A Semanticsbased Framework for ContextAware Services Lessons Learned and Challenges

1
A Semantics-based Framework for Context-Aware
ServicesLessons Learned and Challenges

• Theodore Patkos, Antonis Bikakis, Grigoris
  Antoniou, Maria Papadopouli, Dimitris Plexousakis
• Institute of Computer Science, FORTH, Greece

4th International Conference on Ubiquitous
Intelligence and Computing (UIC-07)
2
System Overview

• The CG-platform is a framework
• for developing advanced location-based and
  context-aware services,
• applying techniques and formalisms from the
  Semantic Web and Ubiquitous Computing domains.
• The system is a Context Pedestrian Guiding
  application for supporting user experience in
  indoor environments through personalized,
  context-aware and context-adaptive services.

3
Talk Outline

• Key Concepts
• Context-Awareness
• Mobile Ubiquitous Computing Systems
• Semantic Web
• System Presentation
• System Architecture
• Supported Services
• Challenges
• Distributed Planning
• Dynamic Planning
• Plan Sharing

4
CG Platform Architecture
CG - Client
CG - Server
XML
XML
HTML/SVG Interactive Map UI
Communication Module
CG - Database
Communication Module
CG Database Server
RDF XML Parsers
RQL/RDF
RDF XML Parsers
CG RDF Schema
Authentication Module
RDF
Cache
ICS-FORTH RDFSuite
Workflow Manager
5
CG RDF Schema Graph
6
Server - Client Communication

• XML message exchanges, adopting the notion of
  envelope.
• Each message includes information about the act
  of communication itself that allows the
  identification of communicating parties, message
  type and intended action.
• Message content can involve
• RQL data, which contain RQL queries to the
  database
• RDF data, which contain (RDF descriptions of)
  updates to the database

7
Server - Client Communication

• An example
• ltcg-messagegt
• ltsendergtdevice18lt/sendergt
• ltreceivergtCG-Serverlt/receivergt
• ltlanguagegtRQLlt/languagegt
• ltcontentgt
• SELECT Z, LastName, Room
• FROM XlnameLastName,
• Zarrangement.initiatedByX,
• ZlocatedAtRoom,
• ZfromStart,
• ZtoEnd
• WHERE Room LIKE G100 AND
• Start gt 2006-03-21 AND
• End lt 2006-03-22
• lt/contentgt
• lt/cg-messagegt

8
UI Design

• Action layer
• Dynamically displays context information from
  ontology

• Interactive Layers
• Service-driven

• SVGs
• Javascript enabled

• Map Layer
• Architectural design under different scales

• Java Applet
• Sophisticated procedures

9
Distributed Planning and Multi-agent Coordination
Challenges

• The vision of Ubiquitous Computing implies
  seamless collaboration of numerous devices
  working together to achieve common objectives.
• Even the more ordinary services of our platform,
  such as the management of a presentation in a
  meeting room may involve continuous cooperation
  between devices, as diverse as the room's
  audio/video equipment, the lighting dimmer, the
  lecturer's laptop etc.
• More complicated services require more
  sophisticated models of teamwork between devices
  that differ in capabilities, characteristics and
  resource limitations.
• Moreover, the expectancy of devices to
  participate in teamwork is not known beforehand
  and is dependent on parameters, such as resource
  availability, mobility, commitments etc

10
Distributed Planning and Multi-agent Coordination
Objectives

• Coordination does not imply either cooperation or
  reciprocation.
• The self-organization and autonomy of devices are
  important design goals.
• To approach the problem of coordinating the
  actions of multiple agents in a ubiquitous
  distributed environment, agents engage in
  Cooperative Distributed Planning, where they are
  endowed with shared objectives and
  representations, with the purpose to jointly
  develop and execute a plan in a coherent and
  effective manner.

11
Distributed Planning and Multi-agent Coordination
Relevant Works

• Generalized Partial Global Planning (GPGP)
• Commitment-based approaches, i.e., Cooperative
  Problem Solving (CPS)
• Extension of the Partial-order Causal Link
  Planning to explore the multi-agent plan
  coordination problem
• Logic-based approaches, i.e., Coalition Logic
• It is important to remember that despite the
  various strategies that have emerged over the
  years, it does not seem possible to devise a
  coordination strategy that works well under all
  circumstances.
• Optimal coordination is desirable but rarely
  feasible, because it generally requires
  substantial computation and communication
  overhead.

