DAMAN (Directed Assembly in Multi-agent Networks)

1
DAMAN(Directed Assembly in Multi-agent
Networks)

• Principal Investigators
• Mikhail Prokopenko, Geoff Poulton, Phil Valencia
  (ICTC), Lech Wieczorek (CIP)
• Other Contributors
• Peter Wang, Vadim Gerasimov, Ying Guo, Jiaming
  Li, Geoff James (ICTC)

2
Aim of DAMAN

• To find general methods for the design and
  control of large complex and intelligent
  networks
• Sensor networks
• Self-assembly networks
• Communications networks
• Power distribution networks.
• Part of the GREMLab collaborative project
• Similar, if broader, aims
• Included support from ICT Emerging science
• Initial Foci
• Robust sensor networks for the Ageless Aerospace
  Vehicle (NASA)
• Directed self-assembly of intelligent particles
• First meso-scale, leading to nanoscale

3
Large Networks Why is it hard to find general
rules to make them do what we want?

• Networks which are sufficiently large and
  connected are usually complex.
• Exhibit emergent behavior, which is
• Hard to predict
• Even harder to model successfully
• Finding general design rules is not easy.

4
DAMAN Project History The Good News

• DAMAN began formally in January, 2003
• Very successful project
• GREMLab collaboration worked well
• Enthusiastic team
• All milestones exceeded
• 15 publications (to June 04)
• including one book chapter
• Two more submitted, plus another book chapter in
  preparation.

5
DAMAN Project History The Bad News

• However
• Large changes to Emerging Science in 2003
• Existing ES funds largely distributed to
  Divisions
• ICT Centre decision
• Combine all existing ES funds
• Call for competitive bids for new projects
• No advantage for existing projects
• Unfortunately DAMAN was not selected by this
  process
• No coherent reasons given for the decision
• In consequence, DAMAN ceased operations on 30
  June.

6
Progress
7
Main Design Approach -Top-Down/Bottom-Up (TDBU)
Solution Space (Desirable Goals)
Achievable Goals
Maybe emergent
Intermediate Entities
Emergent behaviour
emergent behaviour
(Complex) Network Base Components
8
1. Ageless Aerospace Vehicle (AAV)
NASA CONCEPT

• CONCEPT DEMONSTRATOR
• Hexagonal structure, 48 x 1mm Al panels
• 192 tiles, 768 piezoelectric sensors

TILE
SIMULATED MULTI-AGENT SKIN
9
AAV Achievements

• Impact Boundaries
• Local agent response to damage producing emergent
  behaviour
• Stable, reportable
• Initiates next stage diagnosis, reporting,
  action (eg. repair, mitigation)
• Metrics to evolve boundaries
• Entropy-based
• Reward stability (temporal) and uniformity
  (spatial)
• Impact networks
• Link regions of damage on the surface
• Useful in diagnosis
• Local, ant-based algorithm

10
Demo Impact Boundary Formation
11
2. Directed Self-assembly - 2D Mesoblocks

STATE MACHINE
-
0
0
Change
Sense

• Sides , - or 0
• Opposites stick
• State machine can change signs
• On stick or unstick

An analogue of nano- or molecular scale
self-assembly
12
Some Examples a Single-block Virus
E
E
E

• Block polarities (1 1 1 1)
• Single internal rule
• (-1 -1 -1 -1) (1 0 0 0) ? (1 1 1 1)

13
3-block Self-replicating Enzyme

• Enzyme reproduces itself as well as forming
  another 6-block structure

Finish
Start
14
Rectangle Factory Example

• Much more complex
• Capable of self-assembling rectangles of variable
  size and shape
• Matlab version only at present

Initial - enzyme
Final enzyme product
15
Conclusions

• DAMAN was an exciting, relevant and successful
  project, unfortunately cut short.
• Work continues within GREMLab
• More targeted to the needs of existing ICT
  projects