Research Scientist - World Model

Research Scientist on the World Models team developing next-generation world model architectures (diffusion, transformer, autoregressive, or hybrid) with a focus on controllability and physical consistency. You will define and own metrics (physical fidelity, long-horizon coherence, action-following) and run scaling studies to understand compute/data/architecture trade-offs. Expect to publish at the frontier and contribute to an open-source release as the long-term deliverable. The role centers on inventing world-model architectures, building controllability mechanisms for agent interaction, and advancing models that can be rolled out as interactive worlds for embodied reasoning. Strong emphasis on large-scale generative modeling, model-based RL, and multi-node training in PyTorch; prior work in world models, neural simulation or 4D scene representations is preferred. The team is part of Luma, aiming to build unified general intelligence capable of perceiving and interacting with the physical world.