Templates. Template is a pattern used to construct another pattern Used in conjunction with sorting behaviour:

Transcription

1 Templates Template is a pattern used to construct another pattern Used in conjunction with sorting behaviour: Predictable behaviour Parametric Example: Define number and location of clusters Move to cluster, not to arbitrary neighbour

2 Templates in Insects Sort of prepattern in environment Shape to be built is predefined Shape of an insect (for construction of royal chamber) Use temperature and humidity for building nests

3 Chamber building With fake queen, no building stimulated Worker stimulated to build when pheromone concentration within range Walls built ~constant distance from queen Simple model can t predict: Creation of pillars first Rapid rate of increase of building (recruitment isn t enough)

4 Swarm Intelligence: Bonabeau et al Figure 5.3

5 Swarm Intelligence: Bonabeau et al Figure 5.4

6 A Model of Chamber Building Workers follow gradient of cement pheromone Workers within a small radius are attracted The more pellets, the more termites are 2 attracted th = k2p k4h + DH H k 2 = amount of pheromone emitted per unit material per unit time k 2 P = total production of pheromone k 4 H = pheromone decay term D H is diffusion term

7 Reaction-Diffusion Model Termite path= Random walk + Chemotaxis Assume response proportional to gradient Flow of termites with building material Deposit of material Diffusion of termites t C = Φ 2 k1c + DC C γ ( C H ) Attractiveness of the pheromone gradient

8 Reaction-Diffusion Model P = k C k P t 1 2 Amount of material deposited (which generates pheromone) Essentially diffusion (evaporation) γ Attractiveness does not necessarily lead to pillars. Only occurs when steady state: C 0 =φ/k 1, H 0 = φ/k 4, P 0 = φ/k 2 Critical Value, no growth for smaller values of γ. Rewriting implies that there s a threshold number of workers before 1/ 2 1/ 2 pillars emerge c = ( k D ) + ( k D ) ) 2 4 C φ 1 H

9 Swarm Intelligence: Bonabeau et al Figure 5.5

10 Swarm Intelligence: Bonabeau et al Figure 5.6 T ( x, y) = e [(( ) ) (( ) ) 2 ] 2 x x / λ + y y λ 0 x 0 / y

11 Model definition t C = where Φ Fk 1 C + D F ( x, y ) = 1 T ( x, y ) C 2 C γ ( C H ) v ( C T ) Inhibition function: prevents deposition of pellets when too much pheromone is present. This is the template V is force of attraction of queen s pheromonal template. Mechanism inhibits pellet deposition when threshold exceeded P = Fk C 1 k t 2 P

12 Swarm Intelligence: Bonabeau et al Figure 5.7 Model predicts creation of walls around queen Creation of pillars Result: predictable self-organization

13 Nest Building and Self Assembly Insect nest architectures are complex Remember the bee hive? Stigmergy used: Quantitative (continuous) Qualitative (discrete) No evidence of a master builder No evidence of a colony or hive plan Structures are emergent Blueprint is in the environment

14 Nest building Stimuli used to generate structure may start simple but become more complex with structure progression Modularity: Basic structure is repeated

15 Collective Building and Self-Assembly in Natural and Artificial Systems AM, EE141, Swarm Intelligence, W6-2

16 Outline Collective building and self-assembly in natural systems Examples Mechanisms Modeling Reverse engineering with GA Collective building and self-assembly in artificial systems Passive bricks Active mechatronic units

17 Collective Building and Self-Assembly in Natural Systems

18 Natural Examples of Collective Building Guy Theraulaz

19 Natural Examples of Collective Building Guy Theraulaz

20 Natural Examples of Collective Building Guy Theraulaz

21 Natural Examples of Collective Building Guy Theraulaz

22 Natural Examples of Collective Building Guy Theraulaz

23 Natural Examples of Collective Building Pascal Goetgheluck

24 Natural Examples of Collective Building Pascal Goetgheluck

25 Natural Examples of Collective Building Pascal Goetgheluck

26 Natural Examples of Collective Building Masson

27 Natural Examples of Collective Building Masson

28 Natural Examples of Collective Digging

29 Coordination Mechanisms for the Building Activity

30 Collective Building Mechanisms 1 The plan 2 Environmental template 3 Stigmergy

31 Collective Building Mechanisms Scott Camazine

32 Collective Building Mechanisms The plan

33 Collective Building Mechanisms 1 The plan 2 Environmental template 3 Stigmergy

34 Collective Building Mechanisms Environmental template The building plan pre-exists in the environment under the form of spatial heterogeneities. The social insect activity only outlines these pre-existing environmental template. Environmental changes performed by insects play a minimal role in the building activity itself. There are several forms of environmental template: gradients naturally existing in the environment (humidity, temperature ) chemical gradients generated by one or more individuals of the colony

