Annual Report Department


Quality control and total quality management



Yüklə 0,74 Mb.
səhifə7/27
tarix27.10.2017
ölçüsü0,74 Mb.
#15532
1   2   3   4   5   6   7   8   9   10   ...   27

2.6Quality control and total quality management


total quality management

2.6.1Reengineering and TQM in public organisations


TQM is nowadays widespread in private business, i.e. in industrial and service organisations. The development of business models (e.g. EFQM) allowed to integrate TQM in the overall business strategy, sometimes leading to in depth organisational reengineering.

Recently we witnessed a similar drive towards increased accountability and customer satisfaction in government sponsored activities (e.g. education, research,…) and in the public sector in general (e.g. police force, health inspection, judiciary,…).

The increasing demand for funded research in these specific settings generated research into the applicability of TQM techniques and Reengineering models in these areas.

The development of generic knowledge derived from these studies is the subject of this research. In a first step we develop appropriate self-assessment tools for public services.

A specific applied research project is carried out with the Ministry of Justice.

Related projects: Ministerie van Justitie, Parket van Leuven

Publications and reports: WP2003-41, WP2003-27

Scientific staff: L. Gelders, J. Vits

2.7Mechatronics and robotics


design and control of mechatronic systems

2.7.1Influence of hysteresis on the dynamical behaviour of mechanical systems


Most linear guideways that are commonly used in machine tools show significant hysteretic frictional behaviour. When accurate and faster machines are required, these non-linear hysteresis effects should be identified and incorporated in the control design. Therefore, the influence of hysteresis on the dynamical behaviour of a mechanical system is analysed first. The autonomous behaviour of a mechanical system with hysteretic friction can be analytically derived, whereas for the periodic forced excitation case approximate, harmonic balance methods are used. The results of the theoretical analysis have futhermore been experimentally validated on several set-ups. The experiments revealed the difficulty of identifying the actual hysteresis function parameters owing to their sensitive dependence on local guideway conditions (e.g. pre-load, straightness, ...). Therefore a new parameter, the equivalent damping ratio at resonance, which describes the global dynamics of the system in one quantity, is defined and analysed both theoretically and experimentally. These experiments showed that this newly defined parameter is less sensitive to the overall guideway position and is as such a better quantity to characterise the dynamics of the system with hysteretic friction.

The obtained results are to be used for the design of high performance controllers for systems exhibiting hysteretic friction. To reach this goal the influence of different controller parameters on the performance is currently studied, together with scheduling schemes for the controller that are based on the position of the guideway.



Related projects: Growth project GRD1-2000-2527 “Mecomat”, , VW-Foundation

Scientific staff: F. Al-Bender, H. Van Brussel, J. Swevers, W. Symens

2.7.2Control of motion systems with a position dependent dynamical behaviour


Mechatronic systems, e.g. machine tools and robots, consist of combinations of motion degrees of freedom spanning a certain working volume. Consequently, their geometrical configurations are variable and so are the loads seen by the motion actuators. The goal of this research is to find for this category of systems, control strategies that can incorporate these variations. This can be done in two ways: (i) a linear time invariant (LTI) controller that is stable for all parameter variations (robust control) or (ii) a controller that varies with the varying system (gain-scheduling). Both techniques have been investigated for a pick-and-place machine and experiments show that scheduling is necessary to achieve high performance.

Surprisingly, controllers designed using ad-hoc scheduling, where several LTI controllers designed for different configurations of the machine are linearly interpolated, have a higher performance than analytically scheduled controllers that are based on a LPV (Linear Parameter Varying) description of the system. The poor performance of the analytically scheduled controllers is due to the use of a (conservative) single quadratic Lyapunov function to guarantee the stability of the closed-loop system. Future research focuses on ways to reduce this conservatism and on more efficient methods to obtain LPV descriptions for varying systems.



Related projects: IAP P05/06-AMS

Publications and reports: 2002PP019, 2002PP116, 2002PP117, 2002PP170

Scientific staff: H. Van Brussel, J. Swevers, W. Symens

2.7.3Discrete time sliding mode control (DSMC) of a X-Y feed table


A two-degrees of freedom path tracking DSMC has been developed for a X-Y feed table. This controller has been implemented and experimentally validated on this test setup. Its performance has been compared to that of a traditional machine tool controller, i.e. a cascade PI/P tracking controller. The DSMC is designed using the novel reaching law method, where both the reaching mode and the sliding mode are considered in the controller design. The reaching law and the switching surface of the sliding mode are determined in one pole placement procedure. An important property of sliding mode control is its robustness e.g. with respect to input disturbances like friction, which makes this DSMC very attractive for machine tools.

