Siren: Context-aware Computing for Firefighting

Based on an extensive field study of current firefighting practices, we have developed a system called Siren to support tacit communication between firefighters with multiple levels of redundancy in both communication and user alerts. Siren provides a foundation for gathering, integrating, and distributing contextual data, such as location and temperature. It also simplifies the development of firefighting applications using a peer-to-peer network of embedded devices through a uniform programming interface based on the information space abstraction. As a proof of concept, we have developed a prototype contextaware messaging application in the firefighting domain. We have evaluated this application with firefighters and they have found it to be useful for improving many aspects of their current work practices. INTRODUCTION Each year, fires kill about 4,000 civilians and 100 firefighters in United States alone [1]. Firefighting is a dangerous profession that calls for quick decisions in highstress environments, constant reassessment of dynamic situations, and close co-ordination within teams. The smoke, heat and noise in a structure fire mask the environment and force firefighters to operate with an incomplete picture of the situation. “Firefighting is making a lot of decisions on little information,” said one firefighter we interviewed. Improvements in information gathering, processing and integration can help firefighters work more effectively to prevent injury and loss of life, as well as minimize property damage. The pervasive computing community itself can also benefit from research in this area. The nature of emergency response is fundamentally different from office environments, in terms of physical risk, psychological state, and operating conditions. This poses unique challenges for designers and researchers in terms of context awareness, new interaction techniques, and information visualization, to name a few. If we can make an impact in this highly stressful domain, where the systems we offer are secondary to the primary task, we might also be able to apply these results in less extreme environments for a wider audience, such as computing while driving. From an extensive four-month field study of current firefighting practices, we found that firefighters often need to exchange information about their situation and their surrounding environment in a spontaneous and opportunistic manner. Such interaction is especially useful when firefighters need to be alerted about imminent dangers. This type of interaction needs to be spontaneous because the time when information exchange will occur depends on the dynamically changing situation and often has to be done without direct human initiation. It also needs to be opportunistic because the constant movement of firefighters in a complex urban structure makes it impossible to maintain an always-on communication channel among them. The problem is that spontaneous and opportunistic interactions among firefighters are not well-served by current systems. Today, most firefighters rely on two communication channels on the scene of a fire. The first is a broadcast channel for voice communication. The second is a data broadcast channel for status update between incident commanders and dispatchers at a centralized emergency response center. Both channels use the 800MHz to 900MHz radio band. Often, only the voice channel or data channel can be used at a given time. Moreover, both channels are broadcast driven and manually operated to support explicit communication rather than the tacit communication needs between firefighters. Advances in pervasive computing technologies provide us with an opportunity to let firefighters “see through the eyes of fellow firefighters” and provide a greater understanding of the overall situation. Small, cheap, wirelessly networked sensors (such as smart dust [2]) can be deployed on firefighters and in buildings, allowing us to gather contextual information—such as temperature, sound, movement, toxicity, and a person’s location—at a level never seen before. The wealth of sensor data about firefighters and the environment can be exchanged between firefighters to help improve safety and effectiveness. To support spontaneous and opportunistic interactions between firefighters, we have developed Siren, a peer-topeer context-aware computing architecture that gathers, integrates, and distributes context data on fire scenes. To make tacit communication between firefighters more robust in the face of an inherently unreliable transport, Siren offers multiple levels of redundancy in communication and feedback. Siren also simplifies development of emergency response applications by providing a uniform programming interface based on the information space abstraction [3]. Using Siren, we have developed a prototype context-aware messaging application and conducted an evaluation of this application with firefighters. They have found it useful for improving many aspects of their current work practices. The rest of this paper is organized as follows. We discuss related work in section 2. An example search and rescue scenario is given in section 3 to illustrate how Sirenenabled applications may assist firefighters. In section 4, we describe key findings from our field study of current firefighting practices, and how they motivate the design of Siren. Section 5 describes key components of the Siren architecture, including a programming model based on the information space abstraction, storage management, communication, and the context rule engine. We describe a prototype context-aware messaging application developed using Siren in section 6, and our evaluation