Abstract
We introduce ActiveGAMER, an active mapping system that utilizes 3D GaussianSplatting (3DGS) to achieve high-quality, real-time scene mapping andexploration. Unlike traditional NeRF-based methods, which are computationallydemanding and restrict active mapping performance, our approach leverages theefficient rendering capabilities of 3DGS, allowing effective and efficientexploration in complex environments. The core of our system is arendering-based information gain module that dynamically identifies the mostinformative viewpoints for next-best-view planning, enhancing both geometricand photometric reconstruction accuracy. ActiveGAMER also integrates acarefully balanced framework, combining coarse-to-fine exploration,post-refinement, and a global-local keyframe selection strategy to maximizereconstruction completeness and fidelity. Our system autonomously explores andreconstructs environments with state-of-the-art geometric and photometricaccuracy and completeness, significantly surpassing existing approaches in bothaspects. Extensive evaluations on benchmark datasets such as Replica and MP3Dhighlight ActiveGAMER's effectiveness in active mapping tasks.