Solfa
Solfa
Research [ 03 ] Embodied social world model

Social Situation Understanding and Simulation

The same motion means entirely different things across different identities, relationships, and settings.

Published
Author
Chen Li
FIG. 01The shape of the problem
The question
How can AI understand roles, relationships and actions in a shared situation?
Where it stands
A research direction, not a result. We are building the representation; nothing here has been measured or validated on a physical robot.
What the figure is
The figure is a diagram of the problem, not a recording. The explanatory film for this direction is being re-cut and is not published yet.

Our work on embodied social intelligence asks how AI can understand a shared situation. We study an embodied social world model through robot interaction and simulation: people's goals, roles and relationships matter alongside space, objects and motion.

It has to help the robot answer three questions: What is happening right now? What might happen next? How should I take part?

These questions rarely have a single answer. A robot has to keep judging under incomplete information, update its understanding as new evidence arrives, and actively ask people to confirm when uncertainty runs too high.