Developed core locomotion enhancements for the Agibot A2 DOF12 robot within the RoboVerseOrg/RoboVerse repository, focusing on reinforcement learning and robotics simulation using Python. The work involved integrating the robot into the RL locomotion pipeline, updating training and evaluation paths, and refining device handling to support faster and safer experimentation. Implemented joint effort limits and new reward shaping strategies to improve stability and performance, while enhancing evaluation scripts with time tracking and compatibility fixes for Mujoco. Adjustments to actuator configuration and USD path support further improved runtime robustness, translating research advances into production-ready improvements for robotic experimentation workflows.