Skip to main content
Human-Robot Interaction Architecture

S.O.C.I.A.L. HML Framework

A pattern language for inspectable, socially governed HRI.

From context and memory to socially validated behavior — with bounded LLMs, VLMs, agents, and learned policies.

Why this framework?

Human-facing robots need more than perception, planning, and control. They need explicit context, memory governance, social validation, timing, style, embodiment binding, and feedback.

S.O.C.I.A.L. HML provides a layered design language for making these responsibilities visible, testable, replaceable, and governable.

Core position

Modern AI does not make HML obsolete. It makes semantic interfaces, responsibility boundaries, and layered validation more important.

Watch the Overview

A short introduction to the S.O.C.I.A.L. HML Framework and its role in inspectable, socially governed, modern-AI-compatible Human-Robot Interaction architectures.

Architecture at a glance

HML models the full path from interaction evidence to socially appropriate, platform-specific robot behavior.

HumanSceneRobotContext
→
ContextManagementReasoning
→
SocialPlanningTiming + Style
→
NaturalActuationEmbodiment
feedback ↺
Modern AI and SOCIAL HRI

Foundation models as bounded experts

LLMs, VLMs, agents, video models, RL policies, and end-to-end systems can all contribute to HRI. The key is to connect them through HIFs, gates, semantic commitments, and feedback rather than giving them unbounded authority.

Read the Modern AI section
About this work

From HRI consortium lessons to a research framework

S.O.C.I.A.L. HML was created by Dr. Eliahu Khalastchi as an independent synthesis of HRI architecture lessons, informed by applied R&D experience and the rapid shift toward foundation-model-based robotics.

The documentation is being developed alongside an academic paper on inspectable HRI architectures in the foundation-model era.

About the framework

Short overview video

A short video introduction will be added here after the public site is available and can be used as source material.

Read the overview for now