The paper introduces GS-Planner, a planning framework for active high-fidelity reconstruction using 3D Gaussian Splatting (3DGS). The framework aims to enable robots to autonomously collect scene data for full coverage, improving the quality and completeness of the reconstructed scene. Key contributions include: 1. **Online Evaluation**: The framework evaluates the reconstruction quality and completeness of 3DGS maps in real-time to guide the robot's reconstruction process. 2. **Active View Planning**: A sampling-based strategy is designed to explore unobserved areas and improve the geometric and textural quality of the 3DGS map. 3. **Trajectory Optimization**: A safety constraint is integrated with 3DGS to generate executable trajectories for quadrotor navigation in the 3DGS map, ensuring collision-free and dynamic-feasible movements. The paper also discusses the advantages of 3DGS over traditional methods, such as grid maps and neural radiance fields, highlighting its high visual quality, efficient fusion, and fast rendering capabilities. Extensive experiments and ablation studies are conducted in realistic simulation scenes to validate the effectiveness of the proposed method.The paper introduces GS-Planner, a planning framework for active high-fidelity reconstruction using 3D Gaussian Splatting (3DGS). The framework aims to enable robots to autonomously collect scene data for full coverage, improving the quality and completeness of the reconstructed scene. Key contributions include: 1. **Online Evaluation**: The framework evaluates the reconstruction quality and completeness of 3DGS maps in real-time to guide the robot's reconstruction process. 2. **Active View Planning**: A sampling-based strategy is designed to explore unobserved areas and improve the geometric and textural quality of the 3DGS map. 3. **Trajectory Optimization**: A safety constraint is integrated with 3DGS to generate executable trajectories for quadrotor navigation in the 3DGS map, ensuring collision-free and dynamic-feasible movements. The paper also discusses the advantages of 3DGS over traditional methods, such as grid maps and neural radiance fields, highlighting its high visual quality, efficient fusion, and fast rendering capabilities. Extensive experiments and ablation studies are conducted in realistic simulation scenes to validate the effectiveness of the proposed method.