We want to share how we efficiently handle complex long-horizon whole-body tasks by combining Behavior Trees (BT) with a Vision-Language-Action (VLA) framework on our AI Worker.
While learned VLA policies using GR00T N1.7 provide incredible flexibility in perception and language understanding, real-world robot operations require robust safety checks, exception handling, and deterministic rule-based control. To bring these two worlds together, we built a built-in behavior tree system within our ‘Cyclo Intelligence’ orchestrator.
Using our browser-based ‘BT Manager’, you can easily drag and drop actions to model complex sequences, including joint control, mobile control, and managing VLA inference lifecycles. You can also monitor and debug the execution flow in real-time.
We have prepared a demo video of the robot autonomously gathering cups and emptying a trash bin, along with our open-source repository and manual. Check out the links below to see the implementation details!
Demo Video: https://youtu.be/jRcUuwxFk_Y
Source Code: GitHub - ROBOTIS-GIT/cyclo_intelligence: cyclo_intelligence · GitHub
Behavior Trees Manual & Quick Guide: Behavior Trees | ROBOTIS Docs