What Ronin actually is (and who it is for)

Ronin is a humanoid robot that sits on your desk. It has two arms, two hands, and a head with stereo cameras. You teach it what to do by showing it, using a matched controller that mirrors its body. No coding required.

It costs $4,999. That number is not a typo.

What "teach by demonstration" actually means

You pick up the teaching controller. It has the same joint layout as Ronin, the same degrees of freedom, the same proportions. When you move the controller, Ronin moves the same way.

You perform the task you want Ronin to learn. Fold a towel. Sort parts into bins. Place components into a tray. Ronin records everything: every joint angle, every camera frame, every timestep.

After enough demonstrations, typically 20 to 50, you train a policy. The robot learns to do the task on its own. This is not remote control. Ronin is not replaying a recording. It is learning a generalized behavior from your examples.

The more demonstrations you give, the better it gets. The more edge cases you show it, the more robust it becomes. This is the core loop of imitation learning, and it is how most of the interesting work in embodied AI happens today.

Who is building with Ronin

Researchers who need an affordable bimanual platform for imitation learning experiments. The closest comparable platforms cost $50,000 to $200,000. Most labs cannot justify that for exploratory work.

Small hardware teams who want to prototype robot-assisted workflows without committing to a six-figure platform before they know if the concept works.

Educators teaching the next generation of roboticists. You can teach imitation learning from a textbook, but the understanding changes completely when a student can collect data, train a policy, and watch it succeed or fail in front of them.

Builders and developers who want a serious humanoid platform they can actually own and iterate on.

The common thread: people who want to do, not just watch.

What is inside

Two 7-degree-of-freedom arms. That is the same range of motion as a human arm from wrist to shoulder. Matched pair, tuned for fine two-handed work.

Stereo vision from cameras in the head. An active 3-DoF neck for directing gaze. Ronin can track what you are doing and where you are in the space.

Standard grippers included. A 16-DoF dexterous hand called RoninHand is in development as a future upgrade.

No onboard computer. You connect your own laptop or desktop. This keeps the cost down and lets you use whatever GPU you already have. Most researchers already have a capable machine. There is no reason to build another one into the robot and charge you for it.

Open SDK. LeRobot and OpenPI compatible. Everything is documented. You can go as deep as you want.

What Ronin is not

It is not a walking humanoid. It sits on a desk or mounts to a surface. Legs are a distraction from the manipulation problem, which is where the real research bottleneck lives.

It is not a toy. The arms are rated for real tasks with real objects at real-world speeds.

It is not locked to one software ecosystem. It works with LeRobot, HuggingFace datasets, and standard robotics tools. If you already have a workflow, Ronin fits into it.

It is not vaporware. Units ship Q4 2026 through Q1 2027. Reserve with a fully refundable $200 deposit.

Why $4,999

The closest comparable bimanual platforms cost $50,000 to $200,000 or more. Most of that cost is not better components. It is low-volume manufacturing, proprietary software stacks, and margins that assume you are a funded lab.

Polymorph designed Ronin for volume from day one. Standard actuators. Open software. No licensing fees. The goal was always to make the price low enough that the purchase decision is about whether you need a bimanual humanoid, not whether you can afford one.

$4,999 is not cheap. But it is the difference between "maybe someday" and "order it on Tuesday."

The real point

The best way to understand what a robot can do is to have one. Not to read about one, not to watch a demo video, not to wait for a paper. Ronin exists so that the number of people who can answer that question goes from hundreds to thousands.