Towards Robust Spacecraft Trajectory Optimization via Transformers

TL;DR

This work extends the Autonomous Rendezvous Transformer (ART) to robust, chance-constrained optimal control for spacecraft rendezvous in Low Earth Orbit, addressing uncertainty from navigation, actuation, and unmodeled dynamics while ensuring passive safety. ART generates near-optimal, dynamically feasible warm-start trajectories that feed SCP, improving convergence speed and reducing infeasibility, with performance demonstrated across RTN and ROE state representations. A post hoc acceptance framework and a probabilistic margin formulation are introduced to filter outputs and enhance reliability for safety-critical onboard operation. The results indicate substantial gains in efficiency and robustness, marking progress toward reliable AI-assisted autonomous GNC for multi-spacecraft missions and providing a foundation for future closed-loop integration and more challenging mission domains.

Abstract

Future multi-spacecraft missions require robust autonomous trajectory optimization capabilities to ensure safe and efficient rendezvous operations. This capability hinges on solving non-convex optimal control problems in real-time, although traditional iterative methods such as sequential convex programming impose significant computational challenges. To mitigate this burden, the Autonomous Rendezvous Transformer (ART) introduced a generative model trained to provide near-optimal initial guesses. This approach provides convergence to better local optima (e.g., fuel optimality), improves feasibility rates, and results in faster convergence speed of optimization algorithms through warm-starting. This work extends the capabilities of ART to address robust chance-constrained optimal control problems. Specifically, ART is applied to challenging rendezvous scenarios in Low Earth Orbit (LEO), ensuring fault-tolerant behavior under uncertainty. Through extensive experimentation, the proposed warm-starting strategy is shown to consistently produce high-quality reference trajectories, achieving up to 30\% cost improvement and 50\% reduction in infeasible cases compared to conventional methods, demonstrating robust performance across multiple state representations. Additionally, a post hoc evaluation framework is proposed to assess the quality of generated trajectories and mitigate runtime failures, marking an initial step toward the reliable deployment of AI-driven solutions in safety-critical autonomous systems such as spacecraft.

Figures (10)

• Figure 1: Pipeline of ART inference for non-convex chance-constrained trajectory optimization.
• Figure 2: The workflow of SCP for CC-OCP.
• Figure 3: Training datasets for the LEO rendezvous scenario ($N_d=115,000$).
• Figure 4: Pipeline of the experiment (dataset generation, training, inference, and warm-start analysis) with two state representations: RTN state (top) and ROE state (bottom). The two pipelines illustrate parallel approaches for training ART on either RTN or ROE state representations. This work assesses performance across both representations, allowing system designers to choose their preferred representation based on specific application needs.
• Figure 5: Performance comparison of SCP with different warm-starting strategies (deterministic scenario), plotted alongside the initial constraint-violation budget of CVX solutions, $C_\text{CVX}(t_1)$.