OBJS Technical Note

Agent Standards Overview

Frank Manola Object Services and Consulting, Inc. July 1998

OMG MASIF

OMG's MASIF is intended to support high level interoperability among heterogeneous agent systems. The goals are to standardize:

agent management--e.g., creating, suspending, resuming, and terminating agents
agent transfer--(for similar agent system implementations) mechanisms for receiving agents and fetching their classes
agent and agent system names
agent system type and location syntax
agent tracking--tracing agents registered with naming services of different agent systems

It does not address:

multi-hop security
movement of agents between dissimilar agent systems (e.g., code conversion)
agent communication--this is left to implementors; no ACL is defined, nor are communication or negotiation protocols

MASIF defines a reference model which defines concepts such as:

agent system--a platform that can create, interpret, execute, transfer, and terminate agents.
agent system type--describes the vendor, language, and serialization mechanism (e.g., "Aglet" is an agent system type)
place--an exection environment in an agent system in which an agent can execute. A given agent system can support multiple places. Conceptually a location.
region--a set of agent systems that have the same authority, but not necessarily the same agent system type

MASIF defines the MAFAgentSystem interface to support agent management, agent transfer, and agent tracking functions. MASIF also defines the MAFFinder interface, which provides methods for maintaining a dynamic name and location database of agents, places, and agent systems.

The specification discusses its relationship with CORBA Naming, Lifecycle, Externalization, and Security Services.

Further work (for later RFPs) has also been identified, for example:

higher levels of interoperability among Java-based platforms, e.g., serialization formats and internal agent system capabilities
further IDL interfaces, e.g., for low-level agent communications
cooperation with FIPA, e.g., support of high-level communication (ACL) for MASIF agents
integration of further CORBA services, e.g., trader
refinement of the region and place concepts

The MASIF interfaces are defined at the agent system level, rather than at the agent level. Agent systems and agents may be, but are not required to be, CORBA objects. Hence, MASIF does not represent a thorough integration of agent technology into CORBA. However, when agents are defined as CORBA objects, they potentially have access to all CORBA services, any legacy software wrapped by CORBA objects, etc.

FIPA

The FIPA97 specifications define normative specifications for:

agent management (or agent platform services)

white pages via an Agent Name Server (ANS)
yellow pages via a Directory Facilitator (DF)
registration in a given DF defines a domain (agent community)
an Agent Platform defines a logical "place" containing an ANS, DF, management tools, and a collection of agents
interplatform communication takes place via an Agent Communication Channel (ACC), which defaults to CORBA IIOP

agent communication language

based on speech acts
has a formal semantics
also included are several predefined protocols (e.g., contract-net negotiation and auction protocols), and the concept of registered generic and application-specific protocols

agent-software integration

defines Agent Request Broker and Wrapper roles
allows an agent system to integrate non-agent software
details of how wrapper communicates with wrapped software are left to the implementor (a Wrapper is an agent, and communicates with other agents via ACL)

and several reference applications for:

personal travel assistance
personal assistant
network provisioning and management
audio/video entertainment and broadcasting

FIPA98 is extending these specifications, including work on:

agent management support for mobility (identifying the relationship between this work and MASIF is explicitly targeted)
an ontology service, supporting

translation of terms between different ontologies
downloading meanings of terms, axioms, and relationships between terms
querying for relationships between ontologies
uploading and updating of ontologies

additional applications, e.g., product design and manufacturing agents

FIPA does not currently constrain the low-level implementation of agents to any great extent, nor, except for defining agent platform services, does it constrain the infrastructure a great deal. Hence, there is not the sort of reification of the infrastructure (defining lower-level components as agents) currently going on in CORBA (e.g., with respect to portability issues).

Comments

The FIPA and MASIF work are somewhat complementary, in that FIPA has so far primarily worried about high-level agent to agent communication (ACL, negotiation protocols, ontologies), and has not said much about mobility, while MASIF has primarily considered mobility and, to some extent, issues involved in interoperability among heterogeneous agent systems, and does not say much about agent communication (at either high or low level). The MASIF effort is more a bottom-up activity (and doesn't necessarily assume its agents are "intelligent"), while FIPA is more top-down (however, FIPA Agent Communication Channels are required to support at least IIOP as a default method of communication).

Both standards discuss agent management functions and general agent reference models, including "place" and "platform" concepts, white and yellow pages, etc., and, at a relatively high level, they are not dissimilar.

DARPA has identified specific types of agent standards it feels are needed. These include:

Agent reference model

agent management and control
privacy, security, access control, and veracity

Agent-to-Agent communication

communication languages

knowledge sharing
goals and task descriptions

coordination protocols
security protocols

Agent-Software communication

communication (APIs) between agents and non-agent software

These general categories of standards seem roughly the same as those identified by FIPA. Of these, standards relating to security, and to goals and task descriptions, seem least developed. Both FIPA and MASIF address providing an agent reference model, and agent-software communication (MASIF somewhat indirectly). MASIF does not deal with agent-to-agent communication at all.

Both FIPA and MASIF steer away from trying to overly constrain agent implementation technology at this early stage in its development. The contrast in levels of detail is particularly clear in comparing what MASIF tries to specify compared with the level of detail in other OMG specifications. While on the one hand this reluctance provides a useful degree of flexibility in allowing the technology to freely develop, it also means that some obvious approaches to integrating the two technologies, e.g., definding CORBA interfaces for individual agents based on FIPA specifications, are left unspecified (although this would be straightforward for individual developers to do). On balance, given the early stage of agent technology, and the lack of experience with full-blown operational systems, this incremental approach to developing the standards is probablywise.

There is a great deal of potential synergy between the FIPA and OMG activities. One such area is that of OMG's domain-specific work, much of which could be usefully integrated into FIPA's work when completed. (Another area, of course, is the whole area of construction of FIPA agent technology using OMG distributed object technology as a base). FIPA is very much aware of these potential synergies; by the same token, the MASIF submitters are very much aware of FIPA's activities. Annex C of the ACTS baseline document <http://olympus.algo.com.gr/acts/dolphin/AC-baseline.html> discusses some scenarios for OMG/FIPA integration, and general mappings between MASIF and FIPA agent-related concepts.