Objectives
The term computing as interaction refers to computing as something that happens by and through communication among computational entities (also called agents). It arises as a means to tackle the increasing complexity of software systems and the need of autonomous and distributed computation. This has given rise to a new set of agent technologies in which computing is regarded as an activity based on the agents’ social interactions.
The aim of the project is to develop new agent techniques and technologies based on social dynamics for the design and implementation of a multi-agent planning platform (PlanInteraction) composed of autonomous and, possibly heterogeneous, planning entities. The research in PlanInteraction is in line with current efforts in decision theory, game theory, automated planning and scheduling, learning, and other research fields.
The present project is proposed as an attempt to integrate previous scientific and technological contributions of the involved groups, namely MAGENTIX (TIN2008-04446), a platform for the design of secure and optimized management of open multi-agent systems, and PELEA (TIN2008-06701-C03), a general purpose-architecture for planning, execution and learning. We thus envisage the design of PlanInteraction at three levels:
a) Agent level. The key building block is the planning agent with capabilities for planning, acting and learning. More specifically, an agent is a software platform that integrates a full spectrum of planning-related technologies, including sensing, execution, monitoring, re-planning and even learning from past experiences.
b) Interaction level. The key operation is interaction – plugging agents together so that they interact with each other. The essence of the agents’ behaviour is now how they interact with the environment.
c) Organization level. The key notion in agent structure is organization – how agents group together in organizational structures or coalitions that represent a more advantageous position for reaching a particular set of planning goals or carrying out a complex planning task.
And the specific project aims include:
1) Adapting the single-agent planning techniques developed in PELEA to a multi-agent environment, resulting in a model of multi-agent planning
2) Definition and development of new agent technologies for the coordination and interaction of planning agents, resulting in a set of new techniques for multi-agent planning
3) Integration of the PELEA single-agent planning architecture into MAGENTIX, resulting in an open-source platform that allows users to easily develop complex multi-agent planning applications
Examples of domains in which these technologies are starting to be needed arise from recent experience of the groups on solving planning research problems and developing real-world applications.