April 2026 - In this interview, Roland Hechwartner summarizes the latest developments from oneM2M’s 74th Technical Plenary. In addition to chairing oneM2M’s Technical Plenary (TP), Roland is responsible for the coordination of the overall management of the technical work within the TP and its Working Groups (WGs). He is also a representative of Deutsche Telekom (DT).
Q: Would you begin with an overview of the key developments at TP#74?
RH: With all that is happening in the AI sector, there is growing interest at the intersection of AI and IoT. Not surprisingly, our members had many issues to discuss including, the industry trends that are driving AI integration in IoT. Members voted in favour of a new Work Item (WI) to investigate several scenarios and the capabilities that oneM2M standards offer to enable AI integration. We can go through the details later in our conversation. Aside from the AI related activities, I can update you on developments in oneM2M’s three working groups, and progress on the ESTIMED project which is all about Edge-IoT.
Q: Let us begin with AI market trends. What are oneM2M’s members seeing and how do they connect that with oneM2M’s IoT framework?
RH: There are three trends our members see in the market. Firstly, there is growing demand for AI integration in IoT because the two fields are converging rapidly. Any developer building AI-powered applications on top of oneM2M platforms will benefit from concrete guidance on how to leverage oneM2M's semantic and resource management capabilities for AI model training and inference.
Secondly, there are new development around Large Action Models (LAM) for IoT device control. This is a result of recent advances in large language models (LLMs) and LAMs which have opened new possibilities for natural language-based device control. However, a critical bottleneck is the lack of structured, semantically rich training data for IoT device operations. oneM2M’s Smart Device Template (SDT) and oneM2M ontology, including the SAREF ontology, provide exactly the structured device descriptions needed to address this gap. A developer guide on this topic is essential to bridge AI practitioners and oneM2M IoT specialists.
Thirdly, with an eye to distributed IoT systems, there is a need for federated learning using native oneM2M resources. We notice that there is a trend to adopt federated learning for privacy-preserving, distributed model training in IoT deployments. oneM2M's hierarchical architecture that connects common service functions (CSFs) in cloud platforms, gateways, and IoT endpoints (IN-CSE, MN-CSE, ASN-CSE in oneM2M terms) naturally maps to the federated learning topology of a central aggregation server and distributed edge training nodes. Using native oneM2M resources (application entities, containers, contentInstances) to manage the full federated learning lifecycle, without external middleware, is a compelling capability that requires detailed developer guidance.
Q: What existing capabilities in oneM2M lend themselves to tackling AI’s integration in IoT?
RH: We are approaching the work scope by focusing on three scenarios. The first is to combine ontological frameworks with Large Action Models. The second scenario deals with the application of oneM2M resources to support federated learning. And the third scenario involves the use of coding agents to automate the development of Interworking Proxy Entities (IPEs) in multi-protocol IoT networks.
Q: Let us go through these one by one. Would you begin with ontologies and LAMs?
RH: Yes indeed. This scenario involves the use of oneM2M’s Smart Device Template (SDT) and ontologies (SAREF and oneM2M base ontology) in oneM2M for Large Action Models (LAM). Let me remind readers that the SDT is a framework that allows for the development of standardized models for various IoT devices. SDT provides a structured way to represent device functionalities and characteristics, facilitating communication and management within IoT systems. Think of it as a tool that enables interoperability and easier integration across different applications.
We intent to develop a guide that describes how to apply SDT and ontologies, i.e. SAREF and oneM2M base ontology, within the oneM2M platform to enable reliable natural language-based IoT device control using Large Action Models (LAMs). There are four components to this:
- Automated Semantic Restoration will map legacy SDT attributes to SAREF concepts via hierarchical context embedding and inject inferred SAREF URIs into the ontologyRef attribute of oneM2M <flexContainer> resources for zero-shot device provisioning.
- Knowledge Flattening will convert verbose SDT XML/XSD schemas into LLM-friendly system prompt templates.
- Reverse Synthetic Data Generation will algorithmically enumerate valid oneM2M control primitives as ground-truth outputs and uses a Teacher LLM with diversity prompting to back-generate natural language commands for Low Rank Adoption (LoRA) adapter fine-tuning.
- Semantic Guardrails will provide a runtime validation layer that enforces SDT-defined type, range, and enumeration constraints on LLM-generated oneM2M JSON primitives with a self-correction feedback loop before execution.
Q: Next, what does the application of oneM2M resources to support federated learning entail?
RH: For this scenario, we are looking to produce a guide that describes how native oneM2M resources - AEs, <container>, and <contentInstance> - within IN-CSE and MN-CSE nodes can implement a complete federated learning lifecycle without external machine learning middleware.
The IN-AE at the IN-CSE acts as the global model aggregator, while MN-AEs at each MN-CSE perform local training on locally stored sensor data, ensuring raw data never leaves the edge node. The entire federated learning process — round initiation, local training, model upload, FedAvg aggregation with Z-score outlier detection, and global model redistribution — is managed solely through standard oneM2M CREATE, RETRIEVE, and NOTIFY operations.
Q: And lastly, what is involved in applying coding agents for multi-protocol, IPE development?
