This page describes Hewlett-Packard (HP) funded project ContextExplorers and contains information, resources and publications related to this project. This project was made possible through the HP PhD Endowment to CoolCampus initiative of Monash University and efforts of our PhD student Amir Padovitz and advisory team comprising A/Prof Arkady Zaslavsky, Dr Seng Wai Loke, Dr Bernard Burg (formerly with HP Labs), Claudio Bartolini (HP Labs) and our colleagues and visiting scholars including Prof Mihhail Matskin (KTH, Stockholm), Prof Jacques Ajenstat (University of Quebec, Montreal) and others.
We collaborate with LIP-6 (Paris-6) university research group working on context & led by Prof Patrick Brezillon
The emerging research area of context-aware pervasive computing is envisioned to significantly enhance the performance of applications, operating in networked and sensor-based environments. With the advent of the context-aware computing paradigm, adaptation of these applications to various situations of interest will become feasible. This in turn, would enhance the applications' performance and quality of services they provide to their clients. Context-aware applications rely on the existence of sensors and communication facilities in pervasive network settings, and are dependent on the type and characteristics of context they attempt to reason about. Consequently, different challenges, which originate from these dependencies currently exist, and impede the proliferation of context-aware applications in the marketplace. We express our context related philosophy in the context-situation pyramid.
In this research we attempt to deal with some of the challenges associated with context-aware and pervasive computing and address the following research questions.
To address these issues we develop a theoretical model (including algebra to express the model), built on intuitions from state-space analysis and multi-dimensional geometry for capturing context details. We develop practical solutions in the form of algorithms and algebraic operations, which are based on our context theory for achieving reasoning, verification and prediction of context in context-aware, pervasive environments. We aim to provide an architecture, which makes use of our approaches and which provides solutions for challenges concerning context uncertainty and challenges concerning the complex nature of pervasive networks. Within such architecture, we examine the impact of mobility on the process of reasoning about context, using mobile agent technology.
We propose a hybrid architecture consisting of a centralized reasoning engine coupled with light-weight distributed context-aware mobile agents that implement only a subset of the reasoning functionality. A major concern in designing architectures for pervasive systems is providing practical solutions for management and control difficulties associated with the underlying complex environments in which such systems may operate. The emergence of sensor networks consisting of large number of resource limited sensors, and ad hoc networks, in which wireless roaming devices result in continuing changes in network layout can be considered the backbone of future pervasive systems. These complex and untraditional networks necessitate architectural approaches that facilitate context reasoning capabilities, which are also realistic in their ability of achieving them in large scales. We argue that in most cases an architectural trade-off exists between a desire to perform elaborate and meaningful reasoning and the need to successfully control and manage large scale networks. For achieving meaningful awareness of uncertain situations, a centralized, server-oriented, resource consuming mechanism is a natural choice, one which is able to make use of a variety of information, internal or external to the system, and be able to integrate unrelated data in discovering and reasoning about situation discrepancies. However, this approach may not scale well and is not fault tolerant, and would be inappropriate in complex pervasive networks (e.g. large scale sentient building), in particular when the reasoning process needs to account for a variety of entities and perform separate reasoning for each one of them and/or when real-time performance considerations are required. Furthermore, it is possible that information in such systems may often be provided by separate cross-organizational entities, employing different policies and reasoning rules. In such environments we need to consider the decentralization of the system (either for the purpose of a single organization, or as a platform that provides context capabilities for a variety of unfamiliar entities), to improve the system's fault tolerance and scalability as well as the ability to intelligently gather information in a dynamically changing network environment. In addition, we would like to achieve a degree of flexibility by obtaining an ability to ‘inject awareness' or embed context-awareness functionality in specific physical areas, as well as reason about specific situations of interest. We achieve this by pursuing hybrid architecture; one, in which a central knowledgebase and reasoning engine support an overall decentralized architecture that consists of a collection of autonomous mobile agents that are capable of intelligent data gathering and performing local reasoning tasks. As an example, consider a sentient building in which mobile users are considered entities whose context need to inferred by the system. When the number of roaming people is large, it becomes hard to monitor, manage and reason about each person's specific context using a central engine. This becomes the tasks of personal software agents. Alternatively, consider an intruder detection system, which goal is to detect and reason about a single suspicious person rather than reason about all personnel in the building. We would then like to inject specific agents that carry out this task. Once reasoned information, obtained by agents, becomes available to the system, a centralized reasoning engine can infer more elaborate and precise situations, cross-check and cross-reference it. Mobile agents, which are injected into the system, can support the centralized engine by intelligently gathering appropriate information necessary for the reasoning process, therefore reducing communication load and conserving bandwidth. They also provide fault-tolerance by being able to operate during disconnections in the wireless network. The general architecture of ContextExplorers system is depicted below. We conceptually divide this architecture into three layers, namely, a reasoning layer, a data layer and a communication layer.
Our Papers and Presentations
|Maintained by Arkady Zaslavsky. Last updated: 04-Nov-2008 10:36|