State estimation is a fundamental and challenging problem in many applications involving planning and control, in particular when dealing with systems exhibiting a nonlinear dynamics. While the design of nonlinear observers is an active research field, the issue of actively optimizing over time the convergence rate and/or transient behavior of the estimation error has not received, to the best of our knowledge, a comparable attention. In this talk, we discuss some recent results in this direction: we show how to characterize and tune the convergence rate/transient response of a particular class of nonlinear observers stemming from the adaptive control community. This is achieved by suitably acting on both the estimation gains and the inputs applied to the system under observation (thus, the "active" component of the approach). The theory is validated by simulation end experimental results on three classical structure from motion (SFM) problems: depth estimation for point feature, and 3D structure estimation for a planar and a spherical target. The results fully support the theoretical analysis and clearly show the benefits of the proposed active strategy, which is in particular able to enforce an error estimation dynamics equivalent to that of a reference linear second-order system with desired poles.
Paolo Robuffo Giordano is a CNRS researcher at IRISA/INRIA in Rennes since December 2012. He received the M.Sc. degree in Computer Science Engineering and the Ph.D. degree in Systems Engineering from the Dipartimento di Informatica e Sistemistica, Università di Roma "La Sapienza", in 2001 and 2008. He was a PostDoc at the Robotics Institute of the German Aerospace Center (DLR) from 2007 to 2008, and then moved to the Max Planck Institute for Biological Cybernetics in Tübingen, Germany, as Project Leader of the group Human-Robot Interaction from 2008 to 2012. His interests are in the general areas of robotics and nonlinear control. In particular, he has been working on kinematic and dynamical modeling of physical systems, motion control of fixed and mobile manipulators, visual servoing, nonlinear state estimation, nonholonomic systems, control design for VR applications, motion simulation technologies, aerial robotics, bilateral teleoperation, and multi-robot systems.