Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RL-AVIST: Reinforcement Learning for Autonomous Visual Inspection of Space Targets

Matteo El-Hariry, Andrej Orsula, Matthieu Geist, Miguel Olivares-Mendez

TL;DR

RL-AVIST advances autonomous visual inspection for space targets by coupling high-fidelity 6-DOF spacecraft dynamics in the SRB with DreamerV3, a model-based RL algorithm, and comparative model-free baselines. The framework defines a POMDP where an 8-dimensional continuous thrust action space controls a $6$-DOF free-flyer to achieve precise, energy-efficient proximity maneuvers around static or reference-trajectory targets like the Lunar Gateway. Across generalist and specialist training regimes, DreamerV3 delivers superior trajectory fidelity and sample efficiency, demonstrating robust generalization across morphologies and inspection patterns, and transferring to realistic assets such as the ISS and Venus Express with accompanying visualizations. The work substantiates model-based RL as a viable path toward scalable, retrainable on-orbit autonomy and lays groundwork for sim-to-real transfer and perception-enhanced policies for real-time visual inspection.

Abstract

The growing need for autonomous on-orbit services such as inspection, maintenance, and situational awareness calls for intelligent spacecraft capable of complex maneuvers around large orbital targets. Traditional control systems often fall short in adaptability, especially under model uncertainties, multi-spacecraft configurations, or dynamically evolving mission contexts. This paper introduces RL-AVIST, a Reinforcement Learning framework for Autonomous Visual Inspection of Space Targets. Leveraging the Space Robotics Bench (SRB), we simulate high-fidelity 6-DOF spacecraft dynamics and train agents using DreamerV3, a state-of-the-art model-based RL algorithm, with PPO and TD3 as model-free baselines. Our investigation focuses on 3D proximity maneuvering tasks around targets such as the Lunar Gateway and other space assets. We evaluate task performance under two complementary regimes: generalized agents trained on randomized velocity vectors, and specialized agents trained to follow fixed trajectories emulating known inspection orbits. Furthermore, we assess the robustness and generalization of policies across multiple spacecraft morphologies and mission domains. Results demonstrate that model-based RL offers promising capabilities in trajectory fidelity, and sample efficiency, paving the way for scalable, retrainable control solutions for future space operations

RL-AVIST: Reinforcement Learning for Autonomous Visual Inspection of Space Targets

TL;DR

RL-AVIST advances autonomous visual inspection for space targets by coupling high-fidelity 6-DOF spacecraft dynamics in the SRB with DreamerV3, a model-based RL algorithm, and comparative model-free baselines. The framework defines a POMDP where an 8-dimensional continuous thrust action space controls a -DOF free-flyer to achieve precise, energy-efficient proximity maneuvers around static or reference-trajectory targets like the Lunar Gateway. Across generalist and specialist training regimes, DreamerV3 delivers superior trajectory fidelity and sample efficiency, demonstrating robust generalization across morphologies and inspection patterns, and transferring to realistic assets such as the ISS and Venus Express with accompanying visualizations. The work substantiates model-based RL as a viable path toward scalable, retrainable on-orbit autonomy and lays groundwork for sim-to-real transfer and perception-enhanced policies for real-time visual inspection.

Abstract

The growing need for autonomous on-orbit services such as inspection, maintenance, and situational awareness calls for intelligent spacecraft capable of complex maneuvers around large orbital targets. Traditional control systems often fall short in adaptability, especially under model uncertainties, multi-spacecraft configurations, or dynamically evolving mission contexts. This paper introduces RL-AVIST, a Reinforcement Learning framework for Autonomous Visual Inspection of Space Targets. Leveraging the Space Robotics Bench (SRB), we simulate high-fidelity 6-DOF spacecraft dynamics and train agents using DreamerV3, a state-of-the-art model-based RL algorithm, with PPO and TD3 as model-free baselines. Our investigation focuses on 3D proximity maneuvering tasks around targets such as the Lunar Gateway and other space assets. We evaluate task performance under two complementary regimes: generalized agents trained on randomized velocity vectors, and specialized agents trained to follow fixed trajectories emulating known inspection orbits. Furthermore, we assess the robustness and generalization of policies across multiple spacecraft morphologies and mission domains. Results demonstrate that model-based RL offers promising capabilities in trajectory fidelity, and sample efficiency, paving the way for scalable, retrainable control solutions for future space operations
Paper Structure (10 sections, 3 equations, 6 figures)

This paper contains 10 sections, 3 equations, 6 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. RL-AVIST Framework
  4. Control policies and training
  5. Experiments & Results
  6. Comparison of RL Algorithms on Generalist Training
  7. Testing Across Diverse Inspection Trajectories
  8. Visual Inspection Around Realistic Assets
  9. Conclusion
  10. Additional Visual Figures

Figures (6)

  • Figure 1: Training of multiple CubeSat morphologies to follow randomized velocity vectors in the vicinity of the Lunar Gateway. The target structure is shown for visual context only, while the agents learn generalist control policies through diverse randomized conditions.
  • Figure 2: Training performance (mean ± std across 3 seeds) for Dreamer, PPO, and TD3 on the generalist randomized goal-velocity task.
  • Figure 3: 3D tracking performance of the DreamerV3 policy across multiple inspection trajectories: capsule, circle, rectangle, lemniscate, Lissajous, and spiral.
  • Figure 4: Sequence of inspection frames showing the spacecraft trajectory around the Lunar Gateway. Each frame illustrates RGB, depth, and semantic views captured during different stages of the maneuver, highlighting accurate tracking and consistent perception throughout the inspection.
  • Figure 5: Sequence of inspection frames showing the spacecraft trajectory around the Venus Express probe.
  • ...and 1 more figures