The experimental results indeed reveal that the DSMC is superior with respect to tracking accuracy, especially at velocity reversal, thereby avoiding quadrant glitches.



Related projects: IAP P05/06-AMS

Publications and reports: 2001PP037, 2003PP028

Scientific staff: H. Van Brussel, J. Swevers, J. Wang

2.7.4Modelling and Identification of friction


The issue of friction characterisation and modelling has been steadily gaining in importance recently. Adequate knowledge of friction dynamics is important at many levels of machine design, optimisation and control. Our activity in this field followed two lines. Firstly, modelling: where, by combining the phenomenological mechanisms of friction with the contact scenario of asperities on two contacting surfaces, we succeeded in developing a novel generic model of dry friction at asperity level [2003PP091]. This model is able to simulate all experimentally observed aspects of friction behaviour such as hysteresis with nonlocal memory in the pre-sliding regime, frictional lag in the sliding regime, break-away force dynamics, transition behaviour between pre-sliding and gross sliding, and the stick-slip phenomenon. At the same time, since this model is too complex for practical simulation and, in particular, control applications, a simpler heuristic model was formulated, based on the results obtained from the generic model and our past experience with heuristic models (cf. the Leuven friction model). This new model, which we have christened the Generalised Maxwell Slip (GMS) model, has the advantage of capturing all the basic characteristics shown by the generic model in simple, tractable formulations.

Second, to be able to explore, validate and control actual friction force dynamics, we have designed and built an accurate tribometer. With this apparatus, we are able to perform very accurate measurements of pre-sliding hysteresis, break-away force dynamics, transition to sliding and back, and stick-slip behaviour.



Related projects: IAP P05/06-AMS, GOA99-04

Publications and reports: 2003PP091, 2003PP012, 2003PP090, 2003PP011, 2003PP057

Scientific staff: F. Al-Bender, V. Lampaert, J. Swevers, V. Lampaert

2.7.5Comparison of model and non-model based friction compensation techniques in the neighbourhood of pre-sliding friction


Two different friction compensation techniques have been implemented and tested on a machine tool table: a friction model based feedforward technique and a disturbance observer that does not depend on a friction model. For the first technique, the performance of different friction models proposed in literature is compared. The disturbance observer estimates the friction force using a Kalman filter with a second order random walk model. Both approaches are complementary, yielding that their combination results in accurate tracking performance if the Generalized Maxwell-slip friction model is used. The Generalized Maxwell-slip friction model is an advanced novel friction model developed at KUL-PMA.

Related projects: GOA99-04, IAP P05/06-AMS

Publications and reports: 2003PP012, 2003PP011, 2003PP057, 2003PP113

Scientific staff: J. Swevers, F. Al-Bender, V. Lampaert

2.7.6Geometric modeling and control of electro-mechanical systems


The objective of the project is to develop nonlinear energy-based models of electro-mechanical systems, and to derive passivity-based controllers using these models.

A framework has been developed to describe dissipative Hamiltonian and gradient control systems, and their interconnections. It can be used to model mechanical systems with nonholonomic kinematic constraints, and nonlinear electrical RLC circuits (Brayton-Moser model). Passivity-based controllers have been defined, effectively exploiting the geometry of these models. Furthermore, the order reduction of these models under invariance transformations has been described.



Related projects: IST-2001-34166 “Geoplex”

Publications and reports: 2003PP039, 2003PP041

Scientific staff: G. Blankenstein, international collaboration: R. Ortega (CNRS-Supelec, France), T.S. Ratiu (EPFL, Switzerland)

2.7.7Development of a flexible, anthropomorphic robot arm


More and more, robots are entering human-centred environments like offices and households. This shift does not only require research in AI but also has a large impact on the mechanical requirements on the robot. To avoid damage to the environment and the robot during (non-)voluntary contact, the actuators of this soft-robot should have adaptable compliance and the complete robot should have a low weight. Because a human arm is nature’s solution to this problem, this research started with a literature study of human muscles, the anatomy of the human arm, models of low-level muscle control and models of high level planning. In neuroscience literature, it is stated that the human, low-level control of muscles with inherent non-linear stiffness and damping could implement a huge part of the generation of smooth, humanlike movements. One model of this hypothesis, the lambda-model, is used as a reference for further developments of the low level control of the actuators of the soft-robot.

