My group studies machine social intelligence: building AI agents that can understand, reason about, and interact with humans and other agents in complex, open-ended environments. We are particularly interested in agents that learn from embodied and social experiences, build models of the world and other minds, and cooperate with humans and machines over long horizons.

• Online World Modeling for Long-Horizon Planning: Developing agents that construct and update world models for planning in partially observed, dynamic, and long-horizon environments. Related work: 3D-Belief; World-in-World (ICLR 2026 Oral).

• Open-Ended Agent Modeling and Multimodal Theory of Mind: Building computational models of agent behavior, enabling AI systems to reason agents' mental states and social relationships from multimodal inputs in open-ended settings. Related work: MindZero (ICML 2026); AutoToM (NeurIPS 2025 Spotlight); MuMA-ToM (AAAI 2025 Oral); MMToM-QA (ACL 2024 Outstanding Paper Award).

• Multi-Agent and Human-AI Cooperation: Studying how agents communicate and collaborate effectively with humans and other agents in shared physical and virtual environments. Related work: GOMA (IROS 2024); NOPA (ICRA 2023).

• Continual Learning from Embodied and Social Experiences: Enabling agents to continously acquire new knowledge and skills through long-term interaction with humans, other agents, and the physical world. Related work: AgentOdyssey; ThoughtTrace (ICML 2026 RLxF Best Paper Award).

• Large-Scale Open-World Simulations: Creating scalable simulation environments for studying embodied and social reasoning, cooperation, and open-ended learning in rich multi-agent worlds. Related work: SimWorld (NeurIPS 2025 Spotlight); SimWorld-Robotics (NeurIPS 2025).

For more recent research, please check out my lab website.