Recent advances in agent software have tended to confirm early predictions of the use of market mechanisms to achieve robust decentralized control over distributed, parallel, and heterogeneous systems. This is this is different from the original goal of controlling Avogadro quantities of drexlerian assemblers with agoric software, but it turns out to be important for other reasons. In many ways, a molecular assembler is quite similar to a factory, so factory control techniques using software agents that are being developed now will most likely be applicable to molecular assembly.
Nanotech proponents have waxed enthusiastically about market based mechanisms to control large numbers of assemblers, and there has been some success in demonstrating how this might occur (eg. "Spacial Self-Organization in Large Populations of Mobile Robots" by Cem Unsal and John Bay; "Market Organizations for Controlling Smart Matter" by Oliver Guenther, Tad Hogg, and Bernardo Huberman). In addition, some present-day factories have achieved periodic "lights-out" manufacturing (eg. "Lights Out" Part Processing, FANUC Robotics, Considerations For Manufacturing Execution System Implementation). But with molecular assemblers, one must build, initialize subcomponents, and run complex manufacturing processes completely without any manual control. This paper provides a survey of these two emerging trends, and maps out a few milestones on the path to autonomous assemblers.
Not only self-replication, but any nanoscale manufacturing will involve many subcomponents and many steps. The 30 convergent assembly steps necessary to build something the size of a breadbox from atomic parts is somewhat oversimplified, especially when building a wide variety of parts. Getting the right molecule in the right orientation at the right time is not only an issue in mechanosynthesis, but it is also involves issues of inventory, process control, and materials handling -- issues being addressed on a macro scale in manufacturing plants today.
The Environmental Research Institute of Michigan and Deneb, Inc. are jointly developing a distributed agent-based system that can assist and supplant human decision making in efficiently allowing material flow and task scheduling to emerge in an environment of manufacturing assembly. The Agent Network for Task Scheduling (ANTS) uses techniques inspired both by free market economics and insect colonies. This paper describes a new mechanism called least commitment scheduling that defers decisions on process sequences until the last possible moment. A Density-based Emergent Scheduling Kernal (DESK) uses probabilistic committed capacity profiles of resources over time, along with realistic costs, to provide an abstract search space over which the agents can wander to quickly find optimal solutions.