This is the first robot learning hire at a company that builds physical AI data and infrastructure. You will set up policy training and evaluation in-house.
You will work with engineering and product teams to show how synthetic training data changes robot performance, and turn those findings into requirements for the product and the data.
Responsibilities
Fine-tune robot policies on synthetic data, using imitation learning, vision-language-action models, diffusion policies, or reinforcement learning in simulation.
Set baselines on public evaluation suites, and compare runs using fixed seeds, checkpoints, and protocols.
Test manipulation and grasping tasks, and report both gains and failures carefully.
Convert experiment results into data specifications that cover what to generate, how faithful it must be, and how much it should vary.
Publish papers and talks for the robotics community, and join customer technical discussions.
Help recruit and develop the robotics team. You are an individual contributor who also owns part of the team building.
Requirements
6 years of hands-on robot learning work, in industry or university labs.
A master's or doctoral degree in robotics, machine learning, or a closely related field.
Experience training and testing robot policies in simulation with Isaac Sim, Isaac Lab, MuJoCo, or similar tools.
Experience with policy learning methods such as VLA fine-tuning, diffusion policy, ACT, imitation learning, behavior cloning, or reinforcement learning.
Strong Python and PyTorch skills. Familiarity with NVIDIA Isaac, USD, and sim-to-real transfer is expected.
Experience with manipulation or grasping policies. Navigation policy experience also counts.
Background in robotics or simulation products, and a record of using evaluation results to change data or simulation across training cycles.
You can work without visa sponsorship.
Conditions
Base salary ranges from $260,000 to $300,000 per year. No visa sponsorship is offered.
Hybrid role based in San Francisco, United States.