Dirichlet process mixtures and its application to multi-target tracking
Time: 13:15-14:00. Place: Glashuset.
Dirichlet process mixtures (DPM) are a popular class of models widely used in the estimation of probability densities and clustering. In this talk, a tutorial on Dirichlet process will be given. Time varying Dirichlet process mixtures which are proposed to estimate the densities evolving over time will also be introduced. The application of the model to the problem of multi-target tracking via sequential Monte Carlo methods will be demonstrated.
Trajectory prediction and Wind estimation using Multiple aircraft
Time: 13:15-14:00. Place: Glashuset.
Aircraft trajectory prediction is of paramount importance for the safety and capacity of the airspace. It provides air traffic controllers with an instrument to assist in their search for efficient and collision free flight paths. However, meteorological predictions - mainly wind velocity - have inaccuracies that may result in large deviations from the expected path of an aircraft. Fortunately, the position of the aircraft, during its flight, incorporates the effect of the wind on its dynamics. Additionally, wind uncertainty (or else the error of weather forecasts) is not entirely random but is correlated in time and space. We can employ position measurements (from a ground based radar) for multiple aircraft flying in the same airspace, in order to predict their trajectory and estimate the wind in space-time. Finally, by using direct wind measurements, from the aircraft flying in a given sector, and transmit those to the ground, we can further improve our results.
A Nonlinear Observer for Rigid Body Attitude Estimation on SO(3) using Inertial Measurements and Vector Observations
Time: 13:15-14:00. Place: Glashuset.
Attitude estimation is a classical problem with a rich and fascinating history still holding a forefront position as the subject of intensive research. Attitude observers based only on the rotation kinematics are of special interest for applications using inertial sensors and attitude aiding devices, with application to autonomous vehicles. In this seminar, the problem of attitude estimation is motivated by presenting some autonomous robotic platforms under development at DSOR, and a nonlinear attitude observer defined on SO(3), based on inertial and vector measurements, is proposed. The observer has an almost globally asymptotically stable equilibrium point at the origin and convergences exponentially fast for any initial condition in an explicit region. The feedback terms, derived constructively using the Lyapunov's stability theory, are an explicit function of the available vector observations and biased angular velocity measurements. Topological limitations for achieving global stabilization on the SO(3) manifold provide important guidelines and are discussed and illustrated in the present observer. Exponential convergence and convergence bounds are obtained, resorting to the recent results for parameterized linear time-varying (LTV) systems. The properties of the observer are illustrated in simulation for inertial sensor characteristics and initial alignment errors commonly found in practical setups.
Autonomi i bilar och Fordonsframdrivning
Time: 13:15-14:00. Place: Glashuset.
Autonomi i bilar — Darpa Urban Challenge 2007 visade att tekniken finns för att få vanliga personbilar att köra autonomt i stadstrafik. Steget till serieproducerade bilar är dock stort. Föredraget ger några exempel på säkerhetshöjande funktioner i bilar, hur dessa relaterar till forskningsfronten, samt hur olika säkerheter påverkar den slutliga användarnyttan.
Autonomi i bilar och Fordonsframdrivning
Time: 13:15-14:00. Place: Glashuset.
Fordonsframdrivning — Mot bakgrund av den livliga diskussionen om klimateffekter av CO2-utsläpp, den ändliga oljereserven, höjda priser på fossila bränslen mm, har nya principer för fordonsframdrivning kommit i fokus, och olika former av hybridfordon är exempel på detta. Svensk fordonsindustri satsar för att vara med i utvecklingen, och nyligen har industrin i samverkan med akademin och med stöd av Energimyndigheten startat Svensk Hybridfordonscentrum för att ge denna utveckling en forskningsbas. Föredraget belyser några av de tekniska utmaningarna och några av de centrala forskningsfrågorna inom området.
Part 1 : Rank reduction and volume minimization approach to state-space subspace system identification. Part 2: Tensor computations and tensor approximation.
Time: 13:15-14:00. Place: Glashuset.
Abstract, part 1: This will be a short presentation of the content of an article with the same title. We will motivate the need for generalizing the determinant minimization criterion, which is encountered in reduced rank regression problems. We will also present the neat solution and make some geometric interpretations with respect to signal properties.
Abstract, part 2: The second part will be an introduction on tensors and low (multilinear) rank tensor approximation. Tensors in this context are multidimensional data arrays. The objective function of the approximation problem is defined on a product of Grassmann manifolds.
