Section: New Results

Visual servoing

Visual servoing using the sum of conditional variance

Participants : Bertrand Delabarre, Eric Marchand.

Within our study of direct visual servoing, we propose a new similarity function: the use of the sum of conditional variance [31] that replace SSD or mutual information [3] . It has been shown to be invariant to non-linear illumination variations and inexpensive to compute. Compared to other direct approaches of visual servoing, it is a good trade off between techniques using the pixels luminanc,e which are computationally inexpensive but non robust to illumination variations, and other approaches using the mutual information, which are more complicated to compute but offer more robustness towards the variations of the scene.

Photometric moment-based visual servoing

Participants : Manikandan Bakthavatchalam, Eric Marchand, François Chaumette.

The direct visual servoing approaches that have been developed in the group in the recent years, either using the luminance of each pixel, or the mutual information [3] , or the sum of conditional variance described just above, allows reaching an excellent positioning accuracy. This good property is however counterbalanced by a small convergence domain due to the strong non linearities involved in the control scheme. To remedy to these problems, we started a study on using photometric moments as inputs of visual servoing. We expect to find again the nice decoupling and large convergence domain that we obtained for binary moments, without the need of any object segmentation.

Visual servoing using RGB-D sensors

Participants : Céline Teulière, Eric Marchand.

We propose a novel 3D servoing approach [43] that uses dense depth maps to perform robotic tasks. With respect to pose-based approaches, our method does not require the estimation of the 3D pose, nor the extraction and matching of 3D features. It only requires dense depth maps provided by 3D sensors. Our approach has been validated in servoing experiments using the depth information from a low cost RGB-D sensor. Thanks to the introduction of M-estimator in the control law, positioning tasks are properly achieved despite the noisy measurements, even when partial occlusions or scene modifications occur.

Visual servoing of cable-driven parallel robot

Participant : François Chaumette.

This study is realized in collaboration with Rémy Ramadour and Jean-Pierre Merlet from EPI Coprin at Inria Sophia Antipolis. Its goal is to adapt visual servoing techniques for cable-driven parallel robot in order to achieve acurate manipulation tasks. This study is in the scope of the Inria large-scale initiative action Pal (see Section  8.2.7 ).

Micro-Nanomanipulation

Participants : Eric Marchand, Le Cui.

In collaboration with Femto-ST in Besançon, we developed an accurate nanopositioning system based on direct visual servoing [20] . This technique relies only on the pure image signal to design the control law, by using the pixel intensity of each pixel as visual features. The proposed approach has been tested in terms of accuracy and robustness in several experimental conditions. The obtained results have demonstrated a good behavior of the control law and very good positioning accuracy: 89 nm, 14 nm, and 0.001 degrees in the $x,y$ and ${\theta }_z$ axes of a positioning platform, respectively.

We begin a work, within the ANR P2N Nanorobust project (see Section  8.2.4 ), on the development of micro- and nano-manipulation within SEM (Scanning Electron Microscope). Our goal is to provide visual servoing techniques for positioning and manipulation tasks with a nanometer precision.

Autonomous landing by visual servoing

Participants : Laurent Coutard, François Chaumette.

This study was realized in collaboration with Dassault Aviation with the financial support of DGA. It was concerned with the autonomous landing of fixed wing aircrafts on carrier by visual servoing. A complete system has been developed [12] . The vision part consists in detecting the carrier in the image sequence and then tracking it using either dense template tracking or our 3D model-based tracker [2] . The visual servoing part consists in computing particular visual features able to correctly handle the aircraft degrees of freedom. Perturbations due to the wind and carrier motions have also been considered. The complete system has been validated in simulation using synthetic images provided by Xplane simulator and a dynamic model of the aircraft provided by Dassault Aviation.