The physical AI framework MXI-SimWorld v0 Preview, developed by the team led by Professor Yang Chao of the School of Mathematical Sciences at Peking University, has claimed first place on the overall leaderboard with a composite score of 0.9656 in the latest weekly rankings released by RoboColiseum, a leading evaluation platform for embodied intelligence. The framework also took first place in two sub-categories—spatial understanding and general manipulation—and second place in the other two—instruction following and perturbation adaptation—finishing among the top two across all four dimensions. The framework is built on the idea of "supporting policy decisions with simulation." Through a complete pipeline spanning scene reconstruction, interaction calibration, and parallel action rehearsal across multiple environments, it leverages the team's self-developed Harmonic Object Simulation Engine to achieve solving speeds far exceeding those of mainstream physics engines. This addresses a key shortcoming of general-purpose large models—their lack of physical interaction cognition, which makes them prone to errors when deployed in robotic scenarios—by enabling robots to rehearse and screen optimal plans in a trustworthy simulation environment before executing real-world actions. The team is building a full-stack technical system integrating "high-performance computing + physical simulation + AI," and plans to release a series of related models in the future.