Stable and interpretable DDPG controllers: Integrating control Lyapunov functions and symbolic regression via kolmogorov-arnold networks

2026-04-28

PRE

Roham Farhadi

Ramin Vatankhah *

Navid Eqra

DOI：10.1016/j.isatra.2026.04.030

Original

Original request for help

Save

Abstract

En 中文

• Stability-constrained DDPG ensures safe RL via Control Lyapunov Function filters. • Backup controller trajectories improve RL data efficiency via demonstration learning. • Symbolic policies are extracted using Kolmogorov–Arnold Networks (KANs). • Proposed method enables formal stability verification of RL controllers. • Framework improves robustness and convergence over standard RL baselines.

Keywords:

Stable DDPG

Control Lyapunov Function

Symbolic policy extraction

Kolmogorov–Arnold Networks

Reinforcement learning safety

Stable and interpretable DDPG controllers: Integrating control Lyapunov functions and symbolic regression via kolmogorov-arnold networks

Abstract

Journal

Organization

Cited Papers

Citing Papers