Semantic Approach to Dynamic Coordination in Autonomous Systems

Semantic Approach to Dynamic Coordination in Autonomous Systems Artem Katasonov* and Vagan Terziyan University of Jyväskylä, Finland ICAS, Valencia April 24, 2009 * Artem Katasonov is at present with VTT Technical Research Center of Finland University of Jyväskylä

Coordination g Coordination is one of the fundamental problems in systems composed of multiple interacting processes: – to avoid negative interactions, e. g. conflict over a resource. – to exploit positive interactions, e. g. to reuse a shareable result. g The pre-dominant approach has been to hard-wire the coordination mechanism into the system structure (locks, semaphores). g Re-planning instead of just waiting for a locked resource as well as exploiting positive interactions is possible only through additional hard-wiring: the program code of the processes must have incorporated some knowledge about the behavior of each other. University of Jyväskylä

Coordination in open systems g The traditional approach becomes insufficient when considering more open systems, where the processes and resources composing the system may be unknown at design time, or in evolving systems. g In open systems, we want computational processes to be able to reason about the coordination issues in their system, and resolve these issues autonomously (Decker and Lesser, 1995). g Ideally, we would like to allow ad-hoc interaction, where two standalone independently-designed systems are able to coordinate whenever a need arises. g One way towards this is to enable the relevant processes to communicate their intentions with respect to future activities and resource utilization (Tamma et al. , 2005). University of Jyväskylä

Tamma et al. work g Tamma at al. (Univ. Liverpool) developed an ontological framework for dynamic coordination. They stated the need for an agreed common vocabulary, with a precise semantics. – They provided an ontology that defined coordination in terms of agents carrying out activities involving some resources, which can be nonshareable, consumable, etc. – Then, they described the rules for checking for conflicts among activities: e. g. if two activities overlap in time and require the same non-shareable resource, they are mutually-exclusive. – They also described some possible coordination rules to be followed when a conflict of a certain type is detected. g The ontology of Tamma et al. is an upper ontology, i. e. an ontology which attempts to describe the concepts that are the same across all the domains of interest. University of Jyväskylä

Drawbacks of existing work g In Tamma et al. work, roughly speaking, the idea is to make the agents to communicate their intentions and actions using the upperontology concepts (i. e. ”resource”, ”activity”) rather than the domain ontology concepts (e. g. ”printer”, ”printing document”) and in so to resolve the problem of evolving domains or domains not fully known at design time. g Hard-wiring the domain ontology concepts into both communicating systems is substituted with hard-wiring the upper ontology concepts – Design-time alignment is still needed. – Cannot coordinate with agents, for which this is not done. g In cases, when domain ontology is shared, bringing every conversation down to the upper ontology sounds somewhat unnatural. University of Jyväskylä

Insights from the Semantic technology g On the other hand, the Semantic Web research explicitly addresses the possibility of multi-ontology systems. g The standard technologies of the Semantic Web, such as RDF Schema (RDF-S) and Web Ontology Language (OWL), on the level of hierarchies of classes, enable communications in which: – The sender of a message can express it in its own domain ontology and does not need to know any integrating upper ontology. – Only the receiver of the message has to know the upper ontology and to have access to a formal definition of the domain ontology of the sender that links the concepts from that ontology to the upper ontology. University of Jyväskylä

An example Enquirer org: Mary rdf: type person: Woman. org: Mary person: has. Son org: Jack. Responder SELECT ? x WHERE {? x rdf: type family: Mother} org: Mary human: has. Sex human: Female. Sex. org: Mary human: has. Child org: Jack. org: Mary rdf: type family: Mother. ? x = org: Mary family: Mother is a subclass of human: Human with the restriction that it must have a property human: has. Sex with the value human: Female. Se x and must also have at least one property human: has. Child. Query Data Domain ontology family: Definition of family: Domain ontology person: Upper ontology human: Upper ontology rdf: , rdfs: , owl: RDF-S / OWL rules University of Jyväskylä Definition of person: Woman is a subclass of person: Person which is in turn a subclass of human: Human. person: Woman has a restriction to have a property human: has. Sex with the value human: Female. Sex. Also, person: has. Son is a sub-property of human: has. Child.