35 Nest Structure in the Ant Acantholepis Custodiens T Eggs Larvae Pupae 3.00 a.m p.m.

36

37 Convective Air Flows and Complex Chemical Templates

38 Convective Air Flows and Complex Chemical Templates Guy Theraulaz

39 Collective Building Mechanisms 1 The plan 2 Environmental template 3 Stigmergy

40 Stigmergy Definition It defines a class of mechanisms exploited by social insects to coordinate and control their activity via indirect interactions. Response R 1 R 2 R 3 R 4 R 5 Stimulus S 1 S 2 S 3 S 4 S 5 Stop time Stigmergic mechanisms can be classified in two different categories: quantitative (or continuous) stigmergy and qualitative (or discrete) stigmergy

41 Stigmergy The role of the two different stigmergic mechanisms in collective building 1 Sequence of stimuli and answers quantitatively different Positive feedback and self-organization 2 Sequence of stimuli and answers qualitatively different

42 Quantitative Stigmergy Features The successive stimuli quantitatively differentiate in their amplitude and merely modify the answer probability of other individuals Examples : mass recruitment in ants dead ant aggregation in ant cemetery pillar building in termites

43 Quantitative Stigmergy Pillar building in termites Spatial distribution of insects and their building activity are locally controlled by the pheromone density. Termites impregnate with pheromones the building material Pheromones diffuse in the environment Individuals carrying ground bullets follow the chemical gradient; they climb towards the highest pheromonal concentration and drop there their bullets Material drop rate is proportional to the number of active insects in the local region (positive feedback)

44

45

46 Stigmergy The role of the two different stigmergic mechamisms in collective building 1 Sequence of stimuli and answers quantitatively different 2 Sequence of stimuli and answers qualitatively different Self-assembly process

47 Qualitative Stigmergy Features Successive stimuli are qualitatively different. This process generates a self-assembly dynamics. No pheromone involved? A B C S 1 S 2 S 3 time

48 Nest Building in Polist Wasps

49 Guy Theraulaz

50 Nest Building in Polist Wasps Organisation of the building activity It is indirectly carried out via the different, local configurations a wasp can find in the nest A probability of deposing a new cell is associated to each configuration

51 Nest Building in Polist Wasps Potential sites for building S 2 S 1 S 2 S 1 S 1 S 2 S 1 S 3 S 1 S 2 S 1 S 1

52 Nest Building in Polist Wasps Probability of creating a new cell given the configuration of neighboring cells 1,0 0,8 0,6 0,4 0,2 0, Number of adjacent cell walls

53 Nest-Building Modeling

54 Swarm on a 3D Lattice A swarm of nest builders agent features Reactive actions Random movements on a 3D lattice, no trajectories No embodiment (1agent = 1 single cell), no interference (however 2 agents cannot occupy the same cell) n agents working together means n actions (not necessarily n dropped bricks) before next iteration starts; all n agents see the same configuration at a given iteration Agents team is homogeneous No global plan of the whole building Local perception of the environment

55 Swarm on a 3D Lattice Definition of agent s neighborhood for hexagonal cells The neighborhood of each agent is defined as the 20 cells around it (7 above, 6 around, and 7 below). z + 1 z + 1 z z - 1 z z X 2 0 A X Neighborhood Cell contents

56 Swarm on a 3D Lattice Distributed nest building: several sites are active simultaneously

57 Swarm on a 3D Lattice Examples of building rules Z+1 z Building process is irreversible (no cell can be removed) Deposit rules can be strictly deterministic or probabilistic Z-1

58 Example of a rule set (Polist wasps, 7 rules) Swarm on a 3D Lattice

59 Nest-building Modeling Obtained structures (Polists wasps, 7 rules) With deterministic rules With probabilistic rules

60 Looking for Stable Nest Architectures What kind of nest architectures can we build with the same method? What are the rules to implement in order to obtain stable architectures? An architecture is considered stable when several runs of a simulation with the same rule set generate architectures with the same global structure. Mechanisms of stigmergic coordination reduce the number of stable architectures.

61 Agelaia (13 rules) Examples of Stable Architectures

62 Examples of Stable Architectures Parachatergus (21 rules) Vespa (13 rules) Stelopolybia (12 rules)

63 Examples of Stable Architectures Chatergus (39 rules) Artificial Nest Structure (35 rules)

64 Looking for Stable Nest Architectures How to build a stable architecture? The building process is carried out in successive steps. The current local configuration generate a stimulus different from that of the previous and of the successive configuration (qualitative stigmergy). Only this type of building algorithms generate coherent architectures. The whole set of these algorithms generates a limited number of nest shapes.

65 Reverse-Engineering: From the Building to the Individual Rules using GA

66 Similarities to Controller Evolution Prey-Predator (GA, Floreano 98) Obstacle avoidance (RL, Kelly 97) Area Integration (RL,Versino 97) Exploration (RL, Hayes 01) Nest-building (Bonabeau, GA 00) Exploration (RL, Millan 97) Foraging (GA, Jefferson 93)

67 Evolutionary Encoding of the Distributed Nest Building Problem Phenotype: agent endowed with set of microrules Genotype: set of microrules (one-to-one mapping with phenotype); chromosome of variable length; 1 gene = 1 microrule. Life span: number of iteration (e.g. 30,000 iterations, 1 iteration = all 10 agents have applied their microrule) or exhausting of max amount of bricks available (e.g. 500 bricks). Population: 80 individuals Generations: 50 Fitness function: weighted sum of space filling and pattern replication (arbitrary criteria based on 17 human observers who were asked to evaluate the amount of structure in a set of 29 different patterns, [Bonabeau 2000]); Selection: roulette wheel Crossover: two-points, p crossover = 0.2 Mutation: p mutation1 = 0.9 (during life span inactive microrule); p mutation2 = 0.01 (during life span active microrule).

