GS-Planner: A Gaussian-Splatting-based Planning Framework for Active High-Fidelity Reconstruction This paper proposes GS-Planner, a planning framework for active high-fidelity 3D reconstruction using 3D Gaussian Splatting (3DGS). The framework enables robots to autonomously collect scene data for full coverage, ensuring high-fidelity, detailed, and realistic digital representations of scenes. The system uses 3DGS as the scene representation, which offers high visual quality, efficient fusion, and fast rendering. The framework includes a completeness evaluation to identify unobserved regions and a quality evaluation to assess the accuracy of reconstructed regions. A sampling-based active exploration strategy is designed to guide the robot to unobserved areas, improving the geometric and textural quality of the 3DGS map. A quadrotor is selected as the robotic platform due to its high agility. A safety constraint is integrated with the 3DGS map to generate executable trajectories for quadrotor navigation. The system is validated through extensive experiments in highly realistic simulation scenes, demonstrating the effectiveness of the proposed method. The contributions include the first active 3D reconstruction system using 3DGS with online evaluation, evaluation metrics for reconstruction completeness and quality, a safety constraint with 3DGS, and extensive simulation experiments to validate the system. The system is compared with traditional methods, showing its superior efficiency and accuracy in completeness and quality evaluation. The framework is designed to generate safe and high-information-gain viewpoints for the robot, enabling full reconstruction of scenes with high fidelity. The trajectory optimization framework ensures collision-free and dynamic-feasible trajectories for quadrotors. Future work includes deploying the system on real robotic platforms and improving the efficiency of 3DGS.GS-Planner: A Gaussian-Splatting-based Planning Framework for Active High-Fidelity Reconstruction This paper proposes GS-Planner, a planning framework for active high-fidelity 3D reconstruction using 3D Gaussian Splatting (3DGS). The framework enables robots to autonomously collect scene data for full coverage, ensuring high-fidelity, detailed, and realistic digital representations of scenes. The system uses 3DGS as the scene representation, which offers high visual quality, efficient fusion, and fast rendering. The framework includes a completeness evaluation to identify unobserved regions and a quality evaluation to assess the accuracy of reconstructed regions. A sampling-based active exploration strategy is designed to guide the robot to unobserved areas, improving the geometric and textural quality of the 3DGS map. A quadrotor is selected as the robotic platform due to its high agility. A safety constraint is integrated with the 3DGS map to generate executable trajectories for quadrotor navigation. The system is validated through extensive experiments in highly realistic simulation scenes, demonstrating the effectiveness of the proposed method. The contributions include the first active 3D reconstruction system using 3DGS with online evaluation, evaluation metrics for reconstruction completeness and quality, a safety constraint with 3DGS, and extensive simulation experiments to validate the system. The system is compared with traditional methods, showing its superior efficiency and accuracy in completeness and quality evaluation. The framework is designed to generate safe and high-information-gain viewpoints for the robot, enabling full reconstruction of scenes with high fidelity. The trajectory optimization framework ensures collision-free and dynamic-feasible trajectories for quadrotors. Future work includes deploying the system on real robotic platforms and improving the efficiency of 3DGS.