Dynamic ontological coordination Agent 1 “I plan to x: Scan a book y: Bla on org: Ag. PS 4 e” • “I now z: Print a document on org: Ag. PS 4 e” • “Does Agent 1’s intention concern me? ” • “I do not know what x: Scan means. . ” Agent 2 Intention Data Domain ontology DO 1 Definition of DO 1 To interpret action intentions Domain ontology DO 2 Upper ontology: Coordination Upper ontology: BDI Upper ontology rules To resolve conflicts University of Jyväskylä Coordination rules Definition of DO 2 e. g. of Tamma et al. Beliefs-Desires-Intentions model

Upper ontology rules g The definition of a domain ontology of actions have to link it to the two upper ontologies (BDI and coordination), in a way that will enable the upper ontology rules to do the following: 1. Interpret an expression of a mental attitude conveying an action intention to obtain the identifier of the intended activity. 2. Match the activity description in the domain ontology definition with the intention to understand what resources will be utilized (or results produced) by the intended action. g E. g. in FIPA SL communication content language: (I (agent-identifier : name agent 1) (done (action (agent-identifier : name agent 1) (print some. pdf Ag. PS 4 e)))). – to extract the name of the activity ”print”. – then, from the definition of that activity, to understand that ”Ag. PS 4 e” is the identifier of the resource (printer) that is going to be utilized by the intended action. University of Jyväskylä

Ontology linking process No Yes Has unknown concepts? Yes No Is a registered upper ontology? Obtain the definition of own domain ontology in terms of the upper ontology Attempt one or both: • Download online ontology definition - may have to ask the sender for URL • Ask the sender for the ontology definition No Success? Yes Attempt ontology alignment Yes Done University of Jyväskylä Success? No Respond: NOT UNDERSTOOD

Semantic Agent Programming Language (S-APL) g It is a hybrid of semantic reasoners like CWM and agent programming languages like AFAPL. g RDF-based, uses Notation 3 syntax. g One of basic constructs: implication (just like in CWM): { ex: Weather ex: has. Condition ex: Sunny. ex: Weather has: has. Temperature ? t > 20 } => {ex: Current. Weather ex: is ex: good} g Another basic construct: executing an atomic Java component: {sapl: I sapl: do java: ubiware. shared. Message. Sender. Behavior} sapl: configured. As {p: receiver sapl: is “John”. p: content sapl: is “hello”} University of Jyväskylä

With S-APL: g Can define classes of activities: g Can then attach coordination-related properties: org: Scan coord: requires ? scanner. g g g Can write then interpretations rules – to understand what activities take place and what resources are involved. Can write then coordination rules – how e. g. to resolve a conflict if one is detected. Can also define static policies: University of Jyväskylä

Value g Enabling agents to coordinate without assuming any design-time ontological alignment of them: – agent can express an action intention using own vocabulary, – and through the process of dynamic ontology linking other agents will be able to arrive at a practical interpretation of that intention. g g g The definition of the domain ontology in terms of an upper ontology must be provided. However, such a definition is external to the agents and may be added later, when an agent is already in the operation. In result, an intelligent agent can potentially communicate with a ”stupid” agent, e. g. from a legacy system. It is also possible to connect two ”stupid” agents by putting an intelligent middleware in between. University of Jyväskylä

Limitations – Future work g The coordination ontology and the corresponding modeling syntax has to be extended with constructs that will enable describing the effect of an activity on a resource or an attribute of that resource. g Allowing some of an activity’s parameters to come from the background knowledge of the listener agent rather than from the message. g If an activity changes an attribute of a resource, the resource may undergo some follow-up changes due to environmental causes – the identification of conflicts has to be performed as reasoning or planning process rather than based on straightforward rules. University of Jyväskylä

Additional information on S-APL g S-APL web-page http: //users. jyu. fi/~akataso/sapl. html g Katasonov A. and Terziyan V. (2008) Semantic Agent Programming Language (S-APL): A Middleware Platform for the Semantic Web. In: Proc. 2 nd IEEE International Conference on Semantic Computing (ICSC'08), August 4 -7, 2008, Santa Clara, USA © IEEE, pp. 504 -511 University of Jyväskylä

16 University of Jyväskylä