Currently, I focus on world models for physical intelligence, embodied agents, and long-horizon decision making.

• World Model: I am interested in learning predictive models of real and simulated environments that can compress multimodal observations, infer dynamics, and support controllable generation for agents operating in the physical world.
• Model-Based Reinforcement Learning: I study how agents can use world models to plan, evaluate counterfactual futures, improve sample efficiency, and act safely under uncertainty across robotics, games, and open-ended decision-making tasks.
• Embodied and Multimodal Intelligence: I work on bridging language, vision, action, and simulation so that foundation models can build actionable representations of the world and generalize from digital environments to physical systems.