Predictive Missile Guidance with Online Trajectory Learning

Abstract

INtroductIoNVarious missile guidance laws have been developed and studied since the World War II. The proportional navigation 1 (PN) is quite celebrated among these laws, due to its simplicity and efficiency. First studies on modern guidance laws based on modern control and estimation theory were conducted during 1960's. However at that time, implementation of these methods was not feasible because of the computational restrictions. Yet, with the proceeding technology and increasing capability of the targets, the aforementioned attitude towards modern laws began to change during late 1970's 2-8 . The problem of coping with modern day targets (e.g., intercepting a hostile aircraft) with advanced manoeuverability skills, calls for utilisation of estimation techniques within guidance algorithms. By making use of these estimation methods, it is possible to predict the future trajectory of the target. The basic idea behind predictive guidance is enabling the interceptor to take advantage of the estimated future trajectory of the target and modify the guidance law according to these estimations. Talole & Banavar 9 have shown that the PN law modified with predictive control is superior to the PN law itself in terms of control effort. Prabhakar 10 , et al. demonstrated that predictive guidance is capable of exhibiting a significantly improved performance. However, most of the existing predictive guidance laws assume full or partial knowledge of the target's dynamics. In this paper, estimations are based on a learning process, which utilises the noisy measurement of the target positions by passing them through a recursive least squares (RLS) estimation algorithm. This study presents a predictive guidance scheme for tactical missiles. The modern day targets, with improved manoeuverability, have revealed insufficient performance of the conventional guidance laws. The underlying cause of this poor performance is the reactive nature of the conventional guidance laws such as proportional navigation (PN) and pure pursuit (PP). Predictive guidance offers an alternative approach to the classical methods by taking proactive actions by estimating target's future trajectory. However, most of the existing predictive guidance approaches assume that the interceptor have a model of the target dynamics. A guidance strategy is developed in this study, that can learn the target dynamics iteratively and adapt the interceptor actions accordingly. A recursive least squares (RLS) estimation algorithm is employed for learning and estimating the possible future target positions, and a fixed horizon nonlinear program is employed for selecting the optimal interception action. Monte-Carlo simulations show that the guidance algorithm introduced in this work demonstrates a significantly improved performance compared to the alternatives in terms of interception time and miss distance.

Accurate Prediction and Estimation of 3D-Repetitive-Trajectories using Kalman Filter, Machine Learning and Curve-Fitting Method

“…A sequence of observation is gathered which fills the matrix equation. The Least Squares solution of the observation sequence provides the estimation of evolution matrix at every sample step, so that co-ordinates of the center of curvature, i.e (p e0 , p n0 ) in (7), is available at every time step.…”

“…The improvement of tracking/interception performance using several methods, is reported in [2], [3], [4]. Predictive guidance and learning based guidance are also proposed to improve the interception performance, as seen in [5], [6], [7], [8]. Methods pertaining to the field of soft computing applied in predictive guidance also provide promising results as seen in [9], [10], [11].…”

Target State Estimation and Prediction for High Speed Interception

“…The future states are expressed as a function of previous states as shown in equation ( 5), where j is the index of the observations, and the system of eqns. with the evolution matrix ([cosδ sinδ] T ) is as shown in (6). A sequence of observation is gathered which fills the matrix equation.…”

“…The improvement of tracking/interception performance using several methods, is reported in [1], [2], [3]. Predictive guidance and learning based guidance are also proposed to improve the interception performance, as seen in [4], [5], [6], [7]. Methods pertaining to the field of soft computing applied in predictive guidance also provide promising results as seen in [8], [9], [10].…”

Design and analysis of stable differentiator-predictor

“…This paper focuses on the problem of prediction of future signal and its derivatives for stochastic signal. Estimation of future signals is important for many applications, for examples, future trajectory prediction for missile interception (Akcal & Ure, 2017), short-term stock prediction (Weng, Lu, Wang, Megahed, & Martinez, 2018), et al Figure 1 shows the future trajectory prediction for missile-interception system. As the speed of attack missile is very fast, the usual methods of real-time interception control need the motion of defense missile much faster and more flexible.…”