Skip to main content
DimOS manipulation gets a robot arm from “plugged in” to “doing useful work”. You can plan collision-free motion, teleoperate with a VR headset or keyboard, record demonstrations, and train a policy that runs back on the same robot.

​
Choose your workflow

​
How the stack fits together

Three layers, top to bottom:
The planning layer is where you should spend your time. It is where “do the task” turns into motion — and it is deliberately pluggable. The planner, the IK solver, and the world model are all protocol-based: you can swap any of them, or bypass planning entirely and drive the coordinator from a learned policy. The layers below rarely need touching once your arm is integrated. Teleop slots in beside planning, not above it: a Quest headset or keyboard streams end-effector targets to the coordinator, which solves IK in-loop. That path is what makes demonstration collection feel direct.

​
What runs today

Several workflows also run without hardware through a mock arm or MuJoCo, including the quick start below. A1Z hand-teach and learned-policy execution require the real arm. Per-arm setup lives with the platform, not here: A-750, OpenArm, and the Galaxea A1Z hardware and learning guide.

​
Try it in two minutes

A Meshcat window opens with the arm. Drive the end-effector with W/A/S/D (XY), Q/E (Z), R/F/T/G/Y/H (roll/pitch/yaw). Everything you see — the IK, the 100Hz loop, the visualization — is the same stack that runs on real hardware; only the adapter is fake. Then pick your workflow from the table above.