Advancing the frontier of generally capable robotics

Tutor is a frontier research lab that trains foundation models for intelligent dexterous robot manipulation. At Tutor, we believe robotics will only be solved through the unification of research and deployment.

Read the DF1 technical report

Open positions

Our approach

Science done purely in the lab or custom-built industrial automation can only advance the frontier so far. We must both push the frontier of robot technology to enable real world deployed value, and push the frontier of robot deployment to create the economic and data flywheels that enable the right technology development.

Learned robot manipulation is a function of data in

We believe learned robot manipulation is in many parts a function of data in. There is no internet of training data for the physical world, and the greatest capabilities in modern language models have come from doing real work, not reading about it.

Robotics is a no-free-lunch full-stack problem

We believe robotics will be solved through careful intentional tradeoff taking across data, models, software, hardware, and product. It is through these tradeoffs that we iteratively advance the frontier.

The deployment flywheel

We deploy fleets of robots into real-world, economically valuable applications, supervised and taught by Tutors. As deployments scale, our Tutors impart more demonstrations and corrections to more robots, and the resulting data trains policies that make every robot in the fleet more capable over time.

Data Factory 1 (DF1)

DF1 is a 100-robot farm of Sonny robots, to our knowledge the largest robot data factory in the United States. We use DF1 for large-scale real-world teleoperation, evaluation, and online iterative improvement of robot foundation models for dextrous bimanual manipulation.

We are using DF1 to explore a new class of methods where policy improvement is driven autonomously via large-scale human supervision over deployed robot fleets.

Read the DF1 technical report

Learn more about Data Factory 1

Learning from Tutor demonstration

Our Tutors teleoperate robots through new tasks to demonstrate robot behaviors for initial behavior cloning.

Corrections from robots' own mistakes

Teleoperators not only perform initial task demonstrations but also correct mistakes and intervene when robots encounter errors, providing valuable data for model training.

Reward from Tutor feedback

Episodes of robot rollouts are scored by our Tutors to provide positive or negative feedback on the robot's behavior, providing signal for the post-training process to accentuate favorable robot behaviors and down-weight low-quality behaviors.

We build policies that learn and improve through scaled operational processes, providing a path towards autonomously compounding model improvement.

Research across the stack

Tutor engages in research work alongside our engineering teams to conceptualize new robotic systems. Our work is full-stack, ranging across frontier model training, optimization and planning algorithms, and novel hardware.

Join the mission

Solving the hardest problems in robotics requires passionate, creative, and dedicated minds. If that sounds like you, get in touch.

Open positions