Why Social Embodied AI?

Our workshop advocates for socially capable embodied agents by integrating embodied learning and robotics, interaction-centered modeling, and computational social intelligence. We aim to move beyond physical task completion toward agents that perceive, reason, and act appropriately under social cues, implicit intent, and interaction norms.

Social Embodied AI addresses the critical gap between autonomous navigation and manipulation capabilities and the complex requirements of human-robot interaction in real-world social environments. Success in social settings requires more than technical proficiency—it demands understanding of human intentions, cultural norms, and context-dependent social behaviors.

Core Challenges

• Challenge 1: Intention and Belief Understanding
  • Human Intention Prediction: Infer a partner's current goal online from sparse and noisy multimodal cues to support anticipation.
  • Belief State Inference: Represent what a partner knows and believes, including epistemic gaps and false beliefs, rather than assuming shared knowledge.
  • Goal Hierarchy Inference: Model layered objectives where social priorities can preempt ongoing routines, enabling interruption and replanning.
  • Emotional State Inference: Estimate affect and cognitive load from behavioral signals and adapt timing, communication, and assistance level.
• Challenge 2: Collaborative Adaptation
  • Knowledge and Awareness Estimation: Track what others can perceive, know, or access, and close information gaps through communication rather than exhaustive search.
  • Role Inference: Capture implicit and shifting roles such as leader, follower, or advisor that regulate initiative during joint activity.
  • Complementary Action Prediction: Anticipate actions that advance shared task structure and divide labor, avoiding redundant imitation.
  • Interaction History Modeling: Build persistent partner models from repeated encounters, enabling personalization and smoother coordination with fewer explicit instructions.
• Challenge 3: Social Norms and Affective Awareness
  • Attention and Focus Estimation: Model engagement and interruptibility from behavior and context, supporting well-timed communication and assistance.
  • Social Norm Recognition and Adaptation: Identify and adapt to societal, cultural, and contextual social rules, including unwritten conventions that shape predictable and appropriate behavior across settings.
  • Preference Adaptation: Adapt to user-specific needs, privacy conventions, and routines, mapping these preferences to actionable choices such as proximity, approach angle, and timing.