The second main item of this research is the quest for back driveable, lightweight actuators with hardware compliance. Pneumatics has, due to the compressibility of the air, an inherent compliance and two experimental setups have been built. The first one is using an antagonistic pair of pneumatic actuated muscles and using the measured pressures, forces and position, the control system allows one to control the position and stiffness of the joint. The second setup consists of a home-made frictionless, double acting piston. Theoretical maximal acceleration of this piston is up to 25G. The difficulty of controlling this fast actuator is enhanced by the large nonlinearity of the used proportional valve. Instead of trying to compensate this nonlinearity, first steps are taken into the development a piezo-actuated valve with almost perfect linear behaviour.



Scientific staff: H. Van Brussel, B. Corteville

2.7.8Dynamic balancing of reciprocating machinery


Input torque balancing is a well-known way to reduce drive speed fluctuations in reciprocating machinery. A novel mechanism, the cam-based centrifugal pendulum (CBCP), has been designed, optimized and realized in practice. Application of such a device to a weaving machine à blanc, shows that the CBCP (i) functions correctly and (ii) significantly enhances the dynamic behavior of the machine, by reducing the frame vibration by half.

Counterweight balancing of linkages, that is, the addition of counterweights to the mechanism links, so as to reduce the shaking force and shaking moment, exerted on the supporting frame, is a well-known and difficult mechanical design problem. It has been proven that it can be reformulated as a convex optimization problem. This implies that the counterweights are found with great computational efficiency, and guaranteed to be globally optimal. Furthermore, the developed convex framework is easily applicable to linkages of arbitrary complexity.



Related projects: IWT project 000270/Picanol

Publications and reports: 2001PP042, 2003PP009, 2003PP052, 2003PP172, 2004PP047, 2004PP048, 2004PP049, 2004PP050

Scientific staff: J. De Schutter, J. Swevers, B. Demeulenaere

micro and precision engineering


2.7.9Development of ultra precision ELID-grinding machine for free form surfaces


Traditionally, most optical surfaces are produced by grinding followed by lapping and polishing. More recently some lenses were produced by moulding but this also required the moulds to be made in the traditional way. This research aims at producing cheap, smooth and curved optical surfaces with average roughness value comparable to polishing and with a very high dimensional accuracy.

Current highly accurate ELID processing is only applicable to flat surfaces. Other high precision manufacturing processes are limited to symmetric surfaces or cannot guarantee high shape accuracy. The main innovation of this research will be the combination of very stiff and accurate positioning elements with an improved ELID process in order to allow producing highly accurate and smooth curved optical surfaces in an efficient way. The process will be applied to various materials covering metals, ceramics and glass.



This requires the development of the following innovative concepts:

  • A five-axis positioning system with an envisaged stiffness of about 5 times the current industrial practice.

  • A reference measurement system tracing tool and work piece with nanometer precision

  • An industrially reliable ELID grinding technology for curved surfaces.

Related projects: GROWTH Project no. GRD1-2001-40538 "Nano Grind"

Scientific staff: D. Reynaerts, H. Van Brussel, J. Qian, D. Hemschoote, P. Vleugels

2.7.10Development of a linear hydraulic micro actuator


During the last decades, a major research effort has been undertaken towards the development of "agile micro robots". This means, micro robots that exhibit a high number of degrees of freedom, a large workspace, and a high workload, all taking into account a restricted volume. However, to build these robots, several issues remain to be solved. This project will mainly address one of them: the development of intelligent micro-actuators with high power density. The objectives of these actuators are a diameter of 1 mm, a stroke of 10 mm, a speed of 10 mm/s, and a force in the Newton range. The actuator needs an integrated force and position sensor in order to realise a closed-loop controlled actuation. In order to achieve these objectives it is important to choose an appropriate actuation technology. In robotics, electromagnetic motors have replaced their hydraulic counterparts because of maintenance and control considerations. However, hydraulic actuators are still more appropriate for microapplications since they have higher power densities and they develop high forces without reduction. Despite these promising properties, hydraulic microactuators are rare in Micro System Technology (MST) because of inherent sealing problems. In 2003, a new microseal based on surface tension has been developed and tested successfully. Furthermore a new valve technology has been investigated.

