About S.O.C.I.A.L. HML
What This Is
S.O.C.I.A.L. HML is an independent framework and pattern language for designing inspectable, socially governed, modern-AI-compatible Human-Robot Interaction architectures.
It combines:
S.O.C.I.A.L. principles
HML — HRI Modeling Language
HRI design patterns
Layered context-to-action architecture
Modern AI integration patterns
The goal is to help researchers, engineers, and system architects describe how a robot moves from raw interaction evidence to socially validated, styled, and embodied behavior.
In simplified form:
Human, Scene, and Robot Context
→ Context Management and Reasoning
→ Social Planning and Behavioral Synthesis
→ Natural Actuation and Embodiment Binding
→ feedback back into context
The framework is intended for systems that must be more than capable.
They must also be:
inspectable
debuggable
correctable
socially aware
policy-governed
memory-aware
compatible with modern AI
Origin and Motivation
This framework was written from the perspective of one year after the completion of the Israeli HRI consortium, a three-year applied R&D effort that ran during 2022–2025.
That period was unusually consequential for AI.
During those years, the field moved rapidly from architectures built as explicit pipelines with AI components inside them toward systems where more and more logic, code, interpretation, and behavior appear to dissolve into model weights.
This site is not a documentation of the consortium.
It is also not an official output of the consortium.
Rather, it is an independent synthesis of architectural lessons:
what worked well
what was difficult to scale
what became clearer only after reflection
what should be preserved
what should be redesigned
how modern AI changes the architectural tradeoffs
The practical experience of building HRI architectures made one point especially clear:
Human-facing robots need both capability and governance.
Modern models can provide impressive capability.
But HRI systems still need explicit mechanisms for context separation, memory governance, social validation, human correction, actuation control, and feedback.
Why Now
The rise of foundation models, VLMs, LLMs, agents, RL policies, and end-to-end robotics systems creates a real architectural tension.
On one side, more monolithic learned systems can be powerful, adaptive, and fluent.
On the other side, HRI requires properties that do not automatically emerge from capability alone:
clear responsibility boundaries
inspectable intermediate state
bounded authority
semantic memory governance
social and safety validation
human override
explainable feedback loops
S.O.C.I.A.L. HML emerged as an attempt to reason about this tradeoff.
The framework does not reject modern AI.
It also does not advocate a simple return to rigid classical pipelines.
Instead, it proposes a hybrid architectural discipline:
use modern AI where it is strong
connect it through explicit semantic interfaces
bound its authority through gates and policies
make commitments inspectable
preserve feedback and correction paths
A central position of this work is:
Modern AI does not make HML obsolete.
It makes semantic interfaces, responsibility boundaries,
and layered validation more important.
Main Contributions
This documentation presents four main contributions.
1. S.O.C.I.A.L. Principles
The S.O.C.I.A.L. principles define architectural properties that remain important in the foundation-model era:
S — Separated Contexts
O — Open Declarative representations
C — Clear Cognition
I — Interpretable Gates
A — Adaptive Autonomy
L — Layered Validation
These principles are not only values.
They are translated throughout the documentation into concrete mechanisms such as HIFs, semantic cells, policies, gates, memory structures, queues, and feedback loops.
2. HML — HRI Modeling Language
HML is a lightweight modeling language for representing semantic flow in HRI systems.
It is built around:
HIFs — HRI Interaction Frames
semantic cells
λ experts and operators
policies
gates
memory
queues
feedback
HML does not replace ROS, behavior trees, state machines, or agent orchestration frameworks.
It operates at a higher architectural level: the level of semantic interaction flow.
3. HRI Design Patterns
The documentation presents a catalog of reusable design patterns for HRI architectures.
These patterns cover:
human context
scene context
robot context
context management and reasoning
semantic memory
task readiness
social validation
social planning
behavioral styling
actuation
feedback
The purpose of the pattern catalog is to provide reusable architectural solutions rather than one fixed implementation.
4. Modern AI Integration
The documentation explicitly addresses how to integrate modern AI into inspectable HRI systems.
It covers:
LLMs
VLMs
multimodal models
agents
RL policies
end-to-end models
video models
learned motion policies
The key idea is that modern AI components can serve as bounded experts, critics, candidate generators, fallbacks, shadow models, and learned executors inside HML patterns.
They should not silently own the whole HRI loop.
Author
This framework was created by Dr. Eliahu Khalastchi.
Eliahu is a computer scientist and HRI / robotics researcher with experience in anomaly detection, autonomous systems, AI for robotics, and human-robot interaction architecture.
He led the architectural work on the HRI toolkit within the Israeli HRI consortium.
This site reflects an independent synthesis of architectural lessons and research thinking developed after that work.
Relationship to the Israeli HRI Consortium
The Israeli HRI consortium provided an important practical context for the experience behind this framework.
However, this documentation is not a consortium deliverable, official report, or implementation manual.
It does not attempt to document one specific system.
It generalizes from architectural lessons learned while working on HRI systems and asks:
Which architectural ideas remain useful?
Which patterns can be reused?
Which assumptions need to change in the foundation-model era?
How can modern AI be integrated without losing transparency and governance?
This distinction is important.
The framework is presented as an independent research-oriented pattern language, not as a description of a particular consortium implementation.
Academic Paper
This documentation is part of an ongoing research effort.
A condensed academic version of the framework is being prepared for submission, with a public preprint planned after the documentation site is publicly available.
The paper will focus on the conceptual contribution of S.O.C.I.A.L. HML as a pattern language for inspectable HRI architectures in the foundation-model era.
The documentation site serves as the extended, visual, and practical companion to that academic work.
How to Read This Documentation
A suggested reading path is:
1. Start with the S.O.C.I.A.L. principles.
2. Learn the basic HML concepts: HIFs, λ experts, semantic cells, memory, and gates.
3. Browse the design patterns by layer.
4. Review the Modern AI section to understand how LLMs, VLMs, agents, RL, and end-to-end models fit into the architecture.
5. Use the pattern catalog as a reference for designing or analyzing HRI systems.
For a fast conceptual entry point, start with:
HML Overview
Modern AI Integration Overview
Practical Integration Patterns