Odsek za avtomatiko, biokibernetiko in robotiko





 

Seznam napovedanih predavanj / List of planned seminars

(povzetki so pod seznamom / abstracts under the list):

Archive:

2010-11

 

Predavanja / Seminars - sezona/season 2011/2012

------------------------------------------------------------------------------------------------------------------------------

------------------------------------------------------------------------------------------------------------------------------

21.06.2012: Robert Bevec

Autonomously learning and recognizing new objects in uncontrolled environments

The task of learning new objects is unavoidable in uncontrolled environments humanoid robots are supposed to work in. An elegant solution to this problem is a robot capable of autonomously learning properties of new objects with very little or no help from a human operator. A robot can use its manipulating abilities to acquire additional information about the scene, since it is very hard to make conclusion based solely on visual information.

In the seminar I will describe some of the challenges lying ahead for successful cognitive systems and describe our method. Our system  is capable of autonomously learning and recognizing new objects in uncontrolled environments. The only demands are that the objects have some distinctive visual features and move as a rigid body.

 

14.06.2012: Leon Žlajpah

IJS FRI server

The Kuka LWR is a state-of-the-art robotic arm aimed specifically for research institutions. Among its many features, it is possible to externally command the robot and read its state via an ethernet interface called Fast Research Interface (FRI). This functionality enables researchers to control the Kuka LWR robot using their own algorithms running in an remote computer, that send and receive data via ethernet to the FRI. The IJS FRI control server intends to provide a simple user interface to the KUKA LWR robot and hides all communication and set-up issues behind interface. In general it is only an interface, but for special need it can contain also control functionalities. Without much installation efforts, access to different standard control modes of the KUKA LWR system is provided:

  • Joint position control,
  • Cartesian impedance control,
  • Joint impedance control.

Furthermore, the IJS FRI server provides additional control modes:

  • Cartesian position control,
  • Simultanesous control of two (or more) LWR robots,
  • Redundant control,
  • Redundant control of two LWR robots as one system.

The IJS FRI server runs on a remote PC witch is connected to the KUKA Robot Controller (KRC) via an Ethernet UDP connection. In intervals of 1 to 100 milliseconds, UDP packages are periodically sent from the KRC unit to the remote host. These packages (MSR and CMD packages) contain a complete set of robot control and status data (e.g., joint positions, joint torques, drive temperatures, etc.). As the timing requirements of the FRI interface are very strong, the server interface has to run in hard real-time. Therefor, the server is impemented in Simulink xPC Target environment. The user connects to this server using another UDP connection. The communication rate can be different to the FRI communication rate (usually slower) and in some modes the real-time requirements are not so strict. The user has to send the data for the applied controllers (e.g., joint position set-points, joint stiffness set-points, etc.) and the IJS LWR server uses this data corresponding to the selected control mode. In this way, users become able to set-up own motion control and/or control strategies for the LWR robot as it is often desired at research institutions.

 

07.06.2012: Igor Mekjavić

Thermal maniqins, research and development

 

24.05.2012: Lado Lenart

Numeric processing in optimal control

Say, in the last eight years the numeric processing in optimal control (OPC) improved significantly. Direct transcription, spectral collocation and well developped nonlinear optimization with constraints and sparse matrix structure are the keywords. We will explore the base problems with the first order PDE – HJBE ( Hamilton, Jacobi, Bellman equation). Then the numerics of pseudospectral methods will be handled briefly. We finish with several OPC examples: cartesian manipulator, point obstacles, bar obstacles, multiphase program handling obstacles, double manipulator and robot ball throwing

 

17.05.2012: Luka Peternel

Robot learning, humanoid robotics and human-robot interaction

The focus of our research is mainly on three areas of robotics: robot learning, humanoid robotics and human-robot interaction. As we all know, humans are extremely adaptable and have ability to learn many tasks during the lifetime. For example at very early stage we learn how to maintain our stability which allows us to walk. Later we can learn how to operate different tools, drive cars or fly aircraft. The key to achieving this is to learn how to use the feedback information from our senses to make the appropriate body movements. In our research we wish to exploit this human sensorimotor ability to teach humanoid robots.

 

10.05.2012: Rok Vuga

Robot learning by observation

The main topic I am going to talk to about today is robot learning by observation. We all know that humans learn by observing very efficiently. This is a big challenge for robotics. We are perfectly capable of recording a human execution or demonstration and then copy it to the robot, but this rarely works, because it is too simple. When we humans observe something, we not only copy the trajectory of the demonstrator, but 1st, understand the intention of his doing, and 2nd, we are able to modify the execution to compensate for the difference between the bodies of the teacher and the demonstrator.

