Context awareness and information processing in opportunistic ubiquitous systems

OPPORTUNITY Workshop at UbiComp 2010
Daniel Roggen, Wearable Computing Lab, ETH Zurich, droggen@ife.ee.ethz.ch
Alois Ferscha, Institute for Pervasive Computing, University of Linz
Gerhard Tröster, Wearable Computing Lab, ETH Zurich
Paul Lukowicz, Embedded Systems Lab, University of Passau
Hans Scholten, Pervasive Systems, University of Twente


Workshop background material as PDF available here.

PROCEEDINGS

Download the proceedings of the workshop here.

PROGRAM - September 26th, 2010

OBJECTIVES

This workshop has for objectives to:

• Get the community to think about the potentials and challenges of "opportunistic" ubiquitous systems: systems that exploit sporadically available sources of information and communication to provide a ubiquitous computing experience.
  
• Paint a picture of the current state of context awareness and information processing in opportunistic ubiquitous systems
• Identify new research directions and new application domains resulting from this new trend.
  

Opportunistic sensing not only calls for new networking paradigms, but also for new means to infer meaning from acquired sensor data. As a result of surveying the domains where opportunistic sensing enables new forms of applications and
the approaches used in these domains, we will be able to paint a clear picture of the set of algorithms, methodologies and tools that are currently proposed. From this we expect cross-fertilization across the application domains, with methods pioneered in one domain being translated to other domains. As a result we seek an overall information processing architecture to infer context from opportunistically sensed data.

However, other outcomes include a documented set of lesson’s learned and best practices to design opportunistic context-aware systems. Also, we expect to
identify the new research directions in the methodologies (sensing, networking, machine learning, signal processing, and reasoning), as the emerging application domains for opportunistic context awareness.

The papers submitted to this workshop are published on a website and if there is enough interest a proceedings will be published with an ISBN number via an online print-shop.

A concise version for publication in a jointly authored article is also possible.

BACKGROUND

Considering the huge amount, and ever growing number of a vast manifold of heterogeneous, small, embedded or mobile devices shaping the ubiquitous computing landscape, makes traditional design-time-defined sensing approaches more and more incompatible. Ubiquitous computing system designs will have to and are already successfully attempting to revert the principle of design-time-defined sensing, to one where technology in a self-aware approach attempts to opportunistically collect data from whatever sources (physical sensors, data repositories, referenced data in the internet) and try to make meaning out of it. Observably, large scale ubiquitous computing systems are undergoing such a paradigm shift already today, a prominent example being positioning systems
implemented in the current generation of smart phones, opportunistically making use of GPS, GSM cell, WLAN and BT signals to position the device depending on their availability.

As such, opportunistic sensing and information processsing appears as a fundamental principle underpinning self-aware ubiquitous computing systems involving very large numbers of entities in open ended environments, and it is a key towards of future generation ubicomp systems. Thus, opportunistic sensing and information processing is not limited to a specific application domain, but rather it observably represents a paradigm shift in large-scale self-aware systems. Thus, “application domains” span from sensing policies for open ended systems, to system able to meaningfully integrate everything that “emerges” out of technological progress, but cannot be foreseen - and thus not be designed for (at design time).

OPPORTUNISTIC SENSING

Until recently sensing networks consisted of homogeneous clusters of small sensor nodes, communicating wirelessly, executing a fixed set of applications in a distributed fashion. These wireless sensor networks are the interface between the cyber world and the physical world, where data is streamed over, what is assumed to be, a simultaneous end-to-end path from source to destination. This however is in contradiction with what actually happens in mobile networks. Nodes move around while changing associations with their environment constantly because there is no stable infrastructure. This is recognized in the increased interest in delay tolerant, or opportunistic networks. No fixed infrastructure is assumed and simultaneous end-to-end communication paths do rarely exist. Instead, mobile nodes, known as mules or ferries, carry data from cluster to cluster. Sensor data is collected on the way and is disseminated in an opportunistic fashion. This idea can be taken one step further. Crowd sourcing or urban sensing uses mobile phones as input devices in addition to existing infrastructures and depends on people’s mobility. A traditional devicecentric approach becomes a human-centric approach. A mobile phone user not only becomes a collector of data, she also filters and disseminates data and will act on data she receives, knowingly or not. Depending on a user’s needs, opportunistic alliances with other users sharing the
same context will be made to do the necessary sensing, processing and actuation. Personal data is mixed with local and global data, resulting in an enormous amount of information. One of the great challenges in opportunistic sensing is narrowing down this information into chunks that are useful for the users. Which data will be collected and how much? How is the data disseminated and routed through the network? How do we decide on a common context? How can we use opportunistic sensing in new and innovative applications?

