A Hierarchical, Model‐Based System for High‐Performance Humanoid Soccer

The development of athletic humanoid robots has gained increasing attention as advances in actuation, sensing, and control enable increasingly dynamic and practical capabilities. RoboCup, an international competition for fully autonomous robotic systems, provides a challenging benchmark for embodied intelligence, requiring robots to perceive, decide, and act in uncertain and dynamic scenarios. This paper presents the hardware and software innovations underlying an open‐source humanoid platform validated in RoboCup 2024. The hardware features lightweight structural components, quasi‐direct‐drive actuators with high torque density, and a specialized foot design that supports powerful in‐gait kicks while maintaining locomotion stability. On the software side, we develop an integrated perception and localization framework combining stereo vision, object detection, and landmark‐based fusion for reliable state estimation. A mid‐level navigation stack produces collision‐aware and dynamically feasible trajectories, while a centralized behavior manager coordinates decision making and action execution. The tight integration of these subsystems enables fast, precise, and physically consistent behaviors in environments with frequent contact and dynamic movement. Beyond competition success, the system highlights key principles for embodied intelligence, including joint hardware and software design, perception‐planning‐control integration, and reliable autonomy in complex environments.