We will outline the key ideas of optimization methods (Newton and quasi-Newton algorithms) that solve this problem.
Neuro Mechanical Networks
Time: 13:15-14:00. Place: Glashuset.
The term Neuro Mechanical Network (NMN) refers to a system in which the basic structural unit is a multi-functional element. Whereas in most traditional systems the basic structural units provide perhaps one or two functions each, e.g. stiffness, the multi-functional element can also provide mechanical actuation, sensing, signal processing and other functions. The biological inspiration is obvious, one may for instance think of muscle cells in the heart.
Apart from biology, the theoretical framework takes inspiration and knowledge from fields such as smart structures, neural networks, and structural optimization. The present formulation consists of a coupled system of equations describing a mechanical truss with a neural network superimposed onto it. Methods from structural optimization are employed in order to create structures with adaptive stiffness. Possible applications include fixtures which can adapt to various load conditions, and in general situations where stiffness and low weight are conflicting demands.
In this presentation we will give some background to the project, present the theoretical framework, and finally show some numerical examples.
Synchronization and Coordination of Multi-Agent Systems
Time: 10:15-11:00. Place: Glashuset.
In this talk we will address the problem of synchronization and coordination of multi-agent systems from the perspective of passivity based control. Such systems arise in a number of emerging application areas, such as sensor networks,autonomous flying and swimming vehicles, networked communication and control systems, and computer networks. We will discuss how synchronization naturally arises in both natural and engineered multi-agent systems and how feedback in such systems leads to interesting emergent behaviors. Examples will be given in bilateral teleoperation and in attitude synchronization of multiple robots.
Model-Based Vehicle Dynamics Control for Active Safety
Time: 13:15-14:00. Place: Glashuset.
The functionality of modern automotive vehicles is becoming increasingly dependent on control systems. Active safety is an area in which control systems play a pivotal role. Currently, rule-based control algorithms are widespread throughout the automotive industry. In order to improve performance and reduce development time, model-based methods may be employed. The primary contribution of this thesis is the development of a vehicle dynamics controller for rollover mitigation. A central part of this work has been the investigation of control allocation methods, which are used to transform high-level controller commands to actuator inputs in the presence of numerous constraints. Quadratic programming is used to solve a static optimization problem in each sample. An investigation of the numerical methods used to solve such problems was carried out, leading to the development of a modified active set algorithm. Vehicle dynamics control systems typically require input from a number of supporting systems, including observers and estimation algorithms. A key parameter for virtually all VDC systems is the friction coefficient. Model-based friction estimation based on the physically-derived brush model is investigated.
Information, Fusion and Control in Autonomous Networks
Time: 13:15-14:00. Place: Visionen.
Information provides a quantitative metric for describing the value of individual systems components in autonomous systems tasks such as tracking, mapping and navigation, search and exploration; tasks in which the objective is information gain in some form. An information model is an abstraction of system capabilities in an anonymous form which allows a priori reasoning on the system itself. By construction, information measures have properties of composability and additivity and thus provides a natural means of modelling and describing large scale systems of systems.
This talk will begin by describing how information measures arise naturally in autonomous tracking, mapping and navigation, search and exploration tasks. It is then demonstrated that the performance of individual sensors and platforms can be modelled using these information measures and that system-level performance metrics can be computed. These ideas are illustrated in a series of tasks involving mixed air and ground autonomous systems. These include flight-tests of cooperative UAVs engaged in tracking and navigation tasks, mixed UAV, ground vehicles and human operatives, engaged in mapping and picture compilation operations, and operations involving multi-platform search in constrained environments. In each, it is shown how information provides both a performance metric and design objective underpinning large-scale systems of systems operation. Finally current work in using these ideas to design and manage large-scale autonomous systems in applications such as cargo handling and mining will be described.
Estimation and Detection with Applications to Navigation
Time: 10:15-13:00. Place: Visionen.
The ability to navigate in an unknown environment is an enabler for truly autonomous systems. Such a system must be aware of its relative position to the surroundings using sensor measurements. It is instrumental that these measurements are monitored for disturbances and faults. Having correct measurements, the challenging problem for a robot is to estimate its own position and simultaneously build a map of the environment. This problem is referred to as the Simultaneous Localization and Mapping (SLAM) problem. This thesis studies several topics related to SLAM, on-board sensor processing, exploration and disturbance detection.
