We're looking for a Research Scientist or Research Engineer to own the strategy and systems for collecting, curating, and scaling high-quality robot learning data. This role sits at the intersection of robotics, data collection, and research - your work directly determines the diversity and quality of the demonstrations our models train on. What You'll Do • Design and implement teleoperation and demonstration collection systems for high-quality robot learning data • Develop data quality metrics, curation pipelines, and filtering strategies specific to robotic interaction data • Research methods to augment real robot data with synthetic, simulated, or cross-embodiment sources • Identify and source external robotic datasets to expand training diversity across platforms and tasks • Build tooling for researchers to explore, annotate, and iterate on robotic datasets • Collaborate with pre-training and post-training teams to translate model data needs into concrete collection strategies • Measure the downstream impact of data collection decisions on model and policy performance What We're Looking For • Hands-on experience with robotic data collection, teleoperation systems, or demonstration frameworks • Understanding of what makes robot learning data useful: diversity, coverage, temporal quality, and action fidelity • Strong software engineering skills for building reliable data collection and processing systems • Ability to reason across hardware, pipelines, and model performance • Experience working with real robotic hardware in a research or industrial setting Nice to Have (But Not Required) • Experience with sim-to-real transfer and synthetic data generation for robotics • Familiarity with cross-embodiment datasets (e.g., Open X-Embodiment, DROID) • Experience with VR teleoperation, motion capture, or dexterous demonstration collection • Understanding of imitation learning and how data properties affect policy generalization • PhD or strong research background in robotics or ML Why This Role • The data you collect and curate is the direct upstream dependency for all model quality • Unique leverage: improvements to data quality compound across every training run • Work across hardware, systems, and research in a way few roles allow • Direct feedback loop with both robot operators and research scientists to continuously improve data quality