Positive definiteness of truncated random return in distributional LQR with applications to distributionally robust verification and control
Distributional Linear Quadratic Regulation (LQR) is concerned with the cumulative quadratic cost — referred to as the random return — rather than its expectation, as in the classical frameworks. In recent research (Wang et al. 2025, 2023), it has been shown that a closed-form expression for this random return exists and can be well approximated by a finite number of random variables, which we call truncated random return. In this paper, we study the truncated random return in distributional LQR. We remove the potentially conservative condition in Teng et al. (2025) to demonstrate that the truncated random return admits a quadratic, positive definite form. Thanks to the (positive definite) quadratic form, we provide a semidefinite program-based solution to the distributionally robust verification problem of truncated random return. We further study the distributionally robust control problem, which is a challenging bi-level nonconvex problem. To address this, we develop an adjoint-gradient descent algorithm that optimizes a tractable upper bound of the original problem and prove its convergence to a stationary point. These results are illustrated with numerical examples.
Authors
- Yulong Gao (ORCID: https://orcid.org/0000-0003-2338-5487)
- Wei Chen (ORCID: https://orcid.org/0000-0003-4074-3100)
- Ruyi Teng
- Dan Wang
Institutions
- Peking University (CN)
- Imperial College London (GB)
- Nanjing University (CN)
Publication Details
- Journal
- Systems & Control Letters
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1016/j.sysconle.2026.106652
- Primary Topic
- Advanced Control Systems Optimization
- Type
- article
- Field-Weighted Citation Impact
- 0.00