High-resolution Shape Modeling of Ryugu from an Improved Neural Implicit Method

The Hayabusa-2 mission, conducted by JAXA, orbited and returned samples from asteroid (162173) Ryugu in 2018 (Watanabe et al., 2017). The Optical Navigation Camera Telescope (ONC-T) was designed to capture high-resolution images for modeling the shape of Ryugu (Watanabe et al., 2017). Currently, the publicly available shape models of Ryugu have been generated from two approaches: structure-from-motion combined with multi-view stereo (SfM-MVS; Watanabe et al., 2019), and a neural implicit method (NIM) based on neural radiance fields (NeRF; Chen et al., 2024). While these methods effectively reconstruct the asteroid’s overall geometry, they fail to accurately resolve fine surface details, highlighting the need for enhanced modeling techniques. Fortunately, the NIM shows significant promise in achieving high-fidelity 3D scene representation, even with limited image input.

In this study, we used an improved NIM to perform detailed shape modeling of Ryugu and accurately reconstruct its surface morphology (Chen et al., 2025). To capture fine-scale surface features, including boulders of varying sizes and shapes, our approach introduces a multi-scale deformable grid representation that flexibly incorporates neighborhood information with different receptive fields. In addition, the 3D points generated by the SfM-MVS method are used to provide explicit geometric supervision during training, enhancing the accuracy of surface reconstruction.

To evaluate the reconstruction performance, we used 61 ONC-T images acquired during the 'Box-C' operations, with a spatial resolution of approximately 0.6 to 0.7 meters. We demonstrate that our proposed NIM is capable of reliably reconstructing Ryugu’s shape from a limited number of images, yielding volume and surface area estimates that are more consistent with the SfM-MVS reference than those produced by the NIM model of Chen et al. (2024). Compared to the SfM-MVS model and the NIM model proposed by Chen et al. (2024), our method more effectively reconstructs both small-scale and large, irregularly shaped boulders, as evidenced by comparisons between real and synthetic images from 3D models. In addition, it successfully recovers terrain features in the polar regions, despite the limited coverage of the ONC-T imagery. Owing to its reduced dependency on dense image inputs, our approach also presents the potential to simplify mission planning for global shape modeling by relaxing image acquisition requirements.

References:

Chen et al., 2024. Neural implicit shape modeling for small planetary bodies from multi-view images using a mask-based classification sampling strategy. ISPRS Journal of Photogrammetry and Remote Sensing, 212, pp.122-145.

Chen et al., 2025. Modeling the global shape and surface morphology of the Ryugu asteroid using an improved neural implicit method. Astronomy & Astrophysics, 696, p.A212.

Watanabe et al., 2017. Hayabusa2 mission overview. Space Science Reviews, 208, pp.3-16.

Watanabe et al., 2019. Hayabusa2 arrives at the carbonaceous asteroid 162173 Ryugu—A spinning top–shaped rubble pile. Science, 364(6437), pp.268-272.