RH: As before, we are planning to develop a guide that describes how to design a coding agent that automatically generates, tests, and deploys IPEs that connect heterogeneous IoT devices running protocols such as MQTT, Sensor Things API, or Zigbee to a oneM2M CSE.
There are two aspects to this scenario. One is to lower the integration barrier whereby a coding agent abstracts the complexity of oneM2M resource modelling and protocol bridging. This will enable developers with little or no prior oneM2M knowledge to produce functional IPEs through natural language instructions.
The second aspect deals with self-evolving IoT Networks. The aim is to equip coding agents so that CSE-connected networks can autonomously extend their device coverage by generating and deploying new IPEs. This would be in response to the discovery of previously unknown device types or protocols. We want this to happen without manual developer intervention.
In the past, we have talked about modernising standards development for the software era. Scenario#3 is a real possibility because of how we have transformed the standardization process through fully machine-readable specifications and training material in the form of code examples.
Q: How will oneM2M take these ideas forward?
RH: In practical terms, the TP approved the new Work Item (WI) on the topic of a “Developers' Guide for AI Enablement and Usage in oneM2M”. The organizations formally supporting this work item are Sejong University, SBS, Exacta GSS, KETI, and Deutsche Telekom. The objective of the work item is to produce a series of Technical Reports (TRs) that provide developer guidelines for AI enablement and usage within the oneM2M service platform.
The TRs cover advanced AI-related topics including the following:
- use of oneM2M device semantic descriptions and ontologies for AI model development,
- use of native oneM2M resources for distributed AI model training, with a focus on practical implementation guidance such as use cases, message flows, and resource description samples.
This WI will involve a continuation of TR-0045 on semantic topics. TR-0045 provided an introductory guide to SAREF usage in oneM2M. The proposed guide on Large Action Models goes significantly further. It will demonstrate advanced semantic exploitation for AI training data generation, which is not covered in any existing oneM2M TR.
Q: Is oneM2M also looking at how AI Agents might interact with machine readable specification documents?
RH: Yes, that is correct. Another work item objective is to leverage the git-based specification development strategy that we launched to modernize our standardization activities. One task is to describe how coding agents can use machine-readable specifications for automated implementation of software products.
For the new WI, we will also address the use of coding agents to deploy interworking proxy entities (IPEs). This is important because IPEs are a mechanism to connect disparate IoT sub-systems, including legacy and proprietary technology solutions that an organization might want to connect as part of a larger or distributed IoT system.
Q: What progress can you share from the oneM2M Working Group sessions at the TP?
RH: The Requirements & Domain Models (RDM) working group, chaired by Massimo Vanetti (SBS), made progress on the recently approved work item on FlexContainer-based Device Management Maintenance. There were discussions on input contributions about field devices using SDT as well as next steps towards the related Technical Specification (TS-0043). The group’s focus is on planning for Release 6 deliverables.
Peter Niblett (Exacta) chaired the System Design & Security (SDS) working group which made progress on several open documents. These include TS-0001 on the Functional Architecture, TS-0004 on the Service Layer Core Protocol, and TS-0026 on 3GPP Interworking. The group also continued work on three Release 6 TRs. These were: TR-0077 on oneM2M and MEC integration scenario and mechanisms, TR-0079 on Robot Operating System (ROS) interworking, and TR-0081 on - AI Agent Interworking.
There was solid progress on Markdown conversions of the TSs and the availability of baseline documents on git. The goal to have all Release 5 TSs available on git by TP#75 in June. However, the achieve timeline goal for Release 5, the completion of four work items will be deferred to Release 6, with a target completion date set for Q3 2027.
Bob Flynn (Exacta) chaired the Testing & Developers Ecosystem (TDE) working group. Their primary activity was to continue progress and make enhancements to the git process. Their discussions flagged innovative ideas to improve the handling and look-and-feel as well as ideas how to include other than permanent documents (TS, TR, WID) to this new git-based approach.
As part of the newly approved work item on AI, the group discussed three developer guides and agreed on their initial versions. These are TR-0083 - Developer Guide for the Use of SDT and SAREF in oneM2M for Large Action Models, TR-0084 – Developer Guide for the Use of oneM2M Resources to Support Federated Learning, and TR-0085 -Developer Guide for the Use of Coding Agents for Automated IPE Development.
The TDE group also discussed and agreed on a suggestion to host a developer event in the week prior to the TP#76. This oneM2M internal event usually focuses on compliance and interop testing of existing implementations.
One final TDE discussion focused on planning for a hackathon involving the ESTIMED project. The hackathon will take place from October to November 2026. oneM2M and ESTIMED representatives will agree on the scope and evaluation criteria for the ESTIMED project in near future.
Q: To close, what are the plans for oneM2M’s next Technical Plenary?
RH: The hosts for the next meeting, TP#75, are the University of British Columbia (Canada) and Sejong University (South Korea). The event will take place at the UBC campus in Vancouver, Canada from June 1-5, 2026. The main goals will be the finalization of Release 5 including the shift towards full git-based specification development, and progress of Release 6 work items.
