Robotics

AgileX Dual-Arm Nero integrates with Isaac Lab for RL training

Open-source project enables dual-arm robot learning with NVIDIA's Isaac Lab.

Deep Dive

A developer has open-sourced an Isaac Lab extension that integrates the AgileX Dual-Arm Nero Manipulator, enabling researchers to train dual-arm reinforcement learning policies for reach tasks. The project builds on the existing SO-ARM101 implementation, adding a full URDF model with mesh references, actuator configurations with per-joint stiffness (ranging from 10 to 200) and damping, and a dedicated task environment (`dual_nero_reach_env_cfg.py`) for the RL-Games framework.

Key additions include a `dual_nero.py` robot configuration file defining articulation properties, joint limits, and initial states for both arms and grippers. The repository provides ready-to-run scripts for zero-shot agent testing and RL training, along with a UI extension example. This setup allows developers to simulate and train dual-arm manipulation policies without building the simulation stack from scratch, leveraging Isaac Lab's physics engine and RL capabilities.

Key Points
  • AgileX Dual-Arm Nero manipulator integrated into NVIDIA Isaac Lab via URDF and configuration files
  • Supports RL training with RL-Games framework, including reach task environment and joint position control
  • Open-source repository includes training/play scripts, per-joint stiffness settings (10–200), and zero-agent testing

Why It Matters

Lowers the entry barrier for dual-arm robotic manipulation research using reinforcement learning in simulation.

📬 Get the top 10 AI stories daily