cs.ROAug 21, 2026

In-Situ Reconstruction of the International Space Station Using 3D Gaussian Splatting and Astrobee

Authors: Hudson Kim, Ryan Soussan, Brian Coltin, Jordan Kam

Organizations: Department of Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, CA 94720, USA. · Intelligent Robotics Group (KBR Inc.), NASA Ames Research Center, Moffett Field, CA 94035, USA. · Graduate Aerospace Laboratories, California Institute of Technology, Pasadena, CA, 91125, USA.

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

This article presents a novel 3D reconstruction and mapping of the interior of the International Space Station (ISS) using 3D Gaussian Splatting (3DGS). Using existing grayscale images from the Astrobee free-flying robot dataset, we construct a full 3D splat of the ISS' Kibō or Japanese Experiment Module (JEM). 3DGS has in recent years shown promise in providing novel view synthesis of scenes captured from many images or videos, this article applies this approach to human spaceflight systems. We compare our 3DGS architecture to existing methods such as Nerfacto and TensoRF and show that reconstruction improves the state-of-the-art in both scene quality and rendering speed. We show that with as little as 500 in-situ images, a high-fidelity map can be constructed using Astrobee's Navigation Camera (NavCam) during free-flight in the JEM. These reconstructions could enable free-flyers to rapidly create and update interior maps for intra-vehicular habitats like the ISS.

Figures & tables

Fig. 1: Astrobee inside the ISS JEM Module with grayscale images and depth renders from in-situ NavCam images [ 5 , 4 ] .
Fig. 2: Pipeline Diagram. Five Astrobee survey flights are filtered to 5,000 frames. COLMAP takes these images and identifies features. GLOMAP uses these features to simultaneously recover cameras poses and a 3D point cloud. The SfM data is then used as priors for the 3DGS map to be created while adaptive density control ensures that the correct number of Gaussians are used and a learned appearance embedding accounts for lighting differences.
Fig. 3: Four novel views of the ISS Module corridor, Airlock hatch, Crew workstation, and Cargo stowage and their depth map from our 3DGS model.
% of imgsSfMTrainTotalPSNR ↑\uparrowSSIM ↑\uparrowLPIPS ↓\downarrow
100% (4,375)5.0 h47 m5.8 h31.190.9180.208
50% (2,189)1.7 h46 m2.5 h31.080.9180.208
25% (1,095)57 m47 m1.7 h30.930.9160.208
12.5% (549)30 m46 m1.3 h30.040.9080.208
6.25% (277)22 m46 m1.1 h28.270.8860.207
3.125% (140)18 m44 m1.0 h25.250.8310.227
TABLE I: Evaluation of the effect the number of images has on reconstruction quality and overall timing.
IterationsPSNR ↑\uparrowSSIM ↑\uparrowLPIPS ↓\downarrowTraining Time
15,00025.290.8360.3057.4 m
30,00020.610.7160.41016.2 m
45,00029.100.9020.22824.3 m
60,00030.090.9110.21732.0 m
75,00030.890.9160.21139.0 m
90,00031.190.9180.20845.6 m
TABLE II: Evaluation of the effect the number of training iterations has on reconstruction quality and training time.
Fig. 4: Comparison with Splatfacto, Nerfacto, Instant-NGP, and TensoRF.

Explore similar work

CardsList

  1. Return-to-Home Feasible Micro-Aerial Vehicle Exploration for 3D Gaussian Splatting Reconstruction

    Oct 2, 2026Prajit Krisshnakumar, Fan Yang, Koichiro Niinuma3D ReconstructionUAV Navigation

    Fujitsu Research

  2. ZipSplat: Fewer Gaussians, Better Splats

    Jun 3, 2026Alexander Veicht, Sunghwan Hong, Dániel Baráth +13D Gaussian SplattingMulti-View 3D Reconstruction

    ETH Zürich · Microsoft

  3. From Images2Mesh: A 3D Surface Reconstruction Pipeline for Non-Cooperative Space Objects

    Apr 30, 2026Bala Prenith Reddy Gopu, Patrick Quinn, George M. Nehma +4Implicit Neural Representations3D Reconstruction

    Department of Aerospace, Physics and Space Sciences, Florida Institute of Technology, 150 W University Blvd, Melbourne, 32901, FL, USA · Creare LLC, Hanover, 03755, NH, USA