Inverse Optimal Control

This page includes explanations related to our inverse optimal control paper in ICCIA 2025.

Model-Free Inverse Optimal Control via Dynamic Linearization

Authors: Mohammadreza Gilak, Raziyeh Mehraban, Behzad Ahi, Mohammad Saleh Tavazoei
Venue: 11th International Conference on Control, Instrumentation and Automation (ICCIA), 2025

---

Overview

Inverse optimal control asks a subtle question:
Given observed system behavior, what cost function makes that behavior optimal?

In this work, we focus on linear quadratic regulators (LQRs) when the system dynamics are unknown. We propose a model-free, data-driven framework that recovers the state cost matrix directly from measured state and input trajectories—without requiring knowledge of the system matrices.

The core idea is to combine two powerful tools:

• Dynamic linearization from model-free adaptive control, which constructs an exact data-based representation of the system along its trajectories.
• Sliding-window inverse optimal control, which estimates the underlying cost function using short segments of data.

By reformulating the system into an extended state-space form and applying inverse optimal control locally over moving windows, we iteratively refine the estimated cost matrix. A simple filtering step then stabilizes the estimate over time.

For full technical details, proofs, and simulations, please refer to the published paper here.

You can cite this work as

Gilak, M., Mehraban, R., Ahi, B., & Tavazoei, M. S. (2025, November). Model-Free Inverse Optimal Control Based on Dynamic Linearization. In 2025 11th International Conference on Control, Instrumentation and Automation (ICCIA) (pp. 1-6). IEEE.