Research Scientist - Foundation World Models for Robotics

What You\'ll DoDrive research on foundational models and world models for robotics (representation learning, dynamics/prediction, planning, control)Formulate research problems and hypotheses grounded in real robotic autonomy needsDesign and run rigorous experiments at scale, including ablations, benchmarking, and evaluation methodologyDevelop and evaluate model architectures for long-horizon prediction, rollout quality, and downstream robotic task performanceExplore and advance pre-training and post-training (fine-tuning, alignment, evaluation) of large multimodal modelsCollaborate closely with Research Engineers to translate new ideas into scalable training pipelines and reliable systemsCommunicate results clearly through internal writeups, talks, and research reviewsPublish and present work at top-tier venuesWhat We\'re Looking For (Required)PhD in a relevant field (e.g., ML, Robotics, Computer Science, Electrical Engineering, Applied Math, Computer Vision, or closely related)Strong publication record demonstrating high-quality research output (e.g., NeurIPS, ICML, ICLR, CoRL, RSS, ICRA, CVPR, etc.)Deep understanding of modern machine learning, with relevance to at least several of:Deep learning and representation learningSequence modeling / transformersGenerative modeling (e.g., diffusion, autoregressive, latent-variable models)Model-based learning, planning, and/or controlRL / imitation learning for roboticsStrong research taste and independence: ability to define problems, execute, interpret results, and iterate quicklyProficiency with at least one modern ML stack (e.g., PyTorch or JAX) and the ability to implement research ideas in codeClear written and verbal communication skillsComfort operating in ambiguity in a fast-moving startup environmentNice to Have (But Not Required)Prior work specifically on world models (latent dynamics, predictive models, model-based RL/planning, long-horizon rollouts)Experience with large-scale multimodal training (VLMs, video models, action-conditioned models, large policy models)Experience working with robotic learning data (real-world logs, teleop, simulation-to-real, multimodal sensor streams)Hands-on experience deploying learning-based components on real robotsFamiliarity with distributed training and performance debugging (multi-GPU / multi-node)Why This RoleWork with an elite research team from Stanford, Berkeley, Harvard, etc.Research that directly connects to real-world robotic autonomy — not toy benchmarksTight collaboration between research and engineering (no silos)High ownership and ability to shape the research agendaOpportunity to publish meaningful work while seeing it come alive on real robotic systemsJ-18808-Ljbffr