Motivated by the goal of building open-ended agents in open-ended worlds, his research focuses on: (i) training agents to achieve long-term goals over extended periods of time, and (ii) training foundation world models through action-conditioned video generation.