NeuralExplorer: State Space Exploration of Closed Loop Control Systems Using Neural Networks

TL;DR

NeuralExplorer introduces a neural network-based framework for exploring the state space of closed loop control systems, enabling estimation of reachable sets for various dynamical systems including hybrid and neural network controllers.

Contribution

It presents a novel neural network approach to approximate sensitivity and inverse sensitivity for state space exploration in control systems.

Findings

Effective in linear and nonlinear systems

Applicable to hybrid and neural network controlled systems

Provides estimates of reachable sets

Abstract

In this paper, we propose a framework for performing state space exploration of closed loop control systems. Our approach involves approximating sensitivity and a newly introduced notion of inverse sensitivity by a neural network. We show how the approximation of sensitivity and inverse sensitivity can be used for computing estimates of the reachable set. We then outline algorithms for performing state space exploration by generating trajectories that reach a neighborhood. We demonstrate the effectiveness of our approach by applying it not only to standard linear and nonlinear dynamical systems, but also to nonlinear hybrid systems and also neural network based feedback control systems.