Related projects: IAP P05/06 AMS

Scientific staff: D. Reynaerts, J. Peirs, M. De Volder

2.7.11A piezoelectric drive combining a stepping and a resonant actuation mode


Ultrasonic linear piezomotors have several advantages compared to electromagnetic motors in specific applications such as in high-vacuum conditions and where no magnetic field is allowed. These compact direct-drive motors are fast, but low speeds and high-resolution positioning are difficult to obtain. On the other hand, several non-resonant piezosteppers were developed, resulting in very accurate linear motors, but with a low speed. This research focuses on piezoelectric motors that can work in a stepping mode as well as in a resonant mode. The stepping mode allows a very accurate, but slow actuation. The resonant vibration mode allows coarse, but fast positioning. A first linear prototype that was built last year, has been extended this year to a planar version, enabling linear movement in two directions. Also a new type of linear piezomotor has been built for higher speeds. Both motors combine the fast positioning capability of resonant motors with a stepping function enabling fine positioning in the nanometer range.

Related projects: IAP P05/06 AMS, FWO138

Publications and reports: 2003PP064, 2003PP075, 2003PP169, 2003PP179

Scientific staff: D. Reynaerts, H. Van Brussel, S. Devos, W. Van de Vijver

2.7.12Design of a Miniature Zero-backlash Tilt-Pan Mechanism.


Echolocating bats achieve a surprising amount of autonomy primarily based on sonar sensing. In order to use insights from biosonar function to improve technical designs, it is necessary to understand the biosonar tasks (e.g., obstacle avoidance, prey capture, navigation). This research concerns the reproduction, at a functional level, of the echolocation system of bats by constructing a bionic bat head that can then be used to systematically investigate how the world is not just perceived but actively explored by bats. This bionic bat head must be of similar size to a real bat head to reproduce the relevant physics, which means a multi-degree of freedom ‘micro’-mechanical system has to be constructed to allow realistic pinnae movement.

The main mechanical focus of the researchs lies in designing a miniature mechanism that enables the pinna shapes to tilt and pan within a range of 60 degrees with an accuracy of less than 0.1 degrees, while minimizing the distance between both pan-axes. To meet these requirements, a miniature zero-backlash tilt-pan mechanism based on differential gear technology has been designed, constructed and tested. Instead of normal gears, a pretensioned cable is used to transmit the movement of the motor pulleys at the back of the system to the differential pulleys at the front, which allows to construct a compact system that doesn’t interfere with the acoustical field. Furthermore, an actuator that enables tragus deformation has been designed.



Related projects: IST Project no. 35144, “Circe”

Publications and reports: verwijzen naar volgnummer in Section 6

Scientific staff: D. Reynaerts, H. Bruyninckx, F. Schillebeeckx, W. Vandevijver, N. Salustiano

2.7.13Development of a micro gas turbine unit for electric power generation


A miniature gas turbine is under development for application in small portable power generators, producing electrical energy from fuel. Such systems could replace batteries as power supply for portable devices with high power demand. As fuel has a much higher energy density than batteries, these portable devices could gain substantial autonomy. To retain the high energy density of the system as a whole, the power generator itself should be as compact as possible. A system based on a miniaturised turbine was chosen because of the high power density of turbines in general.

The main parts of the power unit are the compressor, turbine, combustion chamber and generator. First prototypes of two key components have been built and tested: a centrifugal compressor and an axial turbine. The turbine, with a rotor diameter of 10 mm, is tested with hot compressed air up to 130,000 rpm, generating 50 W of mechanical power and 36 W of electrical power. The 20 mm diameter compressor produces a pressure ratio of 1.20 at 150,000 rpm.

At the moment, the performance and efficiency of both turbine and compressor are clearly limited by the maximal speed that can be reached. Therefore, current research focuses on the development of high-speed air bearings and balancing methods for both turbine and compressor rotors. In parallel, other project partners are developing the combustion chamber, heat recuperator, generator and monitoring system.

