Workflow Support for Wireless Sensor and Actor Networks A Position Paper Pablo Ezequiel Guerrero ∗ Daniel Jacobi † Alejandro Buchmann Dept. of Computer Science Technische Universit¨ at Darmstadt D-64283 Darmstadt, Germany {guerrero, jacobi, buchmann}@dvs1.informatik.tu-darmstadt.de ABSTRACT As initial challenges of wireless sensor and actor networks (WSANs) are overcome, their application possibilities evolve. For these applications to move mainstream, efficient pro- gramming methods are required which can be used by do- main experts. So far, the question of how can WSANs be efficiently programmed remains unanswered. In this paper we examine proposed middleware approaches, and show that they have focused on data extraction rather than in-network actuation. We thus propose the usage of workflows as a means to define the logic that orchestrates the network ac- tivity, and introduce a language to express WSAN interac- tions. At this time, a concrete system is not given, but the paper discusses the relevant aspects towards one, and poses many questions for future research. 1. INTRODUCTION Ever since the idea of merging fundamental sensing, pro- cessing and wireless communication capabilities into tiny de- vices emerged [26, 27], an enormous progress has been made to get large numbers of low-cost, battery-powered nodes to carry out collective tasks [2]. The tight collaboration be- tween experts from different areas leveraged a technology spread over many domains. Indeed, the last 5 years have seen a number of large experimental deployments of wire- less sensor network (WSN) applications. One of the earliest deployments was on Great Duck Is- land [18]. The study, in cooperation with researchers in the Life Sciences, aimed at unobtrusively learning about seabirds and their environment, by installing sensor nodes in and around their burrows during nesting periods. Another deployment, in this case together with glaciologists, aimed at investigating the behavior of glaciers by inserting sen- sor nodes in them [19]. A third deployment, now involving multi-hop communication, was carried out in dutch potato fields [13]. There, crops must be protected against fungal diseases, which are strongly associated to the climatolog- ical conditions within the field. The deployments in this ∗ Supported by the DFG Graduiertenkolleg 492, Enabling Technologies for Electronic Commerce. † Supported by the DFG Graduiertenkolleg 1362, Coopera- tive, Adaptive and Responsive Monitoring in Mixed Mode Environments. 4th International Workshop on Data Management for Sensor Networks. DMSN’07, September 23-28, 2007, Vienna, Austria. Copyright is held by the authors. first group faced many issues in common such as network longevity, remote administration and unobtrusive monitor- ing. Their application logic, however, is quite simple: the (mostly) raw observed data must be pushed out of the net- work. Note that the parameters of this logic, i.e., the sam- pling rates, are normally specified by the domain experts. In a second group of applications, a continuous observa- tion of the environment is not required. In contrast, the slightly more complex goal is to detect an event of interest and observe the phenomena afterwards. One such deploy- ment was that carried out with volcanologists at Ecuador’s Volc´an Reventador [37]. Each node waits until its seismome- ter readings exceed a certain threshold, and then notifies a base station, which triggers a data collection phase. Another example in this group are structural health monitoring sys- tems like Wisden [40] or the one deployed on the Golden Gate Bridge [12]. The event detection is followed by local data storage and posterior progressive coding (compression) for efficient data transmission. The parameters of the ap- plication logic, i.e., the sampling rates and the thresholds, are also given by the experts in the domain, who may vary them to adjust or refine the experiment. A third type of applications is considerably more com- plicated. The habitat monitoring system deployed at the Coastal Redwood Forests of California [17] allows complex queries to be injected into the network, whose results are ag- gregated as they are streamed back to a base station. In the industrial scenario of the CoBIs project [33], nodes attached to chemical drums cooperate without using any external in- frastructure to check for hazardous situations and violations of safety regulations. Again, the logic behind these appli- cations is precisely stated by domain experts, now in the shape of SQL-like queries or inference rules, respectively. From this categorization, it is clear that the research focus has been on data extraction and event detection. While we observe that many operations have been effectively moved into the network, the decision on how, when and where to perform certain actuation is only taken off the network, ei- ther by a human, or with the help of a decision support system. This is logical, since only after data had been con- solidated in a central server was it possible to reason about it and decide what to do next. However, a WSN can be easily extended with other nodes further capable of actuat- ing, forming a wireless sensor and actor network (WSAN) [1]. As an example, WSANs are needed for the control of autonomous vehicles (AVs) and the coordination of swarms thereof. Individual actuators are as diverse as sensors; in