See our new website at research.ict.csiro.au

Distributed Intelligence - Directed Self-Assembly

The Vision: Self-Assembly as a design problem

  • How to engineer local agent properties leading to desired global emergent behaviour?

  • How to recognise, measure and harness the emergent behaviour?

The CSIRO Complex Systems Science activity was established as an emerging science area to explore and exploit the links between different applications in which complex behaviour arises from the interaction of simple components. 

Directed Assembly in Multi-agent Networks (DAMAN) is a 3-year project supported by the CSS activity (started in January 2003). 

The aim of DAMAN is the development of tools necessary for the design of complex networks of autonomous sensors/agents, self-assembling at a hierarchy of scales from the macro- right down to the nano-scale. We are using a variety of methods including multi-agent systems, genetic algorithms and amorphous computing to develop general tools for designing multi-agent networks self-organising for designated global tasks. Initial research is at the macro-scale, using an existing sensor and communication network developed as a concept demonstrator for a self-monitoring and self-repairing aerospace vehicle. The ultimate, and quite ambitious, aim is to apply these tools to controlled self-assembly at the nano-scale.

 
Berkowitz Group: Molecular Dynamics (Micelle)
The University of North Carolina at Chapel Hill		 
Malin Space Science Systems, MGS, JPL, NASA
The Schiaparelli Basin of Mars (MGS Mars Orbiter camera)

Self-organising Impact Boundary on Ageless Aerospace Vehicle skin.

The proposed methodology is based on an iterative process that includes the following steps:

  1. forward simulation leading to emergent behaviour for a class of localised algorithms; 
  2. quantitative measurement of spatiotemporal stability of the emerging structures, using information-theoretic metrics for phase transitions;
  3. evolutionary modelling of the desired global emergent behaviour, where the fitness functions correspond to the metrics obtained at step 2).

These steps should lead to "axiom sets" suitable for design at the nano-scale.

There are a few related activities: