Researcher, Robot Learning

Asimov’s promise is a humanoid that learns: one that discovers new skills and gets better the more it works. You’ll build the learning systems that make that true.

As a Researcher in Robot Learning on our Brain team, you’ll own the methods that turn data into general-purpose behavior on Asimov. Our locomotion and manipulation researchers push specific skills; you’ll work on the layer underneath and across them. That means generalist policies, the data that trains them, and the loops that let Asimov improve from its own experience. You’ll train on real robot data, ship to real hardware, and open-source what you can.

What You’ll Do

• Own Asimov’s generalist policy research end to end, taking each idea from a hypothesis and a dataset to a trained policy running on the physical robot
• Train and post-train vision-language-action models and other generalist policies that follow instructions and transfer across tasks and environments
• Build imitation learning pipelines on teleoperation and demonstration data, and figure out which data actually moves performance
• Design the self-improvement loop, covering autonomous data collection, success detection, and learning from deployment, so Asimov gets better the more it runs
• Combine learning from demonstrations, reinforcement learning, and simulation to close the sim-to-real gap at scale
• Define the evaluations that tell us whether a policy is genuinely better on real hardware, not just on a benchmark
• Partner with World Models, Locomotion and Manipulation, and Platform so your research lands in the stack that ships, and agents calling Asimov through our SDK have capable skills underneath them
• Publish, open-source, and share what you learn with the community building on Asimov

• You have deep hands-on experience training learning-based robot policies, through imitation learning, reinforcement learning, or both
• You’ve taken a learned policy from research code to a physical robot and dealt with everything that broke along the way
• You’re fluent in Python and PyTorch, and comfortable training at scale on multi-GPU clusters
• You think in data: you know how dataset composition, quality, and diversity shape what a policy can and can’t do
• You have a PhD or equivalent research experience in robot learning, ML, or a related field, or a portfolio that speaks for itself
• You communicate clearly through papers, blogs, talks, or well-documented code
• You use AI as a real collaborator in your research

Nice to Have

• Experience with VLA models or robot foundation models such as OpenVLA, Octo, or π0
• Work with diffusion policies, flow matching, or action chunking
• Experience with autonomous data collection, reward learning, or continual learning on real robots
• Time on humanoids or other high-DoF, whole-body platforms
• Familiarity with large-scale simulation in Isaac Lab, MuJoCo, or mjlab
• Published work at CoRL, RSS, ICRA, NeurIPS, or similar venues
• Contributions to open-source robot learning libraries such as LeRobot