of it with firefighters in section 7. We discuss some lessons we have learned about designing for mission-critical pervasive computing applications in section 8. We conclude in section 9 and discuss future work. RELATED WORK There has also been a great deal of work at providing programming support for various aspects of pervasive context-aware computing. This includes the PARCTab system [4], Cooltown [5], the Context Toolkit [6], Contextors [7], Limbo [8], Sentient Computing [9], Stick-E notes [10], MUSE [11], SpeakEasy [12], Solar [13], XWeb [14], GAIA [15], one.world [16], and iRoom [17]. Most of them are designed to support office work in traditional work environments. Siren, on the other hand, aims to support field work practices of mobile firefighters. Context Fabric [18] is a generalized service infrastructure for context-aware computing that implements an information space abstraction [3] and a P2P infrastructure. Siren implements Context Fabric for a peerto-peer network of PDAs, and provides new capabilities for discovery, multi-hop communication and rule-based adaptation. Other systems have also attempted to support emergency responders using mobile ad-hoc mesh networks (MANET). Draco [19] aims to develop rapidly deployable networking technologies that can support voice and video communications between emergency responders. In the commercial world, companies such as MeshNetworks are developing a self-forming, self-healing technology that automatically creates a wireless broadband network at an incident [20]. In Siren, we employ standard 802.11b networking technology and focus instead on supporting tacit communication needs between firefighters at the application level. While all MANET work focuses on design of the underlying communication protocol, our goal in Siren is to integrate sensing, communication and feedback in a single architecture to support tacit communication needs of firefighters. The Command Post of the Future [21] is a set of projects investigating command in battlefield situations. The focus is on developing technologies for mobility and better decision-making, including multimodal interaction, information visualization, and knowledge-based reasoning. We complement this work by looking instead at tacit interactions between firefighters within a building, focusing on how the underlying software infrastructure can be designed to better support spontaneous and opportunistic communication. Our work is also related to Camp et al, who looked at communication issues in emergencies and prototyped a radio system that would reduce voice congestion while maintaining situational awareness [22]. In contrast, we concentrate on supporting tacit data communication needs between firefighters. SIREN-ENABLED SEARCH AND RESCUE: A SCENARIO In this section, we describe a scenario motivated by our field studies to illustrate what tacit communications needs we intend to support. Three firefighters conduct a search and rescue task in an office building. Each firefighter carries a PDA. Wirelessenabled sensors have either been placed at strategic locations (e.g. on smoke detectors) throughout the building, or have been deployed on the fly by initial fire response team whose duty is to size up the situation. As firefighter F1 walks into the building, his PDA continuously monitors surrounding temperature from nearby sensors. F1 approaches the south exit of the building when his PDA detects an usually high heat level and alerts him of imminent danger. As a result, F1 avoids the south exit. At the same time, his PDA also detects the presence of firefighter F2, who is conducting a search and rescue task in an adjacent room, and automatically sends a message containing information about the south exit to F2’s PDA. A few minutes later, F2 receives an immediate evacuation order from the incident commander. F2 knows that there is a growing fire hazard around the south exit, so he rushes toward north. On his way to the north exit, F2’s PDA detects the presence of firefighter F3, who is running toward the south exit, unaware of the potential danger. F2’s PDA automatically forwards the message received from F1 to F3’s PDA, which immediately alerts him of the danger. As a result, F3 turns back to run toward the north exit, safely leaving the building. FIELD STUDY OF FIREFIGHTING PRACTICES To inform the design of Siren, we conducted a field study that spanned four months and included over 30 hours of interviews and low-fi prototype evaluation with 14 firefighters in three fire departments. The firefighters were from many levels of the organizations, including one assistant chief, four battalion chiefs, two captains, two engineers and five firefighters. We employed several investigation methods for our study, including interviews, training observations, and field observation. We conducted interviews at fire stations while the firefighters were on duty and learned about their organizational structure, tools, routines, regular interactions, and typical environment. By observing one field exercise at a training tower used to train new firefighters, we received a first hand sense of how firefighters tackle an urban structure fire. In addition, we accompanied firefighters on two emergency calls to see first hand how they accomplished their tasks. A full account of our field study can be found in [23]. We have looked at many aspects of current emergency response practice. In this section we give an overview of some findings that help motivate the key design decisions in Siren. Figure 1. We conducted a four-month field study with firefighters in their work environment. This figure illustrates one such environment: a mobile incident command located at the back of a command vehicle.