3D Gaussian Splatting in Robotics: A Survey(1)

Abstract

Rich 3D representations within robotics research have long served as a persistent objective.

While previous NeRF representations have become the predominant approach for their underlying, coordinate-based models, the recent introduction of 3DGS has established remarkable potential in its explicit radiance field representation.

By employing 3D Gaussian primitives to enable explicit representation of the scene and facilitating differentiable rendering, 3DGS has demonstrated distinctive advantages over other radiance fields in real-time rendering and photorealistic performance, offering significant benefits for robotic applications.

Figure

Figure 1

A taxonomy of 3DGS in Robotics.

Figure 2

An illustration of the forward process of 3DGS.

Figure 3

Chronological: 3DGS for Scene Reconstruction.

Figure 4

Categorization: 3DGS for Static Reconstruction.

Figure 5

An illustration of 3DGS for dynamic reconstruction.

Figure 6

Categorization: 3DGS for Dynamic Reconstruction.

Figure 7

Chronological: 3DGS for Scene Segmentation and Editing.

Figure 8

Categorization: 3DGS for Scene Segmentation.

Figure 9

An illustration of 3DGS for scene editing.