Unified streaming embodiment
Bring interaction, navigation and manipulation into a shared architecture. Unified pretraining and a multi-rate runtime connect multimodal representations to speech and whole-body motion.
RESEARCH DIRECTIONS
From perception
to purposeful action.
Exploring how intelligence can understand its environment and participate in it.
◎
Full-duplex realtime agents
Our real-time agents combine streaming speech and actions with tool use, priority-based events, multi-person interaction and interruption recovery.
Social perception & multimodal memory
Connect who is present, what is happening and what has happened before. Context supports expressive interaction and a persistent agent’s goals across time.
Decision models & agent infrastructure
Decision models provide structured judgments, selections and scores. Voice infrastructure and the Agent Harness support the surrounding workflow, from data and compute management to coding and agent orchestration.