Planetary rovers face a distinctive navigation problem: terrain that is visually repetitive and featureless can cause SLAM systems to confuse physically distinct locations. Systems built on 3D Gaussian Splatting (3DGS) produce dense maps but rely on RGB appearance descriptors that degrade under this perceptual aliasing. HyperLoop, a new framework from researchers at DLR and academic partners, is the first to integrate hyperspectral imaging (HSI) into the loop closure module of a 3DGS-based SLAM system, using material-level spectral signatures to complement RGB features.

The authors introduce two hyperspectral descriptor families—Bag-of-Spectral-Words (BoSW) and PCA-K VLAD (PCAK-VLAD)—and evaluate them alongside SIFT-VLAD and NetVLAD RGB descriptors under two fusion strategies. Tests used a custom dataset from DLR's Moon-Mars Outdoor Test Site (MMOTS), with synchronized RGB-D and hyperspectral sequences from four planetary analog traverses totaling 65.5 meters.

The key finding is that the sensor's Red-NIR range (600–860 nm) provides limited discriminability for rock terrain. The best HSI fusion (SIFT-VLAD + BoSW with Strict-AND) achieved 5.04 cm ATE RMSE, within 1.8% of SIFT-VLAD alone (4.95 cm), and the improvement over NetVLAD (8.81 cm) was attributable to SIFT-VLAD rather than to the hyperspectral data. The authors conclude that future planetary SLAM missions should adopt SWIR sensors, which capture diagnostic mineral absorption features absent from the 600–860 nm range. As a single-paper result, this finding has not yet been independently replicated.