Research Scientist/ Engineer - Diffusion Policies for Robotic Learning

Our client is a well-funded world model lab building generative, interactive environments that other companies build on top of. They're focused on making world models physically grounded and reliable in specific domains, where passive video data isn't enough.

This role uses robotics-style policy learning to generate the action-conditioned data that trains those world models. The work is about manufacturing training signal through control, not deploying robots in production. (This is a research role feeding world models, not a robotics product role.)

What you'll work on

• Training diffusion and flow-matching policies to generate action-conditioned interaction data for world models
• Building systems that collect data through control to cover dynamics that scraped video can't reach
• Designing training objectives and evaluation metrics for data quality and ...