If you have ever implemented dual-arm robot control, you know the pain: even a tiny sync error can crush your target object or send it flying.
To solve this, our development team built “cyclo_control”, a QP-based (Quadratic Programming) motion control package for the ROBOTIS AI Worker. And yes, in classic ROBOTIS fashion, we have released it completely open-source!
The core is our “Bimanual Controller” feature. It constrains the relative pose of the two grippers as if they are locked by a rigid virtual link. Instead of manually calculating trajectories for both arms, you only need to specify the target pose of the virtual object, and the controller computes the coordinated trajectories in real-time.
We put this framework to the test under three challenging, real-world scenarios:
- Safely transporting a highly fragile foam board
- Applying tape to a box while keeping the tape’s tension tight
- Lifting a box using only wrist-squeezing force
Even with unsteady teleoperation commands, the controller instantly adapts, avoiding self-collisions and joint limits to protect the object perfectly.
Whether you are working on dual-arm manipulation or Physical AI research, we hope this makes your development much easier.
Video: https://youtu.be/4rwVRsLNf9c
Code: GitHub - ROBOTIS-GIT/cyclo_control: Control packages for the ROBOTIS Physical AI lineup · GitHub
Docs: Cyclo Control | ROBOTIS Docs