Southwest Research Institute shares lessons learned from converting a Robotiq 2F-85 gripper from URDF to a simulation-ready USD asset in NVIDIA Isaac Sim. The key finding: format conversion alone does not produce high-fidelity simulation assets. The team encountered issues with articulation root structure, non-instanceable assets, and mimic joint contact failures. The best community asset found (UW-Lab's calibrated USD) still failed to reproduce real gripper pad angling behavior during asymmetric grasps — a gap that matters for reinforcement learning and pre-deployment validation. Practical takeaways include validating assets against contact edge cases, preserving the real gripper's command abstraction, tracking asset provenance, and treating URDF-to-USD conversion as the start rather than the end of an authoring workflow. The post calls for manufacturers, ROS tooling, and simulation platforms to align around a testable 'SimReady' standard.
