Reproduce “Transformers for Trajectory Optimization with Application to Spacecraft Rendezvous” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tommaso Guffanti, Daniele Gammelli, Simone D'Amico, Marco Pavone
Researchers at Stanford University introduced the Autonomous Rendezvous Transformer (ART), a GPT-architecture-based framework that generates high-quality warm-starts for sequential optimizers in spacecraft rendezvous. This approach resulted in more fuel-efficient trajectories and faster convergence for complex non-convex optimal control problems.
Key insights
- Transformers, particularly when coupled with known system dynamics during inference, can accurately forecast and imitate complex optimal control policies and states over extended planning horizons.
- Representing trajectories as sequences of performance parameters (reward-to-go, constraint-to-go), states, and controls allows generative models to be conditioned for user-specified performance requirements.
- Combining data-driven generative models for warm-starting with robust sequential optimizers offers a hybrid approach that provides both computational efficiency and hard constraint satisfaction for non-convex optimal control problems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.