An Ontological Formulation and an OPM profile for Causality in Planning Applications
Download as PPTX, PDF2 likes551 views
The document presents a framework combining ontological formulation and Open Provenance Model (OPM) profiles for addressing causality in planning applications. It discusses the core elements of domain theories in automated planning, as well as a use case in crisis management training through the Pandora project. The research emphasizes the role of semantic web technologies in supporting domain theory modeling and automated checking.
An Ontological Formulation and an OPM profile for Causality in Planning Applications
- 1. Joint International Technology Conference (JIST2011) Hangzhou, China, December 5, 2011 An Ontological Formulation and an OPM profile for Causality in Planning Applications Irene Celino and Daniele Dell’Aglio CEFRIEL – Politecnico di Milano, Italy daniele.dellaglio@cefriel.it
- 2. Summary Introduction Planning metamodel OWL formalization OPM mapping Inference over the model Use case – PANDORA Conclusions and future work 5/12/2011 JIST 2011, Hangzhou, China 2 /16
- 3. Planning Problem Definition of sequences of actions to reach a desired goal Automated planning & scheduling in AI The task requires a domain theory – a model with the knowledge useful to generate plans Agents, actions, causal relationships, etc. Defined by a modeller Our research focus on helping the modeller in checking the coherence and rationality of the domain theory 5/12/2011 JIST 2011, Hangzhou, China 3 /16
- 4. Domain theory – core elements 1/2 Component: physical or logical subsystem of interest for the planning Controllable vs Uncontrollable Agent vs Resource Action: temporally tagged event Event: an action determined by the planner (related to a controllable component) Decision: an action taken by an uncontrollable component 5/12/2011 JIST 2011, Hangzhou, China 4 /16
- 5. Domain theory – core elements 2/2 Planning Rule: representation of actions’ causality – specifies the consequences of actions Reference Action Rule Targets: actions that could be “caused” by the reference action Rule Conditions: requirements on the actions involved in a planning rule, expressed through rule relations: Temporal conditions Constraints Assignments 5/12/2011 JIST 2011, Hangzhou, China 5 /16
- 6. An OWL2 Formulation Core elements actionTriggersAction Action hasActionValue isActionOf Component hasReferenceAction RuleTarget isRuleEffectActionOf hasReferenceComponent hasRuleTarget PlanningRule Rule ruleTriggersRule hasRuleCondition Condition The whole ontology is available at: http://swa.cefriel.it/ontologies/tplanning 5/12/2011 JIST 2011, Hangzhou, China 6 /16
- 7. An OWL2 Formulation Rule conditions PlanningRule Rule Temporal rule relations hasRuleCondition Condition were modelled using rdfs:subClassOf Allen’s Interval Algebra Assignment Constraint Temporal The three kinds of rule Condition Condition Condition conditions are defined rdfs:subClassOf rdfs:subClassOf extending SPIN vocabulary (SPARQL sp:Let sp:Filter Inference sp:variable sp:expression sp:expression Notation, http://spinrdf. sp:arg1, org/) ... sp:arg2, ... sp:Function 5/12/2011 JIST 2011, Hangzhou, China 7 /16
- 8. Open Provenance Model (OPM) Model for the tracking of the provenance of artifacts Three main concepts: opmv:wasDerivedFrom opmv:wasGeneratedBy opmv:was opmv:Agent ControlledBy opmv:Process opmv:Artifact opmv:used opmv:wasTriggeredBy OPM Profiles We mapped using the OPM Vocabulary (OPMV) to define an OPM Profile 5/12/2011 JIST 2011, Hangzhou, China 8 /16
- 9. Mapping the planning metamodel and OPMV opmv:wasDerivedFrom actionTriggersAction Action hasActionValue isActionOf opmv:Artifact Component hasReferenceAction opmv:used RuleTarget opmv:Agent opmv:wasGeneratedBy isRuleEffectActionOf hasReferenceComponent opmv:wasControlledBy opmv:Process hasRuleTarget PlanningRule Rule ruleTriggersRule hasRuleCondition Condition opmv:wasTriggeredBy 5/12/2011 JIST 2011, Hangzhou, China 9 /16
- 10. Checking of the domain theories Meta-model to represent domain theories Vocabulary Axioms It is possible to model domain theories using the ontology Inference processes on domain theories are available Semi-automated checking to the domain theories Extraction of relevant information from the model for the modeller 5/12/2011 JIST 2011, Hangzhou, China 10 /16
- 11. Domain theory checking Orphan elements 1/2 Orphan Extract from the planning component model the orphan elements: Components C2 Components not involved in C4 C5 any Action C1 C3 Actions not involved in any Planning Rule A4 A1 Actions Allow the modeller to check A2 A3 potential lacks or A5 shortcomings Planning rules P2 P1 Orphan action 5/12/2011 JIST 2011, Hangzhou, China 11 /16
- 12. Domain theory checking Orphan elements 2/2 5/12/2011 JIST 2011, Hangzhou, China 12 /16
- 13. Inference and automated checking Action reachability Reachable action: action with a target role in one or more planning rules Modeller is interested in finding: Unreacheble actions: actions generated by controllable components that are never target Actions triggered by the unrecheable action P2 An P1 Pn A1 A2 A3 triggers Unreachable Action: A1 reference A1 dependent actions: A2, A3 ... An target 5/12/2011 JIST 2011, Hangzhou, China 13 /16
- 14. Use case – Pandora Application in Simulation Learning for decision making in a scenario of crisis management Used in the Pandora EU FP7 project Realization of a platform for the training of gold commanders Planning is used to simulate learning sessions Support at the design time for the building of domain theories Additional info on: http://pandoraproject.eu 5/12/2011 JIST 2011, Hangzhou, China 14 /16
- 15. Conclusions and future work Use of Semantic Web in planning: Modelling of domain theories Semi-automated approach to verify the modelling: Tracking causality Check of elements involvement … Future work In-depth evaluation Relation with PDDL Analysis of executed plans 5/12/2011 JIST 2011, Hangzhou, China 15 /16
- 16. Thank you! An Ontological Formulation and an OPM profile for Causality in Planning Applications Daniele Dell’Aglio CEFRIEL – ICT Institute of Politecnico di Milano, Italy e-mail: daniele.dellaglio@cefriel.it web: http://www.cefriel.it 5/12/2011 JIST 2011, Hangzhou, China 16 /16