AgileX and NVIDIA's Isaac Lab open-source RL arm grasping system

AgileX Robotics has published a complete reinforcement learning workflow for embodied AI manipulation using its NERO robotic arm (SO-ARM101) and NVIDIA Isaac Lab. The open-source project, available on GitHub under the AgileX-College repository, provides a simulation-driven framework for training robotic grasping policies. It leverages Python 3.8+ with the uv package manager for fast dependency resolution, and includes pre-built RL task definitions such as Isaac-Nero-Reach-v0 and Isaac-Piper-Reach-v0. The workflow is designed with simulation-to-real transfer in mind, preparing policies for deployment on physical NERO arms.

To get started, users install Isaac Lab via pip, clone the repository, and run a zero-action agent to validate the environment. The project also includes a URDF model download from the agx_arm_urdf repository and a clear directory structure for extending tasks. Key specifications include the NERO robotic arm hardware (from AgileX), SO-ARM101 base framework, and NVIDIA Isaac Lab simulation platform. The system is licensed under BSD-3-Clause and includes citation and contribution guidelines for academic and industrial use.