12
Complex, Dynamic and Uncertain Environments
Challenges

• The majority of services provided by our platform
  requires a certain degree of reasoning and plan
  management skills by the participating agents.
• For instance, an agent wishing to print a
  document must deliberate on whether to use the
  slow inkjet printer located near the user or the
  faster high resolution laser printer of the
  adjacent room.
• While trying to enhance our platform with more
  aspects of context, the complexity of the
  planning task becomes computationally
  intractable.
• The simplifying assumptions of the classical
  planning problem, such as deterministic, atomic
  and simultaneous actions, omniscient agents,
  static and closed environments, must be relaxed
  or completely eliminated.

13
Complex, Dynamic and Uncertain Environments
Objectives

• The ambient computing environment is an open and
  highly dynamic environment.
• Mobile devices connect and disconnect to the
  network, contributing services with durations
  that vary according to their expected presence in
  the environment and the availability of their
  resources.
• Actions, goals and sensor observations have a
  temporal dimension, whose duration may only be
  partially known in advance.
• The assumption of complete world knowledge can no
  longer persist agents do not know a priori all
  other entities that are present at a specific
  time instance nor can they communicate directly
  with all of them.

14
Complex, Dynamic and Uncertain Environments
Objectives

• They have limited perception to acquire knowledge
  about the world they live in and have to generate
  plans preserving a level of uncertainty on both
  the state of the world, the available actions to
  achieve certain state of affairs and the outcome
  of those actions.
• Even the fact of committing different agents to
  certain tasks cannot be guaranteed to hold, since
  agents might disconnect before plan generation
  completes or new and more beneficial
  opportunities might arise.
• The non-deterministic nature of the environment
  is emphasized by the recognition that not only
  agents, but also exogenous events occur, in
  unpredictable and concurrent manner, affecting
  the state of the world.

15
Complex, Dynamic and Uncertain Environments
Relevant Works

• STRIPS
• Action Theories (Situation, Fluent, Event
  Calculus)
• Action Languages (A, C, GC, K, E)
• MDPs, POMDPs
• We emphasize on the use of intuitive and
  mathematically correct formal frameworks, since
  they allow IS analysts to produce detailed,
  formal specifications of ambient computing
  processes.
• Most approaches still trade simplicity for
  reality, due to the increased complexity of their
  reasoning mechanisms.
• It is difficult to build a unified model that
  combines different phenomena, such as
  nondeterminism, concurrency, knowledge,
  continuous change etc.

16
Plan Representation, Evaluation and Sharing
Challenges

• We need to represent the capabilities and
  specifications of different devices, so that
  distributed services can be adapted accordingly.
• We assume that the existence of visitors
  possessing unknown mobile devices is going to be
  a common situation for our platform.
• Profiles of devices must be flexible enough to
  capture both complex actions and decompositions
  of them to primitive ones, so that planning
  agents can understand and combine them to
  distribute responsibilities during service
  execution.
• In addition, plan execution must be monitored
  online and contextual information should be
  considered when evaluating possible future steps.

17
Plan Representation, Evaluation and Sharing
Objectives

• We need to dynamically associate with plan
  evaluation parameters, such as resource
  preservation, goal prioritization, importance
  value of actions, even trust and privacy metrics.
• It is not enough to just build better planners,
  we also need to be able to recognize which
  planning problems and opportunities to consider
  in the first place.
• We need to be able to weight alternative
  incomplete plans and to decide among competing
  alternatives.
• We need to share common plan representations or
  ontologies for describing plans, goals and
  actions.

18
Plan Representation, Evaluation and Sharing
Relevant Works

• Plan Representation
• Hierarchical Task Networks
• Skeletal Plans
• Plan Evaluation and Monitoring
• Model-based Diagnosis
• Constraints and Important values temporal
  constraints (i.e., action duration), motivations,
  goals with priorities and deadlines
• Very few formal frameworks model complex actions
  and common plan representations.
• Planning based on context and privacy parameters.