Ontologies, Agents, and Plans
Clark & Parsia LLC
Outline
- Language
- OWL-S
- Issues with OWL-S
- SAWSDL
- HTN-DL: from formalism to submission
- Multi-agent HTN
- Infrastructure
- An end-to-end HTN-DL system
- The planning component
- The reasoning component (Pellet)
- Execution engine
- Modeling
Language: OWL-S
- Pros
- W3C submission
- "Mature"
- Tools, papers, examples, mindshare
- Expressive (HTN-DL captures an extended subset)
- Cons
- Development is moribund
- Complex model
- Syntax is a mess
- Semantics ill specified
- Tools are often poor or abandoned
- Extensibility difficult
Language: SAWSDL
- Pros
- Proposed recommendation at W3C
- IBM is using it
- Works with WSDL 2.0
- Some momentum
- Cons
- Radically incomplete
- No PEs, process model, etc. etc.
- Roughly the "grounding" in OWL-S
- Few tools
Language: HTN-DL
- HTN
- Operators (aka WSDL operations)
- Basic actions
- IOPEs, service types, non-functional properties
- Methods
- Decomposes an action into simpler parts
- Have IOPEs, types, and non-functional properties too
- DL
- To represent the state of the world
- SPARQL/DL like language for preconditions
- Service types and IO represented using ontologies
Language: HTN-DL
- HTN-DL a bit of a misnomer
- Pluggability
- Change state of the world language
- Change update method
- Change operator/method description language
- Change selection method
- HTN-DL is a flexible framework
- And associated systems should support that flexibility
Language: HTN-DL over SAWSDL
- WSDL describes operations
- Akin to operators
- SAWSDL adds "enrichment" hooks
- From an operator to a concept about that operator
- No direct support for PEs
- Could embed them in the pointed to formalism
- A la OWL-S
- From an XML Schema type to an OWL class
- OWL-S/HTN-DL could use SAWSDL
- Only small technical gain
Language: Multiagent HTN-DL
- Standard HTN is single (implicit) agent
- HTN-DL has more representational flexibility
- Operators and methods have rich descriptions
- Could route them to different actors
- "Centralized" planning for distributed action
- Information gathering could be used
- HTN-DL world state and domain are "open"
New Language vs. OWL-S
- OWL-S has problems
- Semantics
- Lots of OWL-S is underspecified and confusing
- We spent far too much time trying to make it all work
- Syntax
- Tooling
- We propose an alternative
- Clean slate but "conceptually compatible"
- Resulting in a new W3C submission?
- Could get people interested again
Infrastructure: An HTN-DL System
Infrastructure: The Planner
- Options
- Use/extend JSHOP
- Performance isn't terrific
- Doesn't have clean hooks
- Not a completely naive implementation
- Use/extend HTN-DL planner
- Better hooks
- Not very optimized
- More of a prototype
- Clean slate planner
Infrastructure: The Reasoner (Pellet)
- Pellet well suited
- Expressive (nominals, ABoxes, datatypes)
- Good query performance
- Incremental very helpful
- Things to improve
- Performance as always
- Incremental ABox behavior
- Deletes, esp
- Rule implementation
- Both expressivity and performance!
Infrastructure: Execution Engine
- Plans should decompose into agent instructions
- Coor dination points needed?
- Contingencies in general?
- Agents are services
- So SOAP endpoints?
- Plan routing?
- Other possibilities?
Modeling
- Domain descriptions
- Background knowledge
- "Glue"
About Clark & Parsia, LLC
DC-based R&D firm specializing in Semantic Web, web services, and advanced AI technologies for federal and enterprise customers.