Humanoid-Gym is an open-source reinforcement learning (RL) framework designed to train locomotion skills for humanoid robots, emphasizing zero-shot transfer from simulation to real-world environments. The framework is built on Nvidia Isaac Gym and integrates a sim-to-sim validation tool using MuJoCo to ensure the robustness and generalization of trained policies. It features specialized rewards and domain randomization techniques to simplify the sim-to-real transition. The framework has been successfully tested on two humanoid robots, XBot-S (1.2 meters tall) and XBot-L (1.65 meters tall), demonstrating its effectiveness in zero-shot transfer. The paper details the system design, reward function, and experimental results, highlighting the alignment of MuJoCo dynamics with real-world performance. The framework aims to bridge the sim-to-real gap and enhance the potential for successful RL applications in humanoid robotics.Humanoid-Gym is an open-source reinforcement learning (RL) framework designed to train locomotion skills for humanoid robots, emphasizing zero-shot transfer from simulation to real-world environments. The framework is built on Nvidia Isaac Gym and integrates a sim-to-sim validation tool using MuJoCo to ensure the robustness and generalization of trained policies. It features specialized rewards and domain randomization techniques to simplify the sim-to-real transition. The framework has been successfully tested on two humanoid robots, XBot-S (1.2 meters tall) and XBot-L (1.65 meters tall), demonstrating its effectiveness in zero-shot transfer. The paper details the system design, reward function, and experimental results, highlighting the alignment of MuJoCo dynamics with real-world performance. The framework aims to bridge the sim-to-real gap and enhance the potential for successful RL applications in humanoid robotics.