dLITE: Differentiable Lighting-Informed Trajectory Evaluation for On-Orbit Inspection

TL;DR

The paper introduces ∂LITE, a fully differentiable pipeline that couples a differentiable SGP4 orbit propagator with a photorealistic Mitsuba 3 renderer to optimize on-orbit inspection trajectories based on visual costs. By enabling end-to-end gradient-based optimization, the framework designs non-obvious relative orbits that minimize specular reflections while maintaining useful imaging distances, improving data quality for inspection tasks. Key contributions include a differentiable orbit propagation implementation, a photometrically accurate rendering and lighting model, and demonstrations showing reduced image saturation and more robust feature matches across multiple satellites. The work advances mission planning by integrating visual information quality directly into trajectory design, with potential extensions to more complex attitude dynamics and richer imaging costs.

Abstract

Visual inspection of space-borne assets is of increasing interest to spacecraft operators looking to plan maintenance, characterise damage, and extend the life of high-value satellites in orbit. The environment of Low Earth Orbit (LEO) presents unique challenges when planning inspection operations that maximise visibility, information, and data quality. Specular reflection of sunlight from spacecraft bodies, self-shadowing, and dynamic lighting in LEO significantly impact the quality of data captured throughout an orbit. This is exacerbated by the relative motion between spacecraft, which introduces variable imaging distances and attitudes during inspection. Planning inspection trajectories with the aide of simulation is a common approach. However, the ability to design and optimise an inspection trajectory specifically to improve the resulting image quality in proximity operations remains largely unexplored. In this work, we present $\partial$LITE, an end-to-end differentiable simulation pipeline for on-orbit inspection operations. We leverage state-of-the-art differentiable rendering tools and a custom orbit propagator to enable end-to-end optimisation of orbital parameters based on visual sensor data. $\partial$LITE enables us to automatically design non-obvious trajectories, vastly improving the quality and usefulness of attained data. To our knowledge, our differentiable inspection-planning pipeline is the first of its kind and provides new insights into modern computational approaches to spacecraft mission planning. Project page: https://appearance-aware.github.io/dlite/

Figures (9)

• Figure 1: $\partial$LITE is an end-to-end differentiable simulator for on-orbit inspection which produces non-trivial trajectories based on visual costs. We optimise orbital parameters for passive inspection trajectories that yield high-quality images of a target satellite by minimising specular reflections seen by the sensor.
• Figure 2: Images of an H-IIA upper stage rocket body captured by Astroscale's ADRAS-J mission JAXA2024ADRASJ. The mission was able to capture high-quality images of the rocket body over much of its inspection orbit (left). However, high dynamic range content, saturation and sensor bloom due to specular reflections make some images uninformative (right).
• Figure 3: An overview of the $\partial$LITE pipeline. Given a TLE for the chaser satellite and a simulation epoch, we compute a time-varying sun vector $\hat{\mathbf{l}}$, and satellite position at time $t$. The scene parameters are updated to produce rendered inspection imagery. From these observations we formulate a visual cost $\mathcal{L}_s$ capturing the specularity in the scene to backpropagate gradients through the pipeline (dotted lines).
• Figure 4: Simulated observation from our photometrically accurate simulation. Including photorealistic Earth models alongside representative visual models of a satellite, $\partial$LITE produces high fidelity visual simulations of satellites in orbit.
• Figure 5: Error of our differentiable SGP4 propagator with respect to existing software sgp4py2012github, using the Hubble Space Telescope as an example. We achieve similar results, with the benefit of our implementation being fully differentiable via JAX jax2018github. Similar numerical results were observed for CloudSat and Sentinel-6 (not shown).