Robotics is finally leaving the lab. At Momentum Research Labs, we are building the systems that turn promising models into capable machines in the real world. Our first product is a world-model based simulator for training and evaluating robot policies. As a Founding Research Scientist, you will lead the research behind trustworthy robot policy evaluation: the action-conditioned world models that generate robot rollouts and the reward models that capture task success and guide policy learning. Your work will help customers understand how their policies perform and decide what to improve before deploying them on real robots. This is an unusually broad, high-impact technical role for someone who wants to build, ship, and learn quickly at the frontier of physical AI. You will move fluidly between model training, evaluation design, software development, and customer collaboration. If you are energized by hard research problems, high ownership, and seeing your models help robots operate in the real world, we would love to talk. What You Will Do • Lead the training and research of action-conditioned world models that power our simulator and policy evaluations. • Craft reward models that capture task success and provide useful learning signals for robot policies. Validate them against human judgments and task outcomes. • Design and ship evaluations that reveal where our models succeed, where they fail, and which improvements matter for customers. • Work directly with customers to understand their evaluation needs and support them as they deploy and scale their robots. • Shape the research direction and technical foundation of a founding team working across world models, robot learning, and production systems. Who You Are You have built and shipped in several of the following areas: • End-to-end training of large generative models over video, image, or 3D data, using diffusion, autoregressive, or flow-matching approaches. • Training learned policy models, including vision-language-action (VLA) models, diffusion policies, world-action models, or end-to-end imitation and reinforcement learning policies for manipulation or driving. • Training large models across multiple GPUs and nodes. • Building and shipping evaluations for large models. • Training and evaluating reward models for robot learning, using human feedback or task outcomes. • Deploying robots outside the lab and into the real world. You are comfortable taking an open research question through experiments, implementation, and rigorous evaluation. You care about whether a reward captures meaningful behavior and whether an evaluation reflects real policy performance. You thrive with a broad mandate, move quickly without being micromanaged, and are excited to have an outsized impact early. About Us Our north star is simple: more robots deployed in the real world. We are assembling a small, exceptional team to solve the unglamorous, difficult, and deeply consequential work required to get there. If you want to help build the bridge between breakthrough models and useful robots, join us. Compensation The base pay range for this role is $225,000 – $300,000 per year.