Operational Semantics of Goal Models in Adaptive Agents

Several agent-oriented software engineering methodologies address the emerging challenges posed by the increasing need of adaptive software. A common denominator of such methodologies is the paramount importance of the concept of goal model in order to understand the requirements of a software system. Goal models consist of goal graphs rep- resenting AND/OR-decomposition of abstract goals down to operationalisable leaf-level goals. Goal models are used primarily in the earlier phases of software engineering, for social modelling, requirements elicitation and analysis, to concretise abstract objectives, to detail them and to cap- ture alternatives for their satisfaction. Although various agent programming languages incorpo- rate the notion of (leaf-level) goal as a language construct, none of them natively support the definition of goal mod- els. However, the semantic gap between goal models used at design-time and the concept of goal used at implementa- tion and execution time represent a limitation especially in the development of self-adaptive and fault-tolerant systems. In such systems, design-time knowledge on goals and vari- ability becomes relevant at run-time, to take autonomous decisions for achieving high level objectives correctly. Recently, unifying operational semantics for (leaf) goals have been proposed (Riemsdijk, AAMAS08). We extend this work to define an operational semantics for the behaviour of goals in goal mod- els, maintaining the flexibility of using different goal types and conditions. We use a simple example to illustrate how the proposed approach effectively deals with the semantic gap between design-time goal models and run-time agent implementations.