Abstract: To improve the path quality and planning efficiency for mobile robots in complex indoor environments, this paper proposes a sampling-guided improved Rapidly-exploring Random Tree Star algorithm, named Sampling-Guided Rapidly-exploring Random Tree Star (SG-RRT*). The algorithm performs coordinated optimization from three aspects: sampling distribution, node connection, and convergence control. In the search stage, a state-adaptive hybrid sampling strategy is designed and combined with an Artificial Potential Field (APF)-guided node refinement mechanism to reduce the blindness of random sampling and improve the success rate of node connection in narrow passages. In the convergence stage, an adaptive termination criterion based on Zero-Lag Marginal Benefit (ZLMB) is constructed to dynamically balance path quality and planning efficiency. Simulation results show that SG-RRT* exhibits good search stability and robustness in various complex environments. Compared with the Informed-RRT* algorithm, the proposed algorithm reduces the planning time by 85.7% while increasing the path length by only 1.3%; compared with the A* algorithm, it can effectively reduce the computational burden in large-scale environments while maintaining favorable path quality. The results show that SG-RRT* can improve planning efficiency while ensuring path quality, and is suitable for real-time global path planning of mobile robots in complex constrained environments.