Neural network optimal feedback control with enhanced closed loop stability
This addresses stability issues in neural network-based optimal feedback control for high-dimensional nonlinear systems, which is an incremental improvement in ensuring reliable controller performance.
The paper tackled the problem that neural network controllers with high test accuracy can fail to stabilize dynamic systems, and proposed two architectures that locally approximate LQR to reliably produce stabilizing controllers without sacrificing optimality, as confirmed by numerical simulations.
Recent research has shown that supervised learning can be an effective tool for designing optimal feedback controllers for high-dimensional nonlinear dynamic systems. But the behavior of these neural network (NN) controllers is still not well understood. In this paper we use numerical simulations to demonstrate that typical test accuracy metrics do not effectively capture the ability of an NN controller to stabilize a system. In particular, some NNs with high test accuracy can fail to stabilize the dynamics. To address this we propose two NN architectures which locally approximate a linear quadratic regulator (LQR). Numerical simulations confirm our intuition that the proposed architectures reliably produce stabilizing feedback controllers without sacrificing optimality. In addition, we introduce a preliminary theoretical result describing some stability properties of such NN-controlled systems.