A Comparison of Visual Servoing from Features Velocity and Acceleration Interaction Models

Abstract

Visual Servoing has been widely investigated in the last decades as it provides a powerful strategy for robot control. Thanks to the direct feedback from a set of sensors, it allows to reduce the impact of some modeling errors and to perform tasks even in uncertain environments. The commonly exploited approach in this field is to use a model that expresses the rate of change of a set of features as a function of sensor twist. These schemes are commonly used to obtain a velocity command, which needs to be tracked by a low-level controller. Another approach that can be exploited consists in going one step further and to consider an acceleration model for the features. This strategy allows also to obtain a natural and direct link with the dynamic model of the controlled system. This study aims at comparing the use of velocity and acceleration-based models in feedback linearization for Visual Servoing. We consider the case of a redundant manipulator and discuss what this implies for both control techniques. By means of simulations, we show that controllers based on features acceleration give better results than those based on velocity in presence of noisy feedback signals.

Towards Dynamic Visual Servoing for Interaction Control and Moving Targets

“…Most works have presented their results on DVS in simulation and tackling just few of the real world issues like measurement noise [6], [7], or instead of generating the torque command from the features acceleration, a velocity signal command is obtained from numerical integration and sent to the robot, as in [19] for an MPC controller. In this work we aimed at pushing forward the state of the art by tackling several implementation aspects of a real implementation not covered in previous works.…”

“…In the past decades, most research efforts focused on the modeling of the existing relationships between descriptive features in the image and the motion of the sensor at kinematic level [1], [2], and in the development of control strategies to guide the chosen visual features towards the desired ones as summarized in [3], [4]. Although much research has been conducted in this sense, little attention has been devoted to the use of second order models linking the features accelerations to the robot torques (via the robot dynamical model), with some early attempts dating back more than two decades [1], [5] up to some more recent contributions [6], [7]. As it is well known, explicitly taking into account the robot dynamics allows to design controllers with superior performance, especially for what concerns the regulation of forces and interaction with the environment.…”

A General Visual-Impedance Framework for Effectively Combining Vision and Force Sensing in Feature Space

“…In this work, we instead leverage results on the derivation of the VS dynamics [17], [18] that do not depend on the particular choice of the visual features: this allows us to derive a general framework that can effectively combine vision and force sensing directly in feature space. This differs from previous works on this topic since the derivation of secondorder models has often been formulated ad hoc from the definition of the considered features.…”

“…† is the pseudo-inverse of the matrix in the argument. Even though we have considered the manipulator interacting with the environment in our derivation, if we do not have the force measurement, we arrive at an indirect control of the force through a position controller such as the one derived in [17], [18].…”

Integrating Features Acceleration in Visual Predictive Control

“…with a representing the relative acceleration of the sensor expressed on the camera frame and h s being a function of s, z and v. In particular, this last component can be written as a collection of quadratic forms [16]:…”