Founding Robot Learning Engineer

Train and deploy VLA, imitation learning, and RL policies for contact-rich manipulation.

Develop policies that bridge the sim2real gap, and work out how the simulation has to improve so that policies built in it work in real life.

Help build the data flywheel from simulation: generating data, training policies, testing them on real robots, and feeding what we learn back into the simulation.

Use world models, VLAs, and world action models (WAMs) to build and improve these policies.

Define evaluation metrics that predict a policy's success and its zero-shot sim2real transfer before it reaches the real robot.

Hands-on experience training policies and running them on real robots. If you have trained a policy in simulation, watched it fail on the real robot, and worked out why, you are the right fit for us.

You have high agency and high integrity, and you are willing to learn, apply, and fail quickly.

You take complete ownership with minimal guidance and do everything possible to make the product work.

It does not matter if you do not have the relevant experience in any of the requirements, as long as you have a strong interest to learn and apply quickly.

A degree in CS, robotics, or ML, or equivalent experience shipping learned policies on real hardware.

Strong Python and PyTorch.

Experience with at least one of reinforcement learning, imitation learning, or VLA training.

Experience with Isaac Sim, and working knowledge of ROS2 or other robotics middleware.

Hands-on experience with VLA architectures, such as π0, OpenVLA, or Octo, or fine-tuning foundation models for robotics.

Experience with world model architectures, such as DreamerV3, V-JEPA, or latent dynamics models.

Teleoperation data collection and DAgger-style correction pipelines.

Optimizing policies for deployment on robot hardware, such as with TensorRT or ONNX.