Founding Research Scientist at stealth physical AI startup

About the role

You will lead the research function at a venture-backed startup solving the data bottleneck in robotics. By training end-to-end manipulation policies and designing evaluation frameworks, you will bridge the gap between high-quality real-world interaction data and model performance. This hands-on role offers significant equity and the chance to build a research stack from zero.

Why this role is remarkable

• Rare opportunity to join as a founding member with minority co-founder level equity and full ownership of the research agenda.
• Work at the critical intersection of robotics and data infrastructure, solving the primary bottleneck preventing the scaling of physical AI.
• Collaborate directly with a technical founding team in a high-autonomy environment where you set the research direction and validation strategy.

What You Will Do

• Train and evaluate real-world robot manipulation policies using imitation learning, diffusion policies, or vision-language-action models.
• Build the end-to-end data-to-policy validation pipeline to ensure high-quality tactile and interaction data translates to performance.
• Work alongside hardware and software teams to define the research roadmap and iterate on differentiated data collection methods.

The ideal candidate

• Proven track record training manipulation policies on physical hardware, focusing on dexterous manipulation or tactile sensing.
• PhD from a top-tier robotics laboratory or extensive research experience at a leading organization specializing in physical foundation models.
• Strong desire to work in an early-stage, hands-on individual contributor role with the ambition to build a research function from scratch.

Location: San Francisco, USA