Standardized Isaac Benchmark Tool for Embodied AI Research

We designed a benchmarking framework for humanoid robots using NVIDIA Omniverse's Isaac Sim platform, and enabled simulation of different humanoid robots performing the same actions for standardized comparison. We also collected joint data during simulated actions and compared it with physical robot performance on identical tasks, evaluating how motions generated by large language models translate to both simulated and physical robots. We designed predefined tasks such as waving, saluting, raising arms, and walking to assess motor performance, and compared performance differences among various humanoid robots in simulation and real-world environments.