I am working in conjunction with them on this project and plan on a lengthy on site demonstration of the nodes' functions and AI. It's purpose mainly to demostrate that the theory of these nodes is highly functional in both network research, for exploitation and protection.
To give you a quick surmise. A small LAN will be setup. NodeH (node hacker) will be inserted and printed documents of the timing and actions that NodeH will take, will be passed out to the crowd. The node will perform actions and an oversight of it's AI will be presented to the crowd describing the reasons and purposes behind it's decisions.
Automated exploitation with an attack tree backbone (Bruce Shniers idea from DR Dobb's Journal) are some of the main features. I have currently a 13 page overview which I am working on with Caezar. I have already begun development, the first run being MS compatible, with a Linux port possibly before DefCon.