Search papers, labs, and topics across Lattice.
This paper introduces Reptile-D-learning, a novel framework that enhances Learning-based Lyapunov Control (LLC) by utilizing the Reptile meta-learning algorithm to generalize across systems with varying parameters. By capturing shared dynamical structures, the approach enables the learning of a robust Lyapunov network initialization that adapts quickly to new parameter configurations without extensive retraining. Experimental results show that Reptile-D-learning outperforms traditional methods in both generalization and adaptation speed, addressing the challenges posed by parametric uncertainty in nonlinear control systems.
Rapidly adapting control systems can now leverage a shared Lyapunov network to maintain stability despite significant parameter shifts.
Learning-based Lyapunov Control (LLC) provides formal stability guarantees for nonlinear systems, but its validity relies heavily on accurate system models. Parameter variations and uncertainties may invalidate stability constraints, leading to costly retraining. Although D-learning can estimate Lyapunov derivatives without relying on explicit dynamics models, it remains limited by single-task dynamics and degrades under large parameter shifts. We propose Reptile-D-learning, a framework that leverages the Reptile meta-learning algorithm to capture shared dynamical structures across systems with different parameters, thereby learning a generalizable Lyapunov network initialization and a high-performance controller. Experiments on multiple nonlinear control systems demonstrate that Reptile-D-learning significantly improves both generalization and rapid adaptation to unseen parameter configurations.