The particle filter (PF) solution to the SLAM problem is commonly referred to as FastSLAM and has been used extensively for ground robot applications. Having more complex vehicle models using for example flying robots extends the state dimension of the vehicle model and makes the existing solution computationally infeasible. The factorization of the problem made in this thesis allows for a computationally tractable solution.
Disturbance detection for magnetometers and detection of spurious features in image sensors must be done before these sensor measurements can be used for estimation. Disturbance detection based on comparing a batch of data with a model of the system using the generalized likelihood ratio test is considered. There are two approaches to this problem. One is based on the traditional parity space method, where the influence of the initial state is removed by projection, and the other on combining prior information with data in the batch. An efficient parameterization of incipient faults is given which is shown to improve the results considerably.
Another common situation in robotics is to have different sampling rates of the sensors. More complex sensors such as cameras often have slower update rate than accelerometers and gyroscopes. An algorithm for this situation is derived for a class of models with linear Gaussian dynamic model and sensors with different sampling rates, one slow with a nonlinear and/or non-Gaussian measurement relation and one fast with a linear Gaussian measurement relation. For this case, the Kalman filter is used to process the information from the fast sensor and the information from the slow sensor is processed using the PF. The problem formulation covers the important special case of fast dynamics and one slow sensor, which appears in many navigation and tracking problems.
Vision based target tracking is another important estimation problem in robotics. Distributed exploration with multi-aircraft flight experiments has demonstrated localization of a stationary target with estimate covariance on the order of meters. Grid-based estimation as well as the PF have been examined.
Reglerteknikens framtid
Time: 14:15-15:00. Place: Visionen.
Även om återkoppling användes redan i antiken så är det rimligt att påstå att reglertekniken är drygt 60 år gammal. Föredraget ger några reflektioner över den dynamiska utvecklingen av ämnet. Först presenteras en kort historik. Reglerteknik, det första systemämnet, var ett paradigmskifte i ingenjörsutbildningen. Det var svårt att passa in ämnet i en ingenjörsutbildning som var strikt uppdelad i mekanik, elektroteknik, kemi och väg-och-vatten. Detta är ett problem som ämnet fortfarande brottas med. I Sverige har vi varit ganska lyckosamma eftersom vi på flera håll lyckats skapa centrala grupper med ett totalansvar för all utbildning och forskning. Några centrala ideer i reglerteknikens utveckling presenteras tillsammans med några axplock av en mångfald tillämpningar. Den andra fasen i utvecklingen karakteriserades av en mycket stark teoriutveckling och subspecialisering, men helhetssynen på ämnet förlorades. Enligt min uppfattning står vi nu inför en tredje fas. Föredraget avslutas med några spekulationer om den framtida utvecklingen som starkt påverkas av hur vi själva agerar. Bland annat diskuteras interaktion med andra närliggande ämnen liksom frågor som relaterar till teori, praktik, forskning och utbildning.
Cykelåkning i ett reglertekniskt perspektiv
Time: 09:15-10:00. Place: Visionen.
En cykel är inte bara ett bra och miljövänligt fortskaffningsmedel, den lämpar sig också utmärkt för att belysa matematiskt modellbygge och många reglertekniska begrepp. Många intressanta och tankeväckande experiment kan också utföras med enkla medel. I seminariet presenteras flera olika modeller av cyklar. Vi diskuterar frågor som självstabilisering och styrning. Svårigheter med att hantera cyklar med bakhjulsstyrning presenteras ingående. Det är en bra illustration av att det finns system som är svåra att styra. För att förklara detta utnyttjar vi fundamentala begränsningar orsakade av poler och nollställen i högra halvplanet. Exempel på enkla experiment som kan utföras med cyklar presenteras samt erfarenheter från att använda cyklar i reglertekniska kurser. Slutligen beskrivs några mycket intressanta experiment som utförts av Prof Richard Klein för att lära handikappade barn att cykla.
Treating populations and landscapes as signals
Time: 13:15-14:00. Place: Glashuset.
In many applied biological problems one faces a system consisting of entities, as populations, farms or habitats, spread in space. The systems dynamics is then dependent on both interactions between the entities and their distribution in space. Moreover the dynamics is often also mixed with stochastic components. In my research group we have started to treat these systems as signals. To test the dynamics we generate \u2018virtual\u2019 systems by modeling the spatial component and the stochasticity using Fourier transform and spectral analysis. We also have begun to analyze the dynamics by spectral analysis. These methods are promising yet not part of the toolbox of theoretical biologists. The aim of the seminar is that it can be an introduction to fruitful collaborations by presenting different sets of problems and projects.
