Researcher, World Models

The role

You will design, train, and evaluate world models to help the Asimov humanoid robot perceive, predict, and act. You will work on self-supervised learning stacks, generative architectures, and data pipelines to integrate research with hardware and software.

What you will do

• Design, train, and evaluate world models for visual, proprioceptive, and force/torque modalities.
• Advance the self-supervised learning stack for visual and sensor representations.
• Prototype and benchmark generative and predictive architectures.
• Own the end to end data pipeline for experiments.
• Integrate research with platform, firmware, and software teams.
• Contribute to sim-to-real transfer, inverse dynamics, and multi-modal sensor fusion.
• Publish or open-source work.

Required qualifications

• Proven modeling track record with solid evaluations.
• JEPA fluency.
• Familiarity with VLA models and their trade-offs.
• Strong depth in at least one sensory domain such as vision, audio, or natural language.
• Strong data abilities.
• Solid engineering skills to implement, integrate, and ship builds.
• Conversant or deep in several of: SSL for visual and sensor representations, world models, generative and predictive architectures, robotics ML, sensor fusion, PyTorch, JAX, and distributed training.

Nice to have

• 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.
• Demonstrated hardware or robotics interest or hands-on experience.
• Strong communication through technical blogs, talks, or clear written research.

About the company

Menlo Research is an Applied R&D lab building Asimov, an open-source humanoid robot platform, and the full software stack that powers it. Their mission is to make humanoid labor economically viable.