New RL-Feedback Linearization Hybrid Controls Rotor-Based Aero System with Stability Guarantees

A new paper from Gabriel da Silva Lima and Wallace Moreira Bessa presents a learning-based control framework that marries classical feedback linearization with reinforcement learning for adaptive control of nonlinear mechatronic systems. The control law is derived using Lyapunov stability analysis, guaranteeing closed-loop stability even under modeling uncertainties and external disturbances. The key innovation is a reinforcement learning module based on the REINFORCE-with-baseline algorithm, which estimates and compensates for unmodeled dynamics in real-time. By reducing the variance of policy-gradient estimates, the approach ensures stable policy updates during online adaptation. The framework is validated on a single-degree-of-freedom rotor-based AERO system, a common benchmark for nonlinear control.

Simulation results demonstrate accurate trajectory tracking, fast adaptation to changing conditions, and strong robustness against parameter variations and external disturbances. The hybrid approach combines the analytical guarantees of Lyapunov-based control with the adaptability of reinforcement learning, offering an effective solution for controlling nonlinear mechatronic systems often found in aerospace and robotics. The paper is part of the EURODINAME III proceedings and is available on arXiv (2607.00640). This work bridges a critical gap between theoretical control theory and data-driven learning, potentially enabling more reliable autonomous systems in safety-critical environments.