Connected devices can collect information and expose services, yet many still have limited ability to communicate autonomously across networks or coordinate with other systems. Semantic agent discovery offers one approach to this challenge: software agents describe their resources and capabilities in a shared, machine-interpretable form, then use that information to identify potential collaborators. A Linked Open Agent (LOA) model applies this idea to Internet of Things (IoT) ecosystems that span heterogeneous multi-agent systems.
For industrial automation teams, the model is relevant as a way to think about interoperability among connected objects and software agents. Its central contribution is not simply a new directory. It combines a semantic description of an agent with a workflow that can be completed at runtime, allowing agents to search for partners based on meaning rather than matching only literal terms.
Why semantic agent discovery matters
IoT environments may contain devices and agents built using different technologies and operating within separate multi-agent systems (MASs). Even when these components can exchange messages, differences in terminology or data structure can make it difficult to identify which agent offers a suitable resource or service. Syntactic matching can find identical labels, but it may miss relevant relationships when descriptions use different terms.
The Internet of Agents (IoA) concept addresses this by treating software agents as active participants associated with IoT objects. These agents can act autonomously and collaborate with others. To support that collaboration across heterogeneous environments, they need descriptions that other agents can interpret and a discovery process capable of using those descriptions.
How the Linked Open Agent model works
A LOA brings together two main elements: a Linked Agent Contract (LAC) and a Workflow for Agent Control (WAC). The LAC is a semantic description of the agent, while the WAC specifies the control workflow the agent uses in the IoT ecosystem. Both are stored with the agent and support its discovery and operation.
The Linked Agent Contract
The LAC functions like a service contract, but it uses semantic descriptions formalized through the IoA-OWL ontology. Its descriptors provide a common vocabulary for representing aspects of the agent and its environment. The ontology covers six areas: agent profile, context-aware data, social context, service ecosystem, agent artifacts, and IoT resources.
These descriptions are published as linked data. In a collaborative LOA setup, contract information is represented as RDF triples in a central repository. Agents can use this information to reason about available partners and retrieve descriptions associated with other agents. The shared semantic vocabulary is intended to make descriptions more useful across different MAS platforms.
The workflow and runtime discovery
The WAC directs the agent’s control actions. When an agent needs additional information or a collaborator, its discovery task checks whether the workflow is already complete. If not, it uses the agent’s LAC and selected descriptor values to query and reason over the IoA ecosystem. A selection step filters the results, after which the agent can update its workflow and social-circle information using the contracts of recovered partners.
Discovery is governed by a time constraint defined by the LOA, which determines how often the agent explores the ecosystem. Communication is provided through a REST interface that abstracts FIPA agent communications via URIs. The architecture also includes workflow execution and asynchronous response tasks.
What the discovery evaluation found
The model was evaluated in a collaborative community scenario comprising three MASs. Similar versions of the scenario contained 13, 100, and 1,000 agents. The evaluation compared semantic discovery with the JADE Yellow Pages and jUDDI, using a shutter-control agent as the example requester. The experiments varied the number of descriptors and descriptor values used to identify potential partners.
With one descriptor and two values, JADE and jUDDI each achieved a reported accuracy of 74%, while the IoA method achieved 75%. The semantic approach could use a “same as” relationship to find data that purely syntactic discovery did not identify. When additional descriptors and values were applied, the reported IoA accuracy rose to 77% and then 79% in the tested configurations. The study also reports that the expanded semantic criteria helped distinguish false-positive results.
These figures describe the specific modeled scenarios and discovery criteria used in the evaluation; they should not be treated as a general performance guarantee. They illustrate a key design point: discovery quality depends on selecting descriptors and values that appropriately express what an agent needs.
Practical implications for automation teams
For engineers and system integrators considering agent-based IoT architectures, the LOA model highlights several design considerations:
- Define shared meaning: Semantic contracts require consistent descriptions of resources, context, and capabilities, not just a common message format.
- Choose discovery criteria deliberately: The descriptors and values used in a search affect which partners are returned and how accurately they match the request.
- Plan for heterogeneous systems: Linking agent descriptions to a common ontology can support discovery across separate MAS environments.
- Keep workflows adaptable: A workflow that can be completed using information from discovered partners gives the agent a way to respond to changing collaboration needs.
The model also describes recovered partners as potential collaborators that an LOA can use to self-recover at runtime if current collaborators fail. This is a capability of the proposed approach, rather than evidence of performance in a particular industrial deployment.
Conclusion
Linked Open Agents combine semantic contracts, linked data, and runtime workflows to make partner discovery more meaningful across heterogeneous IoT ecosystems. The evaluation suggests that semantic relationships and carefully selected descriptors can improve discovery accuracy compared with the tested syntactic approaches. For automation decision-makers, the practical lesson is to treat interoperability as a matter of shared meaning as well as connectivity: clear, structured descriptions can help agents identify relevant resources and coordinate across system boundaries.
Access Official Technical PDF