SOALA: Self-Organizing and

Autonomous Learning Agents

Project description

This research project addresses two fundamental issues for autonomous learning and self-organizing agents:

1. How can organizational adaptation be totally distributed to individual members and scale up to very large agent systems?

2. How can an individual agent learn from its task performance and benefit its entire organization?

For totally distributed organizational adaptation, we propose to continue using the hormone-based method to allow
agents to reorganize their social and physical structure with complete autonomy. Compared to traditional,
address-based control protocols, this method is totally distributed, robust for reorganization and reconfiguration, and
capable of scaling up. We will generalize the initial results of this idea and demonstrate the full potential for applications
such as metamorphic robots, distributed sensor networks, and self-reconfigurable systems in general.

For autonomous learning agents that can support adaptive organizations, we propose an affordance-based modeling
technique to allow agents to model their interactions with the environment in a multi-layered internal representation, and
use the model to recognize and manipulate new environmental scenarios to achieve organizational goals. This
sensor-action grounded, multi-layered modeling technique can constrain the unbounded explorations and changes in the
environment as well as enable collaboration and reorganization through re-distributing task-roles among agents.

Funding Sources: AFOSR.