Related projects: IWT SBO 030288 “PowerMEMS”, IAP P05/06 AMS

Publications and reports: 2003PP099, 2003PP121, 2003PP148, 2003PP184

Scientific staff: D. Reynaerts, F. Verplaetsen, J. Peirs, P. Vleugels

robots and intelligent machines


2.7.14Biologically inspired manipulation


This research is focussed on the future use of mobile manipulators in house cleaning and human caretaking; one could characterise these applications as a robotic butler. This type of robots hold great promise for significantly increasing the possibilities for elderly and handicapped people to gain more independence.

Most existing manipulators are designed and controlled based on industrial requirements, making them inappropriate to function around humans. In our opinion, this new environment and these new applications call for a re-evaluation of the methods used in current robotics. We aim at the domestication of manipulators, making them `friendlier', `nicer' and better suited for interaction with humans than their industrial counterparts. To achieve this goal, research into a more biologically-inspired behaviour-based approach to manipulation robotics has been done.

As a test platform, we use LiAS (Leuven intelligent Autonomous System). On the industrial manipulator, we have implemented a realtime controller in Orocos using RTAI-LXRT. It implements impedance control with adaptable impedance parameters at joint level. Above this hard-realtime layer, we have implemented a behaviour-based controller. This behaviour-controller is based on inspiration from human manipulation. As output behaviours start trajectories. The trajectory specification is derived from models describing human motion. Therefore, this approach to manipulation should be very appropriate for human tasks. As a test case, we have implemented the writing on a white board, where the execution of the task is purely behaviour-based in the sense described above.

Related projects: IAP P05/06 AMS

Publications and reports: 2003PP021, 2003PP022, 2003PP023, 2003PP154, 2003PP163

Scientific staff: H. Van Brussel, B.J.W.Waarsing, M. Nuttin

2.7.15Concepts for dynamic obstacle avoidance


The goal of the project was to develop a prototype free ranging AGV (Automated Guided Vehicle) for industrial environments . Dynamic obstacle avoidance is one of the key issues for practical applications of automated guided vehicles or mobile robots. When unexpected or un-modelled objects, static or moving, appear, the robot should have the ability of dynamically avoiding them, without any collision, and returning to the normal path after these objects have been passed or removed. In contrast to static and global path planning, the dynamic obstacle avoidance approach only uses the local information of the surrounding environment. From simulations and the actual experiments in industrial environments, the dynamic obstacle avoidance scheme presented in this paper is verified to have the following features:

- All the parameters involved have physical meanings, easy to be determined.

- The system has a strong ability to select optimal detour direction and to smoothly pass through narrow corridors.

- The system has a strong ability to run in relatively complicated or unknown environments.

- The concepts presented for dynamic obstacle avoidance can also be used in other applications. They provide a possible solution to underground navigation for mining industry, according to our preliminary simulation.

Related projects: IAP P05/06 AMS

Publications and reports: 2003PP008

Scientific staff: H. Van Brussel, M. Nuttin, F. Xu, R. Moreas

2.7.16Experimental identification of robot payload dynamics


Starting from a long experience in the field of robot identification, a new method has been developed to experimentally identify the robot payload inertial parameters. The method is based on periodic robot excitation and the maximum likelihood parameter estimation. The modelling approach however is a little bit different. Since we are mainly interested in the payload parameters, there is no need to identify the full manipulator again. The torque contribution of the robot links is compensated for based on a priori known dynamic robot model. The rotor inertias and inertial parameters of the links are set to a priori determined values. This results in an identification model that only depends on the payload inertial parameters and the friction parameters.

This method provides both (1) accurate estimates of the robot payload inertial parameters, and (2) accurate actuator torques predictions, both of which are essential for the acceptance of the results in an industrial environment. Experimental results on a KUKA industrial robot equipped with a calibrated reference payload have validated the industrial applicability of the developed method.



Related projects: IAP P05/06 AMS

Publications and reports: 2003PP020

Scientific staff: J. Swevers, W. Verdonck

2.7.17Robot programming by human demonstration: data interpretation and task extraction


The programming of robotic systems with integrated sensors is a complex problem, which demands specific programming skills. Because of this, the development of robot programs is still one of the major cost factors in the application of robots. Robot programming by human demonstration could be a solution. During the demonstration, several sensors provide measurement data (6D pose, contact formation out of measured forces and torques, eventually other sensors,...) about the task. In an interpretation step, a model of the relevant objects and environment is deduced from this data by means of Kalman Filter estimation. Using the same estimation techniques during task execution copes with uncertainties in the environment.

