Robot attention engine
Developed at Solfa to make attention visible. This demo shows the engine's gaze behavior on physical hardware.
Solfa · Social AI
AI that takes part in human society.
We study how AI can understand situations and people, building socially intelligent interaction systems that enable AI to communicate and collaborate with people in everyday work and life.
Powered by Solfa.Duet 1 modelIn-house research model · Commercial version in development
Beyond agency, toward social intelligence.
As AI becomes more agentic, everyday devices could act as agents across digital and physical spaces. Taking part means understanding people, coordinating with other agents, and responding to human guidance and feedback. Social intelligence becomes a foundation for this shared world.
AgentAI that takes part
Understanding & actionResponse & adjustment
EnvironmentShared spaces and social relationships
AgentsPeople and other agents
Today's AI reads the words. Ours reads what goes unsaid.
We can know more than we can tell.
Nobody can write down the rules for when to speak, but everyone follows them. Our brains read these cues without being told, and our bodies answer them without thinking. Social intelligence is the clearest case: most of it is tacit.
| Explicit AI | Tacit AI |
|---|---|
| Reads the words | Reads what goes unsaid |
| Acts on commands | Acts on cues |
| Works for whoever is using it | Knows its part among people and other agents |
Making progress, wobbles and all.
These demos are still works in progress. Through robots and simulations, we explore how people and AI pay attention, communicate and coordinate.
Solfa Duet 1 · Interaction model
The in-house multimodal real-time interaction model coordinates dialogue and motion, with generated motion retargeted to Reachy Mini and G1 in simulation.
Inside a social intelligence simulation
SOLFA
Welcome. Are you here for the experience?
Social simulation · Group interaction
The in-house scenario synthesis and simulation training for social intelligence explores who is speaking, whom to respond to and when to wait.
PARTNER 02
ASSEMBLE A DESK LAMP · CONCEPT SCENELet’s finish this lamp. Solfa has the shade.
Preparing the collaboration scene…
Social simulation · Joint assembly
The in-house scenario synthesis and simulation training for social intelligence explores how people and a robot coordinate to assemble a desk lamp.
What becomes possible when people and AI take part together?
Working together
Help teams coordinate tasks, share decisions and keep work moving through clear, natural interaction.
Less repeated confirmation, fewer execution errors and smoother handoffs.
Learning and play
Support learning, practice and play with timely responses that leave people room to explore.
More sustained practice, smoother experiences and stronger ongoing participation.
Service and entertainment
Support services and entertainment in shared spaces, responding to each person’s needs and pace.
More completed services, smoother staff handoffs and greater participation.
Solfa · Social AI
Build for the world we shared.
For teamwork, shared experiences or services in shared spaces, we invite partners to define the question, design experiments and test the value together.