CHALLENGES AND NEW POSSIBILITIES OF OPPORTUNISTIC UBIQUITOUS SYSTEMS

Context awareness is the underpinning of adaptive user centric ubiquitous computing system. Assuming a networking infrastructure delivering sensor data one common approach to context recognition is to use machine learning techniques to classify a set of sensor data into the relevant output classes. The context recognition chain is the set of processing steps used to infer the context from the sensor data. These includes techniques from signal processing, pattern recognition as well as reasoning.

Opportunistic sensing, however, requires to revisit these steps to deal with the opportunistic nature of the underlying substrate.

Conversely, new application domains can arise, as a few examples:

• Mobile crowd sensing
  Understanding in real-time crowd dynamics has great potential to raise the global awareness of first responders during emergency situations and rescue operations. The widespread availability of mobile sensor-enabled devices calls for new methods making a decentralized use of sensor data to infer key building blocks of crowd behavior. From an emergency scenario perspective, the availability of central infrastructure cannot be assumed and an application must resort to decentralized participatory sensing. Nevertheless, through repeated interactions with members of a crowd and identification of pairwise correlations between sensor data a global picture about the crowd structure may be inferred. The underlying challenge amounts to a combination of opportunistic sensing and information spread in the crowd, and information processing.
• Urban Sensing
  Mobile phones, vehicle-mounted sensors and infrastructure in urban environments combined with publicly available information, like weather, train and bus schedules or traffic, enable users to sense, record and share their personal environment. This gives rise to new and innovative applications,
  e.g. social networking, traffic management and car guidance systems, or mapping environmental parameters, like air pollution.
• Activity recognition in opportunistic sensor configurations
  Activity recognition consists in inferring the user’s activities from the data of a set of sensors placed on the body. Current approaches usually require a specific deployed infrastructure for a specific recognition task. This is not desireable for a widespread use of context-aware systems. Users are at times
  in highly instrumented environments and at other times in places with little sensor infrastructure. Users carry various sensor enabled devices, such as mobile phones, watches, or smart garments. As the user leaves devices behind, picks up new ones and changes his outfit, the type and location of onbody sensors varies, and the sensing environment changes. Besides, in highly challenging situations data loss are not unexpected. Thus, a more flexible opportunistic approach to activity and context recognition is desired.
• Smart Appliances and Companions
  Digitally enabled devices acting in spontaneous wireless ensemble configurations, able to adapt to changing service requirements and situations (like positioning of mobile phones, cooperative wearables, implicit wearable to infrastructure
  interaction)
• Traffic, Transportation and Logistics
  Collective vehicular settings (car crowds) and self-aware goods, acting and requesting services while in transit through changing assistance infrastructures (e.g. Web of Things).
• Human Assistive Technologies in Extreme Situations
  Service architectures and their autonomous, self managing service provision to humans with specific goals (e.g. wayfinding, guidance), in extreme situations (e.g. rescue, evacuation) where the setup of available instrastructure cannot be
  foreseen at design-time and is likely to change over time.
  

The methods in the chain are not codified. Various methods may be used, but one common processing structure consists of:
data acquisition from the sensors; segmentation of the data stream into sections of interest likely to contain an activity; feature extraction within these section to reduce their dimensionality; classification of the features into a set of output
classes (activities); and “null-class” rejection. Challenges of opportunistic context awareness The classical context recognition chain assumes a design-time definition of the input space, that remains constant throughout use. By input space, we mean the number of sensors, their position, modality, sample rate, latency, as well as their physical sensing characteristics. These assumptions are not guaranteeed from an opportunistic sensing perspective.

TOPICS OF INTEREST

• Sensing & networking
  • Zero configuration
  • Spontaneous sensing
  • Technologies and implementations
• Context frameworks and architectures
  • Self-*-enabling architectures
• Context processing in opportunistic setups
  • Context reasoning
  • Context processing given Quality of Context
• Opportunistic signal processing and machine learning
  • Compressive sampling
  • Opportunistic ensemble classifiers
  • Imputation techniques
• Applications
  • Deployment and test-beds
  • Experiences and lessons learned
  • Tools
  • Analysis and evaluations
  • Methodologies for opportunistic context recognition
• Case studies
  • Urban sensing
  • Environmental monitoring
  • Activity recognition
  • Social networks
  • Crowd sensing

IMPORTANT DATES

• Submission Deadline: July 31st, 2010 [EXTENDED]
  
• Notification of Acceptance: August 16th, 2010
• Camera-ready deadline: September 6th, 2010
• Workshop date: September 26th, 2010

FORMAT

The workshop is a “traditional workshop format” with submissions orally presented by the authors (optional complementary poster alongside accepted papers are welcome).

Ample room will be left for discussion.

Participants will be selected based on paper submissions of 4 pages around the listed topics of interest. They should be written in UbiComp 2010 workshop format and sent as an attached PDF file to daniel.roggen@ife.ee.ethz.ch.