In order to execute complex tasks on robotic systems with multiple sensors, new task specification methods are needed. More specifically, a framework is being developed that allows for different, distributed specifications and which copes with over- or underspecification.



Publications and reports: 2003PP074

Scientific staff: J. De Schutter, H. Bruyninckx, J. Rutgeerts, P. Slaets, K. Gadeyne

2.7.18Robot programming by human demonstration: signal processing and data fusion


Programming a sensor controlled robot is often a difficult task that requires specific knowledge. An elegant solution to this problem would be programming the robot by demonstrating its task. This demonstration will be characterised by measurements of positions, forces and velocities. All the collected data will then be used to construct a formal task description.

The first step in this process, namely the signal processing and data reduction, has been developed during this year. An explicit geometric 3-D model has been constructed from the fusion of the different measurements. This non-minimal geometric model is then reduced to a minimal representation using a Bayesian statistical approach. This model can then be used to construct a task description that is robust against modification in the environment.



Related projects: GOA 99/04

Publications and reports: 2003PP074

Scientific staff: J. De Schutter, H. Bruyninckx, P. Slaets, J. Rutgeerts, T. Lefebvre, K. Gadeyne

2.7.19Robots in ambient intelligent environments


In the European ITEA project Ambience, K.U.Leuven co-operated with Philips, Barco, Epictoid, and the University of Amsterdam to develop a domestic user-interface robot, which was baptized “Lino”. Lino is able to have a natural human interaction by speech and emotional feedback and is able to navigate in a home environment. The natural interaction with the user is achieved by means of a mechanical head able to express emotions. The robot is aware of the position and identities of the users, both from visual and auditory information. The robot estimates its location in the environment with an appearance-based localization method using a stereo camera system. The navigation to the goal is achieved with a hybrid method, combining planning with reactive control. The robot is designed to operate in an intelligent environment, such that external information can be used to localize users and their intentions (context awareness), and that additional information can be retrieved from various databases in the environment. The result is a service robot that can have a simple dialogue with the user, provide information in a natural way (speech and expressions) and can be instructed to navigate to any specific goal in the environment.

Related projects: ITEA project no. 10260: “Ambience”

Publications and reports: 2003PP128, 2003PP156

Scientific staff: H. Van Brussel, M. Nuttin, E. Demeester

2.7.20An architecture for navigation in human-centred environments


In the context of the Ambience project, experiments with two different architectures for robot navigation in human-centred environments have been performed. The autonomous personal robot Lino is equipped with a hybrid architecture that is implemented using an agent-based software framework. The hybrid architecture combines the advantages of deliberative and reactive approaches. A gross motion planner is used for planning a path from start to goal in an a priori map of the environment. A reactive controller is responsible for executing the path while simultaneously avoiding unmodeled obstacles. The obstacle avoidance adopts the gross motion planning algorithm as well, since this algorithm takes the clearance to obstacles into account, and it is not affected by the local minima problem that potential field methods typically suffer from.

For the wheelchair Sharioto, a model-based, probabilistic framework was devised that continuously estimates the user's intention explicitly before trying to assist him or her. A fine motion planner is adopted to model how humans reason geometrically to move a vehicle from a certain position to another while moving around obstacles. In order to test the framework in practice, initial experiments on localisation and map building have started. Also, PMA co-operated with ESAT-PSI to extract natural features from camera images for localisation and topological navigation.



Related projects: ITEA project no. 10260: “Ambience”, STWW project no. ADV/980371/KUL-TW-MECH-PMA

Publications and reports: 2003PP053, 2003PP162

Scientific staff: H. Van Brussel, M. Nuttin, E. Demeester, G. Vanacker

2.7.21Intention estimation for human-machine interaction


Many elderly and disabled people today experience difficulties when manoeuvring an electric wheelchair. In order to help these people, several robotic assistance platforms have been devised in the past.

