Martin Řezáč: Delay compensation in visual servoing and Accelerometer-based feedforward disturbance rejection for Inertially stabilized camera platform

Fri, 06/08/2012

Open seminar in K14 seminar room at Karlov namesti 13/E. The talk will introduce two different topics related to the ongoing development of inertially stabilized camera platforms conducted by the AA4CC group:

1. Accelerometer based feedforward disturbance rejection

A computational design and an experimental verification of a vibration rejection scheme for an inertially stabilized platform mounted on a mobile base will be presented. The mechanical configuration of the platform is that of a standard double gimbal, which realizes a mass stabilization of the line of sight of an optoelectronic payload such as a camera an/or laser ranger finder. Nonideal static balancing of the payload gives rise to projection of the carrier vibrations onto a disturbing torque acting on the payload around the joint axes. This disturbance can be left for the existing inertial angular rate feedback to attenuate but a better solution can be devised. The cause of this disturbing torque—the linear acceleration of the carrier—can be measured, filtered and fed forward to the two direct drive motors. Experiments prove that significant vibration rejection is achieved with this feedforward compensation scheme.

2. Delay compensation in a dual-rate cascade visual servomechanism: modified Smith predictor and reset observer approaches

A Few simple techniques for compensation of a one-sampling-period delay in a slow-sampled outer (position or angle) loop within a cascade visual servomechanism that also includes a fast-sampled inner (velocity) feedback loop will be presented. The results are mainly relevant for visual servoing applications, since the velocity sensors such as tachometers or MEMS-based gyros usually achieve much higher sampling rates compared to computer vision systems used as position (or orientation) sensors. The proposed solutions only compensate for the motion of the camera and not the observed object; they are particularly useful when visual servoing is combined with inertial stabilization. The problem is solved using two different formalisms: first, the problem is cast as an instance of a reset system with periodic reseting of the observer state. Second, a technique based on the concept of a modified Smith compensator is proposed wherein the undelayed output of a mathematical model is replaced by the measured rate signal from the inner loop. Numerical simulations are used to highlight the behavior of the proposed algorithms. The paper also describes experimental results obtained with a real double gimbal camera system and it discusses some implementation issues.