Dexterous Robotics ML Engineer - Real-World Robot Learning

Origami Robotics; San Francisco (CA); On-site; USD 120,000 - 180,000; Full time

Job summary

Origami Robotics in San Francisco is seeking a hands-on researcher to advance general-purpose dexterous manipulation on physical robots. You will train, deploy, and evaluate policies across learning, controls, software, and hardware, turning failures into better data, models, and systems.

You will build real-time control and state estimation, work with PyTorch or JAX, and collaborate across robot learning, perception, and planning to ship robust robotic systems in the real world.

Qualifications

• You have trained and deployed a learned policy on a physical robot.
• Strong Python and hands-on experience building and debugging robotic systems.
• Experience with PyTorch or JAX and methods such as imitation learning or RL.
• Working knowledge of robot kinematics, controls, and motion planning.
• Strong experimental judgment and ownership from idea to real-world demonstration.

Responsibilities

• Train, deploy, and improve dexterous manipulation policies on physical robots.
• Build teleoperation and data-collection workflows for training and evaluation.
• Apply imitation learning and RL fine-tuning where appropriate.
• Build custom real-time control and state estimation algorithms, integrate learned policies with perception and planning, and debug failures across the full robotic system.

Skills

Python

Robot systems debugging

PyTorch

JAX

Imitation learning

Reinforcement learning

Controls

Kinematics

Motion planning

Tools

ROS2

C++

Sim-to-real

Edge inference

Job description

Origami Robotics in San Francisco is seeking a hands-on researcher to advance general-purpose dexterous manipulation on physical robots. You will train, deploy, and evaluate policies across learning, controls, software, and hardware, turning failures into better data, models, and systems.

You will build real-time control and state estimation, work with PyTorch or JAX, and collaborate across robot learning, perception, and planning to ship robust robotic systems in the real world.