Seminars Fall 2008

Le Laboratoire d’Automatique de l’EPFL a le plaisir de vous inviter aux séminaires selon la liste ci-après. Une mise à jour régulière des informations concernant ces séminaires est disponible à l’adresse sur cette page. En particulier, il est conseillé aux visiteurs externes de vérifier que les séminaires soient dispensés comme prévu ci-dessous.

Where: Salle de séminaire LA-EPFL, ME C2 405 (2nd floor), 1015 Lausanne

When: Friday at 10.15am
 


Fall 2008 seminars – updated 29.09.2008

 

Properties of Partial Least Squares (PLS) Regression and Differences between Algorithms

26.09.2008  Prof. B. Wise – President, Eigenvector Research, Inc., Wenatchee, WA, USA.

Partial Least Squares (PLS) regression is used extensively for multivariate calibration, i.e. to relate a number of easily measured variables, x, to a less easily obtained property value, y. The most common applications are spectroscopic and multi-sensor array calibrations. There are, however, several algorithms in common use for developing PLS models, including Non-linear Iterative PArtial Least Squares (NIPALS), SIMPLS and Lanczos Bidiagonalization.
The PLS method is introduced and compared with the similar Principal Components Regression (PCR). Examples of its use are given. The properties of PLS and the differences between the algorithms are highlighted. The recent controversy regarding different approaches to calculating model residuals is discussed. Examination of the properties of PLS leads to an understanding of when the methods lead to different results.

 

Embedded Model Predictive Control with Explicit Computational Constraints

17.10.2008 – Dr Collin Jones – Automatic Control Laboratory, ETH-Zürich, Switzerland.

One of the major challenges in modern embedded controller synthesis is the ability to specify available computational power and storage as input design parameters. Our current work develops new methods for making design tradeoffs among online processing time, storage space and controller performance for high-speed applications. This talk will focus on approximate solutions to ‘explicit model-predictive controllers’, in which the optimal control laws are pre-computed offline, making on-line computation fast and simple. The proposed framework allows the design engineer to specify the available online computational power and storage resources, and then synthesizes an approximate controller that guarantees both system stability and feasibility while maximizing performance within the allocated complexity.

 

Dextrous Manipulation with Multifingered Robotic Hands

14.11.2008 – Prof. Zexiang Li – Dept of ECE, Hong Kong Univ of Science and Technology

In this talk, I will present research done by my group and others in an effort to develop a theory of manipulation by multifingered robotic hands. This work started in the middle 1980’s and continues to be an active area of research today in the robotics community. Problems to be addressed include: Motivation of multifingered robotic hands; grasping kinematics and grasp planning; force optimization subject to covex constraints; contact kinematics and nonholonomic motion planning; and a unified control system architecture for multifingered manipulation.

 

Nonlinear Control Strategies for an Overhead Travelling Crane

21.11.2008 – Prof. Harald Aschemann – Lehrstuhl für Mechatronik, Universität Rostock, Deutschland.

In this lecture, two nonlinear trajectory control schemes for an overhead travelling crane are presented: extended linearisation and passivity-based control. Aiming at an increased handling frequency and a fully automated crane operation, the focus has to be on the crane load motion. The design of the control structure is based on a control-oriented multibody system. Feedback control provides tracking of desired trajectories for the crane in the 3-dimensional workspace with small tracking errors. However, the tracking capabilities concerning desired trajectories for the crane load can be significantly improved by introducing feedforward control based on an inverse system model. Furthermore, a reduced-order disturbance observer is utilised for the compensation of nonlinear friction forces. At this, feedforward and feedback control as well as observer-based disturbance compensation are adapted to the varying system parameters rope length as well as load mass by gain-scheduling techniques. Thereby, the capabilities of the overhead crane can be extended in order to use it as a robot manipulator for the handling of heavy loads in a large cartesian workspace. Experimental results of an implementation on a 5 t-crane show both excellent tracking performance and high steady-state accuracy.

 

Title not available at this time

5.12.2008 – Prof. Jan van Impe – Dept. Chemische Ingenieurstechnieken, Heverlee, Belgium.

Postponed in 2009.

 

Time-scale change with an application to car autopilots

19.12.2008 – Prof. Balint Kiss – Budapest University of Technology and Economics (BUTE), Budapest, Hungary.

A time-scaling scheme is presented such that the scaling function depends on the state variables of the system and on an external variable referred to as the time-scaling input. The time-scaling scheme is applied to create a semi-autonomous robot-pilot for a car with one input which is the steering angle. The longitudinal velocity of the car is a measurable external signal and cannot be influenced by the controller. Using an on-line time-scaling, driven by the longitudinal velocity of the car, exponential tracking of any sufficiently smooth reference trajectory with non-vanishing velocity is achieved, similar to the differentially flat case with two control inputs. The time derivatives of the longitudinal velocity and the time-scaling output are no longer necessary if one designs the tracking controller for the linearized error dynamics around the reference path. The tracking controller is realized on a Ford Focus type car the measurement results illustrate the applicability of the method.