Characterizing Satellite Geometry via Accelerated 3D Gaussian Splatting

TL;DR

This work addresses autonomous geometry characterization of unknown non-cooperative RSOs to enable safe rendezvous and proximity operations. It proposes a low-compute 3D Gaussian Splatting pipeline that initializes from Structure-from-Motion data and renders novel views via differentiable, end-to-end optimization, with covariances constrained to ellipsoidal shapes and a loss combining $L_1$ and $L_{ ext{D-SSIM}}$ using $\lambda=0.2$. Hardware-in-the-loop experiments on a spaceflight-relevant testbed demonstrate that the method can train on-board and render high-quality views much faster than NeRF-based baselines, including under challenging lighting, with up to $45.84$ FPS on a consumer GPU. Compared to dynamic NeRF variants, static 3DGS provides superior efficiency and robustness for static satellite geometries, supporting autonomous GNC and RPO for OOS/ADR missions. The results establish 3D Gaussian Splatting as a practical tool for onboard 3D scene understanding in space, enabling component recognition, pose estimation, and trajectory planning with limited compute resources.

Abstract

The accelerating deployment of spacecraft in orbit have generated interest in on-orbit servicing (OOS), inspection of spacecraft, and active debris removal (ADR). Such missions require precise rendezvous and proximity operations in the vicinity of non-cooperative, possible unknown, resident space objects. Safety concerns with manned missions and lag times with ground-based control necessitate complete autonomy. This requires robust characterization of the target's geometry. In this article, we present an approach for mapping geometries of satellites on orbit based on 3D Gaussian Splatting that can run on computing resources available on current spaceflight hardware. We demonstrate model training and 3D rendering performance on a hardware-in-the-loop satellite mock-up under several realistic lighting and motion conditions. Our model is shown to be capable of training on-board and rendering higher quality novel views of an unknown satellite nearly 2 orders of magnitude faster than previous NeRF-based algorithms. Such on-board capabilities are critical to enable downstream machine intelligence tasks necessary for autonomous guidance, navigation, and control tasks.

Figures (5)

• Figure 1: Concept for DSS-based target object characterization, trajectory planning, approach and detumbling mahendrakar_real-time_2021
• Figure 2: ORION Testbed at Florida Tech
• Figure 3: Satellite mock-up with green screen (left) and after chroma key compsiting post-processing (right).
• Figure 4: Ground truth images of the satellite mock-up
• Figure 5: Comparison of different test cases and SfM methods.