Unsupervised Change Detection for Space Habitats Using 3D Point Clouds

TL;DR

This paper tackles unsupervised 3D change detection in space habitats by detecting appearances and disappearances directly from depth point clouds. It introduces the EM-EMD algorithm, which first clusters two time-separated point clouds with Gaussian Mixture Models via Expectation-Maximization, uses a Minimum Description Length penalty to select the optimal number of components, and then measures change with Earth Mover's Distance to greedily extract the most changed regions. Validation on Astrobee data from the Granite Lab demonstrates the method's ability to detect multiple changes without labeled data, using both reconstructed maps and single-frame depth data, with a runtime analysis that highlights where computation dominates. The work provides open-source code and lays groundwork for robust, scalable, depth-based change detection in autonomous robotic maintenance of future deep-space habitats.

Abstract

This work presents an algorithm for scene change detection from point clouds to enable autonomous robotic caretaking in future space habitats. Autonomous robotic systems will help maintain future deep-space habitats, such as the Gateway space station, which will be uncrewed for extended periods. Existing scene analysis software used on the International Space Station (ISS) relies on manually-labeled images for detecting changes. In contrast, the algorithm presented in this work uses raw, unlabeled point clouds as inputs. The algorithm first applies modified Expectation-Maximization Gaussian Mixture Model (GMM) clustering to two input point clouds. It then performs change detection by comparing the GMMs using the Earth Mover's Distance. The algorithm is validated quantitatively and qualitatively using a test dataset collected by an Astrobee robot in the NASA Ames Granite Lab comprising single frame depth images taken directly by Astrobee and full-scene reconstructed maps built with RGB-D and pose data from Astrobee. The runtimes of the approach are also analyzed in depth. The source code is publicly released to promote further development.

Figures (7)

• Figure 1: The EM-EMD algorithm first clusters input point clouds $\bm{\mathrm{S}}^{t_0}$ and $\bm{\mathrm{S}}^t$ into Gaussian distributions using Expectation-Maximization (EM). Changes are detected using the Earth Mover's Distance (EMD) to compare the distributions.
• Figure 2: The Granite Lab at NASA Ames Research Center simulates the visual environment of the ISS and mimics 3-DOF microgravity by mounting Astrobee on a near-frictionless air bearing carriage. This system is used for testing Astrobee perception, motion planning, and control.
• Figure 3: Three large objects were placed in, moved, or removed from each scene. These images are captured with the NavCam on Astrobee for Scene 5. (Left) An Astrobee was placed near the edge of the scene. (Middle) A crate was placed in one corner of the scene. (Right) A cargo bag was placed in one corner of the scene.
• Figure 4: Scene change detection with the EM-EMD algorithm identifies changes on reconstructed maps of the environment. (Left) Two objects, an Astrobee and cargo bag, are added along one wall of the scene, and this region along the wall is identified as a change region. (Right) Three objects, an Astrobee, cargo bag, and crate, are each placed along separate walls in the scene. The cargo bag and crate are correctly detected areas of change (TP), while significant noise at the map boundary led to a shift in the change region. Astrobee was not correctly identified as a scene change (FN).
• Figure 5: A cargo bag and Astrobee robot are detected by comparing two single-frame depth images taken at the same location at different times. (Top) Cargo bag detection. (Bottom) Astrobee detection. (1 Stage) Images and input point clouds. (2 Stage) GMM clustering using EM. (3 Stage) Change detection using EMD.