ActiveSplat is an autonomous tool leveraging Gaussian Splatting to create realistic 3D representations by walking through indoor environments, eliminating the need for pre-captured images.
Developed by researchers from Beijing Institute of Technology and Tsinghua University, ActiveSplat can 'build a 3D map on the fly' with exemplary efficiency and accuracy.
Using a hybrid map representation, ActiveSplat combines dense environmental information and sparse topology for effective viewpoint sampling and path planning.
The system's hierarchical planning strategy minimizes repetitive movements while its adaptability to various conditions ensures effective mapping even in challenging areas.
Collection
[
|
...
]