NVIDIA Isaac Lab tutorial enables robotic arm data collection for imitation learning
Learn to teleoperate AgileX NERO arm and collect HDF5 demonstrations for AI training.
NVIDIA Isaac Lab now offers a step-by-step tutorial for keyboard teleoperation and demonstration data collection on the AgileX NERO 7-DoF robotic arm. The guide covers multimodal teleoperation for end-effector SE(3) pose alignment, enabling users to control the arm in simulation for a block-stacking (Pick and Place) task. It includes seamless HDF5 data logging compatible with Robomimic, plus trajectory replay for instant validation of collected human demonstrations. Both the AgileX NERO (7-DoF) and PiPER (6-DoF) arms are supported out of the box.
The tutorial begins by merging separate XACRO description files into a unified URDF for Omniverse import. It then walks through environment configuration, keyboard teleoperation, and data collection. The collected HDF5 datasets are ready for imitation learning pipelines, allowing researchers and hobbyists to train AI agents on human demonstrations. The project is open-source and available on GitHub, making it accessible for robotics labs and educators looking to accelerate robot learning.
- Keyboard teleoperation enables SE(3) pose control for the AgileX NERO 7-DoF arm in NVIDIA Isaac Lab.
- Demonstration data is logged in HDF5 format, compatible with the Robomimic framework for imitation learning.
- Supports both AgileX NERO (7-DoF) and PiPER (6-DoF) arms with cross-hardware compatibility.
Why It Matters
Accelerates robot learning by providing an easy pipeline to collect and replay human demonstrations in simulation.