Lifelong Learning Control of Nonlinear Systems with Constraints Using Multilayer Neural Networks with Application to Mobile Robot Tracking

This paper presents a novel lifelong multilayer neural network (MNN) tracking approach for an uncertain nonlinear continuous-time strict feedback system that is subject to time-varying state constraints. The proposed method uses a time-varying barrier function to accommodate the constraints leading to the development of an efficient control scheme. The unknown dynamics are approximated using a MNN, with weights tuned using a singular value decomposition (SVD)-based technique. An online lifelong learning (LL) based elastic weight consolidation (EWC) scheme is also incorporated to alleviate the issue of catastrophic forgetting. The stability of the overall closed-loop system is analyzed using Lyapunov analysis. The effectiveness of the proposed method is demonstrated by using a quadratic cost function through a numerical example of mobile robot control which demonstrates a 38% total cost reduction when compared to the recent literature and 6% cost reduction is observed when the proposed method with LL is compared to the proposed method without LL.