Research topics
Control of Hard Disk Drives

Hard-Disc-Drives (HDDs) have been at the center of the Personal Computer (PC) revolution during the last few decades. Indeed, in the last 20 years alone, HDD storage capacity grew from few Gigabits to several Terabits. Future evolution of these storage devices will require continued innovation by research communities to improve their performance and capacities. A typical HDD is an assembly of several rotating disks called platters controlled by a spindle motor to ensure their high speed rotation. Data, which are arranged in concentric disks, are read or written by means of electromagnetic heads mounted at the end of several head arms. Such components are allowed to displace on the platters surfaces. Through a current control, a Voice-Coil-Motor (VCM) actuator is dedicated to move the heads to reach the desired data track. The main components of a typical Hard-Disk-Drive are illustrated in figure below. There are two main functions for a HDD servo-system. First, it must ensure a prompt displacement of the R/W head from its current track to another target position using a limited control effort (track seeking mode). Second, it must accurately position the head on the required track while information are being read or written (track-following mode). The control is often subject to various run-outs which have to be treated properly and compensated as much as possible. The low frequency modeling of the actuator is influenced by nonlinear effects of frictions and flex ures generated from the movements of the reading or writing phases of the head. The high frequency VCM dynamics can be expressed as a linear transfer function including several resonance modes. Both nonlinearities and resonance modes should be taken into account in the controller design. They present a great challenge since they are the major source of degradation of the servo-system performance.


Control of Underwater Vehicles

Underwater vehicles have gained an increased interest in the last decades given the multiple tasks they can accomplish in different application fields ranging from scientific to industrial or military. One of the potential applications concerns underwater inspection of structures (such as bridges, hydraulic dams, boat hulls, etc). This operation consists in primarily visual observations of the structure being inspected with quantitative measurements. In our research team, we are particularly interested in the autonomous control of tethered underwater vehicles also called Remotely Operated Vehicles (ROV). Different challenges in autonomous control of such systems arise from the inherent high nonlinearities and time varying behavior of the vehicle’s dynamics subject to hydrodynamic effects and external disturbances. In order to avoid degradation in the performance of the controlled system during a specific mission, the vehicle is expected to possess a self tuning ability and compensate for different kind of disturbances. That is why adaptive controllers are very popular for such systems. However, various problems are related to the implementation of an adaptive controller on an underwater vehicle such as i) the need for the persistency in excitation that can lead to a bad transient behavior, ii) the tuning of the adaptation gains that can lead either to instability (high gains) or slow down the convergence rate (small gains), iii) the need of an appropriate initialization of the estimated parameters requiring an a priori knowledge of the system. I aml particularily interrested in developping control schemes for such systems ... with their applications to the following two underwater vehicles (AC-ROV and L2ROV)

The AC-ROV (left) and the L2ROV (right)

Control of Parallel Robots

Parallel Kinematic Manipulators (PKM's) consist in using at least two kinematic chains to support the end-effector (called also traveling plate or nacelle), each of these chains containing at least one actuator. This configuration allows a distribution of the load between the different chains. Even though parallel manipulators have important advantages in terms of stiffness, speed/acceleration, accuracy and payload compared to their serial counterparts, it was shown that they have many singularities in the workspace. These singularities can be eliminated through actuation redundancy. A degree of actuation redundancy in a parallel manipulator is the integer representing the difference between the number of its actuators (actuated joints) and its degrees-of freedom (dof). The actuation redundancy also allows to increase the traveling plate accelerations and to homogenize the dynamic capabilities of the robot throughout its workspace, and can also allow for more safety in case of breakdown of individual actuators. Parallel robots are used in many industrial applications such as food packaging. Looking for short cycle times in these applications means obviously to look for short motion time, as well as stabilization time as short as possible, while guaranteeing the robustness of performances with respect to perturbation and changes in conditions of use. One interesting solution can lies in searching for the ways to multiply pick-and-place speed by 5 to 10: this is a major technological breakthrough enabled on strong scientific innovations in two basic sciences which are among the foundation of robotics, kinematic and dynamic design and automatic control. The first topic is the general architecture of the robot, where for instance actuation redundancy, can be used together with direct drive actuation. A second topic is vibration minimization (at stop points), and this problem can be tackled with a twofold strategy: as far as design is concerned, the solution can be the proposition of dynamically balanced architectures and on the control point of view, and the solutions can include nonlinear and adaptive control schemes

The PAR-2 (left) and R4 (right) parallel manipulators


Control of Underactuated Mechanical Systems

Underactuated mechanical systems are those systems possessing fewer actuators than degrees of freedom (generalized coordinates) i.e they have generalized coordinates that are not actuated, and this is a source of dynamic constraints which are generally non integrable and therefore second order non holonomic.

Many examples of such systems exist, and mainly in robotics, they include, among others, flexible joint and flexible link robots, space robots, the gymnast robots and particularly the acrobot, the pendubot, the planar vertical take-off and landing aircrafts (PVTOL), the undersea vehicles and other mobile robots.

Many difficulties are often exhibited by such systems, like nonlinear dynamics, complex internal dynamics, non holonomic behaviour, lack of feedback linearizability, nonlinear coupling between the actuated and the unactuated coordinates.

The under-actuation in the mechanical systems is generally introduced by intentional design to reduce the manufacturing cost, the weight, and/or failure rate, so the obtained systems may be able to perform complex tasks with a reduced number of actuators, but they require new approaches to design effective control strategies, therefore they constitute a rich framework of nonlinear control problems of both practical and theoretical interest, and for that they are attracting more and more attention of researchers from nonlinear control community. ...

The inertia wheeel inverted pendulum : An underactuated mechanical system

Control of Humanoid Robots

Legged robots are those mechanical structures which are able to move on ground by alternating support legs. The problem of control of such robots is more complex than the case of classical robot manipulators with a fixed base, inasmuch as the legs are not attached to the ground and consequently they can lift off during movement. Furthermore, control of legged robots should take into account stability of the whole structure of the robot during walking. Most successful legged robots have 2 (biped), 4 (quadrupeds) or 6 (hexapods) legs. This legs-over-wheels approach lends itself for use in all-terrain purposes seeing as legs are more effective in an uneven ground than wheels. One interesting class of legged robots is biped walking robots where mainly two sub-classes are studied, namely 2D and 3D biped robots. In biped walking locomotion, two gaits could be underlined.

Static walking which refers to a system that stays balanced by always keeping the center of mass (COM) of the system vertically projected over the polygon of support formed by feet. On the contrary, dynamic walking is not constrained in such a manner, therefore the COM may leave the support polygon for periods of time. When interesting in 3D biped walking robots, many control approaches have been proposed to resolve the problem of locomotion control. The basic control architecture in this context includes mainly three components, namely the trajectory generator, the controller and the stabilizer. The classical way to deal with stable trajectories generation is to use the concept of Zero Moment Point (ZMP). However, other criteria have also been proposed, such as the Contact Wrench Sum (CWS) [9] or the Foot Rotation Indicator. The aim of the second component is to track reference trajectories, generated by the pattern generator. In order to improve the performance and robustness of the controller with respect to numerical errors and external perturbations, the control architecture needs an outer control loop, this is achieved through the stabilizer. The stabilization could be performed at different levels by modifying trajectories either in joint space, in operational space, or in both.

The HOAP 3 humanoid robot (left) and SHERPA biped robot (right)