In this paper, a disturbance observer (DOB) based predictive control approach is developed for the image-based visual servoing of an inertially stabilized platform (ISP). As the limitation in degrees of freedom of a two-axes ISP, it is hard to estimate the variable feature depth of the target at each control cycle when using an uncalibrated camera, which brings the challenge in the design of the visual servoing controller. To this end, a depth-independent kinematic matrix that only involves nominal parameters is obtained by employing the partitioned scheme in the system modeling. The uncertain kinematics arising from the unknown feature depth, angular velocity tracking errors, and uncalibrated intrinsic parameters is considered as the lumped uncertainty. A discrete-time DOB is then constructed to estimate the lumped uncertainty in real time. Instead of taking an integral action to eliminate tracking errors induced by the uncertain kinematics, the disturbance estimation is actively incorporated into the receding optimization process of the predictive controller. The stability of the closed-loop system is fully analyzed. Experiments on tracking a moving target are performed to validate promising qualities of the proposed approach.

在本文中，为惯性稳定平台 (ISP) 的基于图像的视觉伺服开发了一种基于干扰观测器 (DOB) 的预测控制方法。由于两轴 ISP 自由度的限制，使用未标定的相机时难以估计每个控制周期中目标的可变特征深度，这给视觉伺服控制器的设计带来了挑战。为此，通过在系统建模中采用分区方案，获得仅涉及标称参数的深度无关运动学矩阵。由未知特征深度、角速度跟踪误差和未校准的固有参数引起的不确定运动学被认为是集总不确定性。然后构建离散时间 DOB 以实时估计集总不确定性。不是采取积分动作来消除由不确定运动学引起的跟踪误差，而是将扰动估计积极地纳入预测控制器的后退优化过程。对闭环系统的稳定性进行了全面分析。执行跟踪移动目标的实验以验证所提出方法的有希望的质量。