Menlo - Researcher, World Models
Requirements
• Publications at NeurIPS, ICML, ICLR, CoRL, or RSS (or arXiv work with comparable citations). • PhD or equivalent research experience in ML, robotics, or computer vision. Not required with a strong portfolio. • Demonstrated hardware or robotics interest or hands-on experience. • Strong communication: technical blogs, talks, or clear written research.
Responsibilities
• Design, train, and rigorously evaluate world models that let Asimov predict the consequences of actions across visual, proprioceptive, and force/torque modalities. • Advance our self-supervised learning stack for visual and sensor representations, building on and extending the JEPA family (V-JEPA, I-JEPA, and related predictive-embedding approaches). • Prototype and benchmark generative and predictive architectures (diffusion, DiT, flow matching, VAEs) against JEPA-style objectives for embodied prediction and planning. • Own the data pipeline for your experiments end to end: curation, tooling, and scaling, without depending on a separate data-engineering team to move. • Integrate what you build with our platform, firmware, and software teams so research reaches the robot, not just the paper. • Contribute to sim-to-real transfer, inverse dynamics, and multi-modal sensor fusion, and publish or open-source work where it strengthens the field and the team. • What We Look For • What We Look For • Proven modeling track record: you have trained models and can show solid, honest evaluations, not just training curves. • JEPA fluency: you understand the joint-embedding predictive approach and can reason about where it fits versus alternatives. • Breadth across approaches: familiarity with prior and adjacent work, including VLA (vision-language-action) models, and a view on their trade-offs. • Depth in a modality: strong depth in at least one sensory domain (vision, audio, natural language, or similar). • Strong data abilities: you get things done without depending on a whole data-engineering team. • Solid engineering: you can implement, integrate, and ship what you build alongside platform, firmware, and software teams. • Conversant, ideally deep, in several of: SSL for visual and sensor representations; world models (JEPA, V-JEPA, I-JEPA, LeJEPA, MJEPA); generative and predictive architectures (diffusion, DiT, flow matching, VAEs); robotics ML (VLA, inverse dynamics, sim-to-real, optical flow); sensor fusion (vision, proprioception, force/torque, multi-modal encoders); PyTorch, JAX, and distributed training.
Benefits
• World models are the bet that lets a humanoid generalize instead of memorize. This is a rare seat where your research runs on real hardware in short cycles, your data and modeling choices are yours to own, and your work ships in the open. If you want the distance between an idea and a walking robot to be measured in weeks, this is the room. • A Note on AI • You don't need deep AI expertise for every role, but we do expect everyone at Menlo to be intellectually curious, drawn to tinkering and discovery, and excited to use AI as a real collaborator in their work. For some roles, AI fluency is a core requirement. When that's the case, we'll say so explicitly in the qualifications. People who thrive here don't treat AI as a novelty. They use it to think better, and make their work easier for others to build on. • Equal Opportunity and Accommodations
Apply in one click
Upload My Resume
Drop here or click to browse · Tap to choose · PDF, DOCX, DOC, RTF, TXT