68 Evolutionary Encoding of the Distributed Nest Building Problem Biased evolution: Probabilistic templates (reduction of stimulating configurations containing a large number of bricks). Start microrule. No diagonal deposits (space not filled). Problems: Fitness function: Functional? Esthetical? Biological plausible? Mathematical definition? Episthatic interactions (sequence of microrules needed for coordinated algorithms).

69 Collective Building and Self-Assembly in Artificial Systems

70 Passive Bricks

71 Passive Bricks In-line 2D structures using Kheperas [Martinoli 99] -> [Easton??] Lionel Penrose ( ) Self-assembly mechanisms of genetic relevant molecules reproduced with passive wood bricks External energy source (shaking, human action) Evolution and self-assembly tightly coupled Video-tape!

72 Passive Bricks Evolutionary architecture: Nicolas Reeves (UQAM Montreal, Canada) Cellular automata approach. No genetic operator: population manager replaced by direct intervention of the human being. 3D CAD simulations Stereolytographic sculptures

73 Passive Bricks Evolutionary architecture: Pablo Funes (Brandeis University, US) GA approach, geometry- and force-based fintness functions Off-board evolution and reproduction of results with Lego bricks 2D (bridge, crane) and 3D structures (table) Set-up and objective Diagram of forces

74 Passive Bricks Evolutionary architecture: Pablo Funes (Brandeis University, US) 2D Ex.: Bridge Simulated bridge Real Lego bridge

75 Passive Bricks Evolutionary architecture: Pablo Funes (Brandeis University, US) 2D Ex.: Crane 3D Ex.: Table

76 Active Mechatronic Units

77 MEL Mechanical Engineering Laboratory Research on Self-Reconfigurable Mechatronic Systems in MEL Eiichi Yoshida Satoshi Murata Haruhisa Kurokawa Kohji Tomita Shigeru Kokaji

78 MEL Mechanical Engineering Laboratory Self-Reconfigurable Mechatronic Systems Distributed mechanical system composed of many identical units using local communication only dynamically reconfigurable Self-assembly / Self-repair arbitrary initial state self-assembly trouble cut off self-repair using spare units

79 MEL Mechanical Engineering Laboratory Self-assembly and Self-repair Self-Reconfigurable Mechatronic Systems difference / diffusion var. per units difference diffusion var. remove Step

80 MEL Mechanical Engineering Laboratory 2D Self-Reconfigurable Systems SMA Torsion Coil Springs SMA Coil Spring Vertical Direction Female: Auto-locking (releasing by SMA) pin Pin holes Male: Rotating Drum Weight: 50[g] (approx. ) Span: 50[mm] Basic Motion

81 MEL Mechanical Engineering Laboratory 2D Self-Reconfigurable Systems Experiment using 6 units Initial Final Moving Unit

82 MEL Mechanical Engineering Laboratory 2D Self-Reconfigurable Systems

83 MEL Mechanical Engineering Laboratory 3D Self-Reconfigurable Systems Basic Motion A

84 MEL Mechanical Engineering Laboratory 3D Self-Reconfigurable Systems Grey: distance from target different from zeo -> to be moved Green: distance from target = 0 -> ok! Red: current moved unit (once at the time)

85 MEL Mechanical Engineering Laboratory 3D Mechanical Unit 3D Self-Reconfigurable Systems connecting hand rotating arm span: 27.5cm weight: 6.8kg DC motor: 7W

86 Modular Robotics (PolyPod) Mark Yim (Stanford University, Xerox Park Research Center, ) Parallel structure, 10 links Two types of modules: segment and nodes Nodes: power modules Segments: HC11 microcontroller, angle position (potentiometer), IR local communication

87 Modular Robotics (PolyPod) Mark Yim (Stanford University, Xerox Park Research Center, )

88 Modular Robotics (PolyBot) Mark Yim (Stanford University, Xerox Park Research Center, ) The future PolyBot This generation of PolyBot included onboard computing (Power PC 555) as well as the ability to reconfigure automatically via shape memory alloy (SMA) actuated latches.

89 Self-Configurable and Modular Robotics Other researchers: Hajime Asama, RIKEN Center, Saitama, Japan Distributed assembling and disassembling of stair-like structures Peter Will and Wei-Min Shen, ISI-USC, L.A.,US Modular robotics

90 Conclusion

91 Self-Assembly, Distributed Building, Modular Robotics Very promising domain but System design and hardware development are crucial Energetic problem still unsolved Scalability and distributed control still unsolved problems: how to control individual units for obtaining a given team performance? Evolution and bio-inspiration can help: incremental/modular evolution, reverse engineering