Robot Learning Engineer - Manipulation

Job description

Job responsibilities

• Work on the full learning loop for manipulation tasks: task definition, demonstration collection, data curation, training, real-robot evaluation, and deployment.
• Train and fine-tune manipulation policies, from large pretrained models such as vision-language-action models to compact task-specific policies, and choose the right approach for each task.
• Develop repeatable recipes for industrial tasks such as pick-and-place, bimanual handling, and contact-rich assembly, adding force or tactile signals where they help.
• Deploy policies on edge compute and validate observation processing, action interfaces, and control timing on the robot.
• Turn failures and human interventions into better data, better models, and better evaluation.
• Measure what customers care about, including success rate, cycle time, intervention rate, and the data and time needed to reach a target, and improve those numbers task after task.
• Package recipes and models so the next task, and the next robot, starts from what was learned on the last one.

Job requirements

• Trained or fine-tuned a learned manipulation policy and deployed and evaluated it on a physical robot.
• Strong Python and PyTorch skills, with the ability to write maintainable training, evaluation, and deployment code.
• Practical depth in imitation learning and at least one modern policy family, such as vision-language-action models, diffusion policies, or action-chunking transformers.
• Working knowledge of robot kinematics, coordinate frames, camera calibration, and the interface between learned actions and low-level control.
• The habit of diagnosing failures with controlled experiments across data, sensing, model, and execution.
• Comfort taking on open-ended problems and communicating tradeoffs clearly to teammates at the robot.