English

Transformers As Generalizable Optimal Controllers

Systems and Control 2026-03-17 v1 Robotics Systems and Control

Abstract

We study whether optimal state-feedback laws for a family of heterogeneous Multiple-Input, Multiple-Output (MIMO) Linear Time-Invariant (LTI) systems can be captured by a single learned controller. We train one transformer policy on LQR-generated trajectories from systems with different state and input dimensions, using a shared representation with standardization, padding, dimension encoding, and masked loss. The policy maps recent state history to control actions without requiring plant matrices at inference time. Across a broad set of systems, it achieves empirically small sub-optimality relative to Linear Quadratic Regulator (LQR), remains stabilizing under moderate parameter perturbations, and benefits from lightweight fine-tuning on unseen systems. These results support transformer policies as practical approximators of near-optimal feedback laws over structured linear-system families.

Keywords

Cite

@article{arxiv.2603.14910,
  title  = {Transformers As Generalizable Optimal Controllers},
  author = {Turki Bin Mohaya and Maitham F. AL-Sunni and John M. Dolan and Peter Seiler},
  journal= {arXiv preprint arXiv:2603.14910},
  year   = {2026}
}

Comments

6 pages

R2 v1 2026-07-01T11:21:39.979Z