Agile Tradespace Exploration for Space Rendezvous Mission Design via Transformers

TL;DR

The paper tackles rapid, multi-objective design of spacecraft rendezvous trajectories under nonconvex constraints by marrying an extended Autonomous Rendezvous Transformer (ART) with Sequential Convex Programming (SCP). It formalizes a MO-OCP for RPOD missions, extends ART to generate near-Pareto trajectory sets in batched inference, and validates generalization across flight times and diverse orbital dynamics using perturbed Keplerian models and SpaceTrack debris data. The results show ART can produce high-quality initial guesses that accelerate SCP convergence and serve as an effective surrogate for the Pareto front, with runtimes approaching those of convex relaxations. This framework enables agile tradespace exploration for real-world rendezvous planning and can extend to broader autonomous space missions.

Abstract

Spacecraft rendezvous enables on-orbit servicing, debris removal, and crewed docking, forming the foundation for a scalable space economy. Designing such missions requires rapid exploration of the tradespace between control cost and flight time across multiple candidate targets. However, multi-objective optimization in this setting is challenging, as the underlying constraints are often highly nonconvex, and mission designers must balance accuracy (e.g., solving the full problem) with efficiency (e.g., convex relaxations), slowing iteration and limiting design agility. To address these challenges, this paper proposes an AI-powered framework that enables agile mission design for a wide range of Earth orbit rendezvous scenarios. Given the orbital information of the target spacecraft, boundary conditions, and a range of flight times, this work proposes a Transformer-based architecture that generates, in a single parallelized inference step, a set of near-Pareto optimal trajectories across varying flight times, thereby enabling rapid mission trade studies. The model is further extended to accommodate variable flight times and perturbed orbital dynamics, supporting realistic multi-objective trade-offs. Validation on chance-constrained rendezvous problems with passive safety constraints demonstrates that the model generalizes across both flight times and dynamics, consistently providing high-quality initial guesses that converge to superior solutions in fewer iterations. Moreover, the framework efficiently approximates the Pareto front, achieving runtimes comparable to convex relaxation by exploiting parallelized inference. Together, these results position the proposed framework as a practical surrogate for nonconvex trajectory generation and mark an important step toward AI-driven trajectory design for accelerating preliminary mission planning in real-world rendezvous applications.

Figures (7)

• Figure 1: This paper presents a framework for AI-powered agile mission design. Mission configurations (target orbits, boundary conditions, and flight-time ranges) serve as inputs to a trajectory generation pipeline (top), where a Transformer model autoregressively predicts control sequences that are propagated through the system dynamics (middle). Batch inference efficiently produces multiple Pareto-optimal trajectories across flight times that are encoded into a tensor of $[B (\text{batch size}),N(\text{time horizon}),D (\text{subsequence length})]$, which can be projected onto the multi-objective solution space as surrogate mission models. Selected point solutions can then be further refined using sequential convex programming (bottom).
• Figure 2: ART inference process. During autoregressive generation, ART only predicts the optimal control $\boldsymbol{u}_k$ while states are propagated via an available dynamics model.
• Figure 3: Sampled 100 trajectories from the dataset for two rendezvous scenarios. The lower panels provide zoomed-in views of the corresponding upper panels.
• Figure 4: Performance analysis of the SCP with ART initialization compared to that with CVX initialization, tested for 2,000 scenarios. The statistics are binned across the cumulative constraint violation of the CVX solution.
• Figure 5: Distributions of the orbital elements of 500 sampled debris.