Neuromorphic robust estimation of nonlinear dynamical systems applied to satellite rendezvous

Abstract

State estimation of nonlinear dynamical systems has long been driven by the goals of accuracy, computational efficiency, robustness, and reliability. With the rapid evolution of various industries, the demand for estimation frameworks that can simultaneously fulfill these factors has grown significantly. Leveraging recent advances in neuromorphic computing architectures, this study presents a neuromorphic approach for robust filtering of nonlinear dynamical systems called SNN-EMSIF (spiking neural network-extended modified sliding innovation filter). SNN-MSIF benefits the computational efficiency and scalability of SNNs with the robustness of EMSIF, which is an estimation framework for nonlinear systems with zero-mean Gaussian noises. Notably, the weight matrices of the networks are designed according to the system model, eliminating the need for a learning process. The efficacy of the approach is evaluated by proposing a spiking framework based on an extended Kalman filter (EKF). Through comprehensive Monte-Carlo simulations, the performance of EKF and EMSIF. Additionally, SNN-EMSIF is compared with SNN-EKF in the presence of modeling uncertainties and neuron loss by means of obtained RMSEs. Results demonstrate the validity of the proposed methods and highlight the superior performance of SNN-EMSIF in terms of accuracy and robustness. Furthermore, investigations into obtained runtimes and spiking patterns generated by the SNN-EMSIF provide compelling evidence of the achieved computational efficiency, with an impressive reduction of approximately 85% in emitted spikes compared to possible spikes. The SNN-MSIF framework presents a promising solution to address the challenges of robust estimation in nonlinear dynamical systems, opening new avenues for efficient and reliable estimation in various industries benefiting neuromorphic computing advantages.