ActiveGAMER

Active GAussian Mapping through Efficient Rendering

We introduce ActiveGAMER, an active mapping system that utilizes 3D Gaussian Splatting (3DGS) to achieve high-quality, real-time scene mapping and exploration. Unlike traditional NeRF-based methods, which are computationally demanding and restrict active mapping performance, our approach leverages the efficient rendering capabilities of 3DGS, allowing effective and efficient exploration in complex environments.

The core of our system is a rendering-based information gain module that dynamically identifies the most informative viewpoints for next-best-view planning, enhancing both geometric and photometric reconstruction accuracy. ActiveGAMER also integrates a carefully balanced framework, combining coarse-to-fine exploration, post-refinement, and a global-local keyframe selection strategy to maximize reconstruction completeness and fidelity.

Our system autonomously explores and reconstructs environments with state-of-the-art geometric and photometric accuracy and completeness, significantly surpassing existing approaches in both aspects. Extensive evaluations on benchmark datasets such as Replica and MP3D highlight ActiveGAMER's effectiveness in active mapping tasks.

At each keyframe step, HabitatSim generates posed RGB-D images, which are stored in a keyframe database, with certain frames designated as Global Keyframes. These observations are used to update a Gaussian Map comprising a collection of 3D Gaussians. Map optimization is achieved by minimizing color and depth rendering losses, based on the rendered RGB-D images and silhouette masks. Using the up-to-date Gaussian Map, rendering-based planning evaluates the information gain across sampled candidate viewpoints and choose the one with highest information gain as the next-best-view.

Our method contains two maps: 3DGS Map and Exploration Map. The 3DGS Map captures high-fidelity scene geometry and appearance using 3D Gaussian Splatting, enabling accurate rendering and geometry. The Exploration Map, represented as a 3D occupancy grid, guides active view selection by tracking free, occupied, and unexplored space. These two maps work together—3DGS refines reconstruction quality, while the Exploration Map accelerates coverage and planning—ensuring efficient and complete scene modeling.

Quantitative Results

We evaluate our method on both MP3D and Replica benchmarks, demonstrating strong performance in reconstruction accuracy, completeness, and rendering quality.

As shown in Table 1, our approach achieves the best completeness and compactness ratio on MP3D, significantly outperforming prior methods such as ANN and NARUTO.

Table 2 presents novel view rendering results on Replica, where our method consistently outperforms SplatAM and NARUTO in PSNR, LPIPS, and L1-D metrics, indicating superior visual fidelity.

Table 3 provides ablation studies validating the effectiveness of each component, showing that both refinement and global keyframes contribute to improved completeness and PSNR.

Finally, Figure 5 illustrates reconstruction progress over time, highlighting the impact of fine exploration, which leads to a notable jump in both completeness and rendering quality.

Qualitative Results

We present qualitative results of our method on Replica and MP3D datasets, showcasing the high-quality reconstruction and rendering capabilities of ActiveGAMER.