In contrast with previous approaches, a user-centered architecture for shared wheelchair control was proposed that continuously estimates the user's intention explicitly before trying to assist him or her. The actual navigation assistance is performed by a fine motion planner that takes the kinematic and dynamic constraints explicitly into account. Wheelchairs are only one of the many applications in which user plan recognition is important. This research topic plays an important role in all applications where human and machine co-operate, and where graphical user interfaces may be imperative for reliable man-machine interaction. Typical examples include driving a car and desktop applications.



The research currently focuses on how to model humans that interact with machines, how to recognise the plans users have, and how to adopt the information regarding their plans to execute the task more robustly and easily.

Related projects: ITEA project no. 10260: “Ambience”

Publications and reports: 2003PP050, 2003PP155, 2003PP159

Scientific staff: H. Van Brussel, M. Nuttin, E. Demeester

2.7.22Shared autonomy for wheelchair control: a learning approach


Moving constitutes one of the crucial problems to obtain independent living for severely disabled people. Electrical wheelchairs can be a great help, but some users lack the fine steering capacities required for "every-day-life" manoeuvres such as e.g. driving through a door. There are many explicit wheelchair user interfaces available. These enable users to convey their intention explicitly to the wheelchair control system. However, many users find it extremely difficult to do so. This research explores ways of implicit communication to assist the user to perform daily (but difficult for the user) manoeuvres safely. Implicit communication consists in the first place of: processing of the user signals and of the signals from environmental sensors. Work on this processing resulted already in some assisting algorithms: a collision avoidance algorithm (decelerates the wheelchair near obstacles), an obstacle avoidance algorithm and a table docking algorithm. Further research was done on how these algorithms can be combined in a general framework that activates the appropriate algorithm at the appropriate moment. This depends of course on the individual needs and handicap of the wheelchair user. A beginning was made to explore how learning techniques can be used to adapt the activation of the correct algorithm (in this case an algorithm or agent corresponds with a user intention) to the specific user.

Related projects: ITEA project no. 10260: “Ambience”; STWW project no. ADV/980371/KUL-TW-MECH-PMA

Publications and reports: 2003PP160, 2003PP051

Scientific staff: H. Van Brussel, M. Nuttin, D. Vanhooydonck, G. Vanacker, E. Demeester

2.7.23A learning approach to acquire new skills by demonstration


A new trend in mobile robotics is the increasing interest for the field of service robotics. Service robots will more and more appear in several aspects of human life: e.g. as a pet (like the Aibo dog from Sony), as an intelligent butler, as an interactive guide, as an intelligent wheelchair, etc. One of the challenges here is that non-professionals (non-programmers) are intended to operate these complex robot systems. Therefore it would be good to equip the robot with learning capabilities. Learning capabilities allow the robot to adapt to dynamic and varying environments and they allow the (non-professional) user to teach the robot new skills by demonstration (e.g. an ergotherapist that wants to teach a specific navigational skill to an intelligent wheelchair for a specific handicap). Research was done in this area on how learning techniques like neural networks and neuro-fuzzy networks can be used for the acquisition of skills like wall-following and obstacle avoidance.

Publications and reports: 2003PP051

Scientific staff: H. Van Brussel, M. Nuttin, D. Vanhooydonck, G. Vanacker

2.7.24A robotic assistant for intelligent meeting rooms


In the framework of the Ambience project, the question was put what could be the role of a mobile robotic assistant in an ambient intelligent environment. The robot Maktub was used to explore this role in an intelligent office environment or an intelligent meeting room. Maktub has exteroceptive and proprioceptive sensors, a pan-tilt-zoom camera and a wireless ethernet link. The first scenario that was considered was that of embodied telepresence: a company might want to participate to a videoconferencing meeting, but not dispose of an appropriate videoconferencing room with cameras, etc. The mobile robot could take the role of an intelligent cameraman here: it has a map of the office environment for safe navigation and sends images over the internet to the remote partners, thus creating the required telepresence. This telepresence has also successfully been shown to prince Filip and the rector of the university during a presentation of the Ambience project. In a second scenario, Maktub is a mobile assistant in the intelligent meeting room itself: it avoids obstacles, presents the interface to the intelligent meeting room and carries an extra movable camera at eye level.

