuristic desirability) is static, i.e. not changed during the problem solution 134
![Page 135: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/135.jpg)
AS for TSP- Individual Ant Behavior
amb …
ii
jr
ij
ir
j ik
k
ij, quantity of virtual pheromone deposited on the link between the node i and j135
![Page 136: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/136.jpg)
Assumptions on TSPAssumptions on TSP• Usual assumption: fully connected graph (i.e. there is a direct p y g p (
route with a given distance from any city in the problem to any other); city list work ok
• Real problem: not fully connected; problem with city listp y ; p y• Possible solutions:
– Assume virtual routes so that fully connected; give very bad scores to ants choosing virtual routes (e.g., high but not infinite virtual distance; D i ’ ti )Dorigo’s suggestion)
– Alternative: break not valid tours asap and either relaunch a new ant or consider less ants for updating pheromones at the next iteration (Martinoli’s suggestion); computationally more efficient but risk to ( gg ); p ylose constructive aspect of trail laying/following; does not work for dead end edges with end criterion being at the start city
– Graph connectivity: full – dense – sparse; probably different solutions work better as a function of the connectivity degree; interestingwork better as a function of the connectivity degree; interesting problem
136
![Page 137: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/137.jpg)
AS for TSP - AlgorithmAS for TSP AlgorithmLoop \* t = 1*\
Place one ant on each node \*there are n = |N| nodes \*
For k := 1 to m \* each ant builds a tour, in this case m=n\*
For step := 1 to n \* each ant adds a node to its path \*For step := 1 to n \ each ant adds a node to its path \
Choose the next node to move by applying a probabilistic state transition rule
End-forEnd-forUpdate pheromone trailsUpdate pheromone trails
Until End_condition \*e.g., t = tmax *\
137
![Page 138: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/138.jpg)
AS for TSP – Transition RuleskkiJ : list of nodes still to be visited for ant k when it is at node i; starting from an
exhaustive list of all the cities in the problem, nodes get scratched during a tour T; at the beginning the list contains all nodes but i; also called tabu list
t it l t |N| t (N b f d i th bl ) i hi h thT : tour, it last n= |N| steps (N = number of nodes in the problem) in which the probabilistic transition rule below is applied: iteration index: number of times the whole algorithm is run; 1 ≤ t ≤ tmax
T
t
During a tour T, an ant k at the node i decided to move towards the nodej with the following probability (idea: roulette wheel):
kpij(t) = k
0 , if node j have been visited by ant k alreadybecause of tabu list
[ij(t)] [ij]
, if the node have not been visited yetpij(t) = k
l Ji
k
[il(t)] [il]
: parameter controlling the influence of the virtual pheromone : parameter controlling the influence of the local heuristic (visibility)
138
![Page 139: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/139.jpg)
AS for TSP – Virtual Pheromone Update
At the end of each tour T, each ant k deposits a quantity of virtual pheromoneon the link (i,j); pheromones sum upijij
kk
0 , when (i,j) has not been used during the tour Tij = k
Q h (i j) h b d d i th t T k Q , , when (i,j) has been used during the tour Tij = k
Lk(t)
Lk(t) = length of the tour T done by ant k at iteration tQ = parameter (adjusted by heuristic, not sensitive)
Note: the longer the tour, the lower is the quality of the solution, the smaller the quantity of pheromone dropped 139
![Page 140: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/140.jpg)
AS for TSP – Default Virtual Pheromone Update
m
)()()1()1( ttt ijijij
with ij(t) = k = 1
ijk
ρ = evaporation coefficientAt iteration t = 0 each link is initialized with a small homogeno s pheromone q antithomogenous pheromone quantity τ0
140
![Page 141: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/141.jpg)
AS for TSP – Virtual Pheromone Update with Elitism (EAS)
m
)()()()1()1( tettt eijijijij
with ij(t) = k = 1
ijk
LQteij /)(
0)( e
if (i,j) belongs to the best tour T+ out of the mtours generated by ants at a given iteration
0)( teij otherwise
e = parameter (adjusted by heuristic, not sensitive)p ( j y , )
Note: idea, best tours get extra reinforcement141
![Page 142: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/142.jpg)
AS for TSP – Evolution of the Best Tour Length
Example: 30 nodes problem
142
![Page 143: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/143.jpg)
AS for TSP – Results 50 citiesAS for TSP – Results 50 citiesExample of solution found on Eil50 problem
143
![Page 144: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/144.jpg)
AS for TSP – Performance as a Function of the Problem
Di iMean number
Dimension
Network n (dimension) best solution
Mean number of iterations
for to the near-optimal solution
Simulation time (seconds)
4 X 4 16 160 5,6 8
5 X 5 25 254,1 13,6 755 X 5 25 254,1 13,6 75
6 X 6 36 360 60 1020
7 X 7 49 494 1 320 134407 X 7 49 494,1 320 13440
8 X 8 64 640 970 97000144
![Page 145: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/145.jpg)
Summary of ASSummary of AS• Ants are launched at each iteration from each node to
explore the network• Ants build their paths probabilistically with a
probability function of: (i) artificial pheromone values, and(ii) heuristic values (in TSP: city visibility)(ii) heuristic values (in TSP: city visibility)
• Ants memorize visited nodes• Once reached their destination nodes (in TSP the lastOnce reached their destination nodes (in TSP the last
city on their list) ants retrace their paths backwards, and update the pheromone trails
145
![Page 146: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/146.jpg)
C l iConclusion
146
![Page 147: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/147.jpg)
Take Home Messages1. Differences between artificial and natural SI2 Differences between computational and physical SI2. Differences between computational and physical SI3. Key mechanisms for natural SI: self-organization and
stigmergyg gy4. Self-organization ingredients: positive feedback,
negative feedback, randomness, multiple interactions5 SI based systems exploit careful balance between5. SI-based systems exploit careful balance between
exploration and exploitation6. Microscopic models help understanding SI-based p p g
systems7. Ants exploit trail laying/following mechanisms and
other strategies for foragingother strategies for foraging
147
![Page 148: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/148.jpg)
Take Home Messages
8. Ants are able to generate efficient inter-nest networks9. Trail laying/following mechanisms can be expanded
with other properties of the agent easily implementable i ft ( d l ti f thin software (e.g., memory, modulation of the pheromone quantity, etc.)
