Building Physical AI systems often means dealing with fragmented tools for data, training, simulation, and hardware control. To solve this, our team developed CYCLO: a modular, open-source framework designed to connect the entire Physical AI workflow into a single stream.
Key Features:
• Cyclo Intelligence: Data recording, LeRobot dataset conversion, and imitation/reinforcement learning policy training & inference
• Cyclo Control: QP-based motion controller with joint limits and self-collision avoidance
• Cyclo Lab: Simulation training environment built on NVIDIA Isaac Lab
• Cyclo Manager: Web browser-based UI for robot operation and management
• Ecosystem Integration: Seamless compatibility with ROS 2, Zenoh, Hugging Face, and more
Whether you are training policies or deploying on real hardware, CYCLO streamlines your entire development process. Check out our repository, overview video, and documentation below to get started:
Video: https://youtu.be/0fw_7_cnlEI
Code: GitHub - ROBOTIS-GIT/cyclo: Cyclo is an open, modular framework for building and operating Physical AI systems across AI integration, data workflows, robot control, simulation, and operations on real hardware. · GitHub
Docs: What is CYCLO? | ROBOTIS Docs