Controllability-Constrained Deep Network Models for Enhanced Control of Dynamical Systems

Suruchi Sharma,Volodymyr Makarenko,Gautam Kumar,Stas Tiomkin

2023-11-11

Abstract:Control of a dynamical system without the knowledge of dynamics is an important and challenging task. Modern machine learning approaches, such as deep neural networks (DNNs), allow for the estimation of a dynamics model from control inputs and corresponding state observation outputs. Such data-driven models are often utilized for the derivation of model-based controllers. However, in general, there are no guarantees that a model represented by DNNs will be controllable according to the formal control-theoretical meaning of controllability, which is crucial for the design of effective controllers. This often precludes the use of DNN-estimated models in applications, where formal controllability guarantees are required. In this proof-of-the-concept work, we propose a control-theoretical method that explicitly enhances models estimated from data with controllability. That is achieved by augmenting the model estimation objective with a controllability constraint, which penalizes models with a low degree of controllability. As a result, the models estimated with the proposed controllability constraint allow for the derivation of more efficient controllers, they are interpretable by the control-theoretical quantities and have a lower long-term prediction error. The proposed method provides new insights on the connection between the DNN-based estimation of unknown dynamics and the control-theoretical guarantees of the solution properties. We demonstrate the superiority of the proposed method in two standard classical control systems with state observation given by low resolution high-dimensional images.

Systems and Control,Artificial Intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to ensure that these models are controllable in the sense of control theory when using deep neural networks (DNNs) to estimate dynamic system models from high - dimensional data (such as images). Specifically, the paper points out that although modern machine - learning methods, such as deep neural networks, can estimate models of dynamic systems from control inputs and the corresponding state - observed outputs, and these data - driven models are often used to derive model - based controllers, there is usually no guarantee that the models represented by DNNs will be controllable in the formal sense of control theory. This lack of controllability guarantee often limits the use of DNN - estimated models in applications that require formal controllability guarantees. Therefore, the paper proposes a control - theory approach to explicitly enhance the controllability of models estimated from data by adding controllability constraints to the model - estimation objective. This method not only improves the controllability of the model, but also reduces long - term prediction errors and makes the model easier to be interpreted by control - theory quantities. The paper verifies the effectiveness of the proposed method through experiments on two standard classical control systems (the inverted pendulum and the cart - pole system).