10 Ant System has been the first metaheuristic taking10. Ant System has been the first metaheuristic taking advantage of the ant inspiration
11 The first NP hard problem it has been applied was the11. The first NP hard problem it has been applied was the Traveling Salesman Problem
148
![Page 149: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/149.jpg)
Additional Literature – Week 1Books
K ll K "O f C l Th N Bi l f M hi S i l• Kelly K., "Out of Control: The New Biology of Machines, Social Systems and the Economic World", Perseus Press, 1995.
• Ormerod P., "Butterfly Economics: A New General Theory of S i l d E i B h i " P th B k 2001Social and Economic Behavior", Pantheon Books, 2001.
• Crichton M., “ Prey: A Novel”, HarperCollins, 2002.• Bonabeau E. and Theraulaz G., “Intelligence Collective”. Hermès,
1994 (in French).• Camazine S., Deneubourg J.-L., Franks N. R., Sneyd J., Theraulaz
G., Bonabeau E., “Self-Organization in Biological Systems”. P i S di i C l i P i U i i PPrinceton Studies in Complexity, Princeton University Press, 2001.
• Hölldobler B. and Wilson E. O., “Journey to the Ants: A story of S i tifi E l ti ” Th H d U i it PScientific Exploration”, The Harvard University Press, Cambridge, MA, 1994.
149
![Page 150: Distributed Intelligent Systems · • Ud dh dld i l l dUnderstand how to model, design, control, evaluate, and optimize distributed intelligent systems • (f d f lt i ith j t) L](https://reader036.vdocument.in/reader036/viewer/2022070916/5fb69d142cc3d804853d3df5/html5/thumbnails/150.jpg)
Additional Literature – Week 1Papers• Beni G. and Wang J., “Swarm Intelligence”. Proc. of the Seventh Annual
Meeting of the Robotics Society of Japan Tokyo Japan 1989 pp 425 428Meeting of the Robotics Society of Japan, Tokyo, Japan, 1989, pp. 425-428.• Bonabeau E. and Theraulaz G. "Swarm Smarts". Scientific American, March
2000, pp. 72-79.• Peter Miller “Swarm Theory”, National Geographic, July 2007, pp. 126-147.y , g p , y , pp• Aron S., Beckers R., Deneubourg J.-L., and Pasteels J.-M., “Memory and
Chemical Communication in the Orientation of two Mass-Recruiting Ant Species”. Insect Society, 1993, Vol. 40, pp. 369-380.
• Deneubourg J L Aron S Goss S and Pasteels J M “The Self• Deneubourg J.-L., Aron S., Goss S., and Pasteels J. M., The Self-Organizing Exploratory Pattern of the Argentine Ant”. J. of Insect Behavior, 1990, Vol. 3, pp. 159-168.
• Beckers R., Deneubourg J.-L., and Goss S., “Trails and U-Turns in the S l ti f P th b th A t L i i ” J Th Bi l V l 159Selection of a Path by the Ant Lasius niger”. J. Theor. Biol., Vol. 159, pp. 397-415, 1992.
• Jackson D. E., Holcombe M., and Ratnieks F. L. W., “Trail geometry gives polarity to ant foraging networks”. Nature, No. 432, pp. 907-909, 2004.p y g g pp
• Dorigo M., Maniezzo V., and Colorni A., “The Ant System: Optimization by a Colony of Cooperating Agents”. IEEE Trans. Systems, Man, and Cybernetics, Part B, Vol. 26, pp. 29-41, 1996. 150