Assessing the SALSA architecture for developing agent-based ambient computing applications

We have proposed the use of autonomous agents for coping with some of the challenges of creating ambient computing systems. The motivation of this research is that ambient computing environments are characterized by the distribution, reactivity, collaboration and adaptation of their artifacts, which are also characteristics attributed to software agents. To assist developers in creating the software entities of an ambient computing environment, the Simple Agent Library for Smart Ambients (SALSA) was created. The SALSA middleware and architecture enables the creation of autonomous agents reactive to the context of the ambient computing environment. SALSA agents can represent users, resources, or wrap complex system functionality of the environment. The aim of this paper is to provide evidence that SALSA facilitates the implementation of ambient computing services through autonomous agents. Unfortunately, the literature on Ubicomp development frameworks has, aside from a few exceptions, not reported experimental evaluation of their usability. The evaluations of Ubicomp development platforms have not addressed this issue since their evaluations have been mostly focused on performance and to prove feasibility. We present the results of an empirical evaluation conducted to assess the use of SALSA. This study included in-lab programming experiments and design exercises to evaluate the programming facilities provided by SALSA agents. Even though for some of the evaluation participants the use of autonomous agents as an abstraction for the development of ubiquitous computing systems was not innate, the evaluation results demonstrate that the execution model of SALSA and its facilities to implement Ubicomp systems are comprehensible.

Research highlights
► We evaluate the use of software agents to create ambient computing applications.
► Subjects completed programming and design tasks using an agent framework.
► Participants successfully incorporated agent abstractions into their design.
► Inferring context using XSL filters was the task that caused more difficulties.
► The use of an ontology could facilitate inferring secondary context information.