Page 129 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

Box 6: ITU standardization activities on embodied AI

ITU-T Study Group 21 Part 3: Future AI
ITU-T SG21 is actively developing standards that shape the foundations and evaluation
of Embodied Artificial Intelligence (EAI).

Two following core work items directly target EAI by defining system capabilities
(perception, decision-making, action, and interaction) and establishing standardized
benchmarking frameworks:
• F�RF-EAI (Requirements and framework for embodied artificial intelligence
systems);
• F�EAI-AC-BK (Assessment criteria for Embodied Artificial Intelligence system:
Benchmark).

The following approved ITU-T Recommendation contributes to embodied AI interaction
scenarios:

• F�746�9 (Requirements and architecture for indoor conversational robot system).
The following complementary efforts address enabling technologies. These introduce
evaluation methods for foundation models, including aspects of visual embodied
intelligence and multimodal reasoning critical to Embodied AI:

• H�FDM-AC-Gen (Assessment criteria for foundation models: General);
• H�FDM-AC-V (Assessment criteria for foundation models: Vision).

Domain-specific applications are being studied in F.MTTIR, which operationalizes
embodied perception-action loops in inspection contexts for safety and efficiency:

• F�MTTIR (Requirements and framework of computer vision and audition-based
transportation tunnel inspection robotic systems).

Additionally, SG21 is advancing ontology-based standards to enable context
modelling and interaction in intelligent spaces, supporting adaptive and interactive
EAI deployments:
• F�DCOR-reqs: (Requirements for construction of ontologies knowledge base for
household service robots in intelligent spaces)
• F�DCHE-Reqs: (Requirements for construction of ontology knowledge base of
home environment for robot interaction in intelligent space)
• F�DCHU: (Requirements for construction of home user profile ontology
knowledge base for robot interaction in intelligent space)

Collectively, these initiatives position SG21 as a key driver in defining requirements,
benchmarks, and enabling ecosystems for embodied intelligence.

117