Suggested prereading:
Keitt, T. H. 2000.
Spectral representation of neutral landscapes.
Landscape Ecology 15:479-493
Greenman J. V, Benton T. G. 2005. The frequency spectrum of structured
discrete time population models; its properties and their ecological implications. OIKOS. 110:369-389
Linstrom T, Hakansson N, Westerberg L, Wennergren U. 2008
Splitting the tail of the displacement kernel shows the unimportance of kurtosis. Ecology. 89:1784-1790
available for download at:
http://people.ifm.liu.se/unwen/isy_seminar
Predictive Approaches for Vehicle Dynamics Control
Time: 13:15-14:00. Place: Glashuset.
Automatically reproducing complex evasive manoeuvres, which a skillful driver would execute in extreme limit handling situations (i.e., high speed on slippery surfaces), requires the solution of a complex multivariable control problem where strong nonlinearities and model uncertainties significantly complicate the control design. Model Predictive Control (MPC) represents an effective design procedure for systematically developing active safety systems and controlling the vehicle at the limit of handling capabilities by coordinating multiple actuators (such as steering, braking, engine torque, active differential and active suspensions). In this talk we will present our most recent results on the application of low complexity nonlinear Model Predictive Control approaches for designing real-time Active Front Steering (AFS) and combined AFS and braking control systems. The underlying design methodology, an in-vehicle experimental setup, as well as simulation and experimental results of a controlled vehicle at high speed on icy and snowy roads, will be presented.
How to control Vattenfall
Time: 13:00-14:00. Place: Glashuset.
Is modern control theory of any use in power plants that has basically functioned in the same way for about a hundred years? Yes, if you want stable operation, increased efficiency and decreased emissions. Mikael will present how modern control theory has been introduced in Vattenfall, how Vattenfall works with these tools today and also dig into some recently performed projects.
Regression on Manifolds with Implications for System Identification
Time: 10:15-12:00. Place: Visionen.
The trend today is to use many inexpensive sensors instead of a few expensive ones, since the same accuracy can generally be obtained by fusing several dependent measurements. It also follows that the robustness against failing sensors is improved. As a result, the need for high-dimensional regression techniques is increasing.
As measurements are dependent, the regressors will be constrained to some manifold. There is then a representation of the regressors, of the same dimension as the manifold, containing all predictive information. Since the manifold is commonly unknown, this representation has to be estimated using data. For this, manifold learning can be utilized. Having found a representation of the manifold constrained regressors, this low-dimensional representation can be used in an ordinary regression algorithm to find a prediction of the output. This has further been developed in the Weight Determination by Manifold Regularization (WDMR) approach.
In most regression problems, prior information can improve prediction results. This is also true for high-dimensional regression problems. Research to include physical prior knowledge in high-dimensional regression i.e., gray-box high-dimensional regression, has been rather limited, however. We explore the possibilities to include prior knowledge in high-dimensional manifold constrained regression by the means of regularization. The result will be called gray-box WDMR. In gray-box WDMR we have the possibility to restrict ourselves to predictions which are physically plausible. This is done by incorporating dynamical models for how the regressors evolve on the manifold.
The elements of Weather Forecasting - from observations through physical principles to computational modelling
Time: 13:15-14:00. Place: Glashuset.
The ability to forecast the weather has been developed over the last 150 years through observations of large scale pressure systems and gradually extended knowledge of the flow of the atmosphere. The governing equations of the flow and thermodynamics are basically the same as in many other geophysical applications, ocean, plasma, etc. but the ability to exploit them in models has been, and still is, limited by the capacity of large scale computing.
The principles behind forecast models will be described and some different ways of implementing them. The weather forecasting is an initial value problem, and the most important aspect for the success is an accurate initial state. This is achieved by means of data assimilation, where observations and model values are combined in more or less optimal ways in a quasi-continuous process.
There are many sources of errors in the process: the atmosphere is never sufficiently well observed and only within a certain accuracy, the data assimilation and model solutions involve approximations and, most importantly, the governing equations are non-linear so any small errors may grow rapidly if conditions are right. There are ways of quantifying this by running ensembles of perturbed forecasts and estimate probabilities of weather events rather than just a precise value (which is unlikely to be right).