Related projects: ITEA project no. 10260: “Ambience”

Publications and reports: 2003PP157, 2003PP161

Scientific staff: H. Van Brussel, M. Nuttin, D. Vanhooydonck

2.7.25Autonomous compliant motion


This research applies advanced Bayesian techniques for estimation and hypothesis testing to force-controlled robot tasks, such as assembly or deburring. The robot controller receives a nominal task specification and an indication of which geometric parameters in that task are insufficiently known and have to be identified on line. This identification is performed with (nonlinear) Kalman Filters. In many situations, the contact situation is not known exactly, and multiple candidates are available; these candidates are ordered according to probability through Bayesian hypothesis tests. We have also succeeded in constructing simple geometric models of the contact situation from scratch; in this case, the hypothesis testing algorithms are used to generate contact models starting from the simplest models that explain the data.

Related projects: GOA 99-04

Publications and reports: 2003PP074

Scientific staff: J. De Schutter, H. Bruyninckx, T. Lefebvre, K. Gadeyne, J. Rutgeerts, W. Meeussen, P. Slaets

2.7.26Nonlinear state estimation


The control literature has many examples of linearized versions of the classical Kalman Filter, but all of them are presented as ad-hoc, stand-alone approaches. This research has produced a technique to deal with a subset of nonlinear estimation problems, by transforming them into a higher-dimensional parameter space, in which the estimation problem is linear. The original estimates are recovered by, at any desired moment, applying the nonlinear constraints that connect the components of the higher-dimensional state vector. The estimation is experimentally verified on a cube-in-corner assembly.

Related projects: GOA 99-04

Publications and reports: 2001PP066, 2003PP027, 2002PP057, 2003PP100

Scientific staff: J. De Schutter, H. Bruyninckx, T. Lefebvre

2.7.27Model recognition


One of the problems faced in trying to make robots more autonomous, is the estimation of unknown parameters of the robot or its (model of the) environment. While performing a task autonomously, the robot must also be capable of choosing what model to use, eg. when performing a task that consists of several subtasks (such as assembly problems). This research uses a Bayesian approach to simultaneous model selection and state estimation for sensor-based robot tasks. The approach is illustrated with an example from autonomous compliant motion: simultaneous contact formation recognition and estimation of geometrical parameters. Previous research in this area mostly tries to solve one of the two subproblems, or treats the Contact Formation recognition problem separately, avoiding interaction between the Contact Formation detection and the geometrical parameter estimation problems. This limits the application area to task execution under small uncertainties. This research allows the robot to handle large uncertainties during the execution of its sensor-based task.

Related projects: GOA 99-04

Publications and reports: 2003PP026

Scientific staff: J. De Schutter, H. Bruyninckx, T. Lefebvre, K. Gadeyne

2.7.28Active sensing for autonomous manipulation


This part of the research focusses on planning compliant motion tasks to provide informative measurements to the estimators. A task plan consists of a sequence of contact formations (i.e., groups of contacts for which the same elements of the objects are in contact) and compliant paths in each of them (i.e., paths which do not break the established contacts). Task plans which guarantee the estimation of the inaccurately known geometrical parameters to the desired accuracy are generated by decoupling the planning of the sequence of contact formations from that of the compliant paths. The planning is simulated for a cube-in-corner assembly under different uncertainty conditions.

Related projects: GOA 99-04

Publications and reports: 2002PP090, 2002PP094

Scientific staff: J. De Schutter, H. Bruyninckx, T. Lefebvre

2.7.29Distributed realtime machine control software


This research builds a software framework for all possible realtime and non-realtime feedback control systems, with an emphasis on motion and machine control. The framework is component-based, and designed to run over a computer network. Currently, the lowest, application-independent layers are functional: integration with (realtime) Linux, hardware abstraction, basic motion control, simple CORBA interfaces. The project is available as Free Software via http://www.orocos.org.

Related projects: IST-project no. 2001-37394, “Ocean”, IST-project no. 2000-31064, “Orocos”, OMC, FMTC, GOA/JDS/99-04

Publications and reports: 2002PP097

Scientific staff: H. Bruyninckx, P. Soetens, P. Issaris, K. Gadeyne, B. Koninckx, J. Rutgeerts, W. Meeussen

Yüklə 0,74 Mb.

Dostları ilə paylaş:
1   2   3   4   5   6   7   8   9   10   ...   27




Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©muhaz.org 2024
rəhbərliyinə müraciət

gir | qeydiyyatdan keç
    Ana səhifə


yükləyin