Abstract:Traditional navigation algorithms for humanoid robots often fail to account for the effects of periodic center-of-mass variations, torso sway, and lateral gait drift on spatial safety margins. Consequently, planned trajectories may pass too close to obstacles, while local navigation strategies may become rigid or trapped. To address these issues, this paper proposes a hierarchical navigation framework specifically designed for humanoid robots. The framework consists of RP-Theta* (Risk-Perception Theta*) for global path planning, FCA-DWA (Fuzzy Cognitive Arbiter Dynamic Window Approach) for local obstacle avoidance, and a dynamic multi-goal scheduling module. RP-Theta* incorporates continuous spatial risk into the path search process. FCA-DWA adjusts local velocity decisions online according to the current environmental state. The multi-goal scheduling module switches navigation goals based on the actual connectivity cost and triggers replanning when the robot deviates from the planned path. MATLAB simulation results show that RP-Theta* reduces the average path risk from 0.51 to 0.15, while the integrated method shortens the task completion time by 14.5% in the most complex scenario. The proposed framework is further deployed and validated in Gazebo using a Unitree 12-degree-of-freedom humanoid robot model.