The Role
As a member of technical staff, you will bridge the critical gap between passive video generation and active environmental control. You are an expert at the intersection of generative AI and reinforcement learning, focused on transforming video models from mere pixel-predictors into actionable, physically grounded simulators.
Job Responsibilities
- Architect and train action-conditioned generative models, enabling precise temporal control and manipulation of complex simulated environments.
- Develop structured latent representations that explicitly capture physical dynamics, positional information, and multi-object interactions.
- Integrate generative models with reinforcement learning pipelines, mitigating latent drift and ensuring stable policy learning in visually complex, generated environments.
- Design rigorous evaluation frameworks to test the physical accuracy, compositional generalization, and sample efficiency of model-based control methods.
Minimum Qualifications
- Extensive practical experience in Machine Learning, Reinforcement Learning, Robotics.
- Direct, hands-on experience building model-based RL systems, or action-conditioned video generation architectures.
- Deep technical understanding of latent dynamics, object-centric representations, and policy learning.
- Expert-level software development experience with deep learning frameworks (e.g., PyTorch, JAX) and large-scale training infrastructures.
Preferred Qualifications
- PhD in Computer Science, Artificial Intelligence, Robotics, or Machine Learning.
- A track record of significant research contributions in robotic manipulation within simulated environments (e.g., publications in top-tier venues like NeurIPS, ICLR, or CoRL).