Our work focuses mostly on the first problem, that is understanding what the demonstrator is doing before copying trajectories. This is achieved by Semantic event chains, which we also use to segment the observation of the task. Semantic event chains have already been presented to you by our colleague Eren a few weeks ago, but I will briefly review them anyway. I will also present you parametric hidden Markov models. They are useful for many things, so I will give you an explanation of how they work and what can you do with them. In the end, I will tell you what we are doing to combine these two approaches in order to improve robot learning.

 

23.04.2012: Fares Abu-Dakka

TRAJECTORY PLANNING FOR INDUSTRIAL ROBOT USING GENETIC ALGORITHMS
A new methodology is presented, that uses genetic algorithms to solve and evaluate path and trajectory planning problems for industrial robotic systems operating in 3D environments with static obstacles.
Obstacles have been modeled using combinations of simple geometric objects (spheres, cylinders, and plans) which provide an efficient algorithm for collision avoidance.
Path planning algorithm is based on global genetic algorithm optimization techniques, which aim to minimize the sum of the distances between significant points of the robot along the path considering the restrictions to avoid collisions with obstacles. The path is composed of adjacent configurations obtained by an optimization technique using genetic algorithms, seeking to minimize a multi-objective function that involves the distance between significant points of the two adjacent configurations, and the distance from the points of the current configuration to the final one. 
The trajectory planning algorithm is similar to the one for path planning problem, but with some differences in the objective function and some details related to the conceptual difference between path and trajectory planning. The objective is to minimize the time required to move the robot from an initial configuration to a final one without colliding with obstacles, taking into consideration the limitations on the actuators. Each trajectory is constructed by means of adjacent configurations obtained through an optimization process using genetic algorithms aims to minimize a function of time required to move the robot between two adjacent configurations, the distance from the points of the current configuration to the final one, and the distance between significant points of the adjacent configurations along the trajectory. The restrictions of this algorithm may be one or a combination of the following: torque, power, and energy limitations.
An evaluation method is designed according to the problems presentation by defining individuals and genetic operators capable of providing efficient solutions to the problems. 

19.04.2012: Adam McDonnell

Combined effects of hypoxia and sustained recumbency

The aim of the present study is to investigate the combined effects of hypoxia and sustained recumbency (bedrest), on human physiological systems. The partial pressure of oxygen in the environmental gas inside future planetary habitats will be lower than in atmospheric air. Prolonged exposure to low gravity will result in deconditioning of vital physiological systems and may consequently constitute a threat to the health of the astronauts.  However, it is unknown how prolonged exposure to both reduced gravity and hypoxia will affect health. One specific aim of this study is to assess whether the previously observed inactivity-induced reduction in peripheral perfusion is augmented by hypoxia (Golja et al. 2002), and whether the decrease in perfusion causes sleep disruption (Krauchi et al. 1999). 

 

12.04.2012: Miha Deniša

Discovering New Motor Primitives in Transition Graphs

I will present a methodology for discovering new movement primitives in a database of example trajectories. The initial trajectory data, which is usually acquired from human demonstrations or by kinesthetic guiding, is clustered and organized into a binary tree, from which transition graphs at different levels of granularity are constructed. We show that new movements can be discovered by searching the transition graph, exploiting the interdependencies between the movements encoded by the graph. By connecting the results of the graph search with statistical generalization techniques, we can construct a complete representation for new movement primitives, which were not explicitly present in the original database of example trajectories.

 

06.04.2012: Bojan Nemec

Towards efficient sensorymotor learning

A truly autonomous robot should be able to generalize known actions to new situations and to autonomously refine its knowledge base. Autonomous learning of new actions is a difficult problem because the search space that has to be explored is potentially huge.

In the presentation, we will propose how to make sensory motor learning more efficient by reducing the search space based on previous experience. We will also review a learning algorithm PoWER and suggest some modifications.

 

29.03.2012: Nejc Likar

Virtual mechanism approach for dual-arm manipulation

The presentation will talk about dual-arm robots and present a novel control approach for cooperative dual-arm object manipulation.

The dual-arm system consists of two seven degree of freedom Kuka LWR robots. Our scheme has three typical features:

(1) the two arms with the object together form a new kinematic chain, where the base of the second arm is endeffector of the new robot;

(2) the object is defined as a virtual mechanism, therefore manipulating the object is accomplished by controlling the virtual mechanism;

(3) the proposed scheme is simple to use with cooperative dual-arm systems on mobile platforms.

The effectiveness of the proposed approach, is tested with different experiments.

 

08.03.2012: Tadej Petrič

Smooth Transition between Tasks on a Kinematic Control Level: form Stability to Obstacle Avoidance

Common approaches for kinematically redundant robot consist of a definition of several tasks properly combined in priority. However, in some cases the task priority needs to be changed in order to successfully perform the desired task without changing the initial strategy. I will present a method for control of kinematically redundant robots, where the focus is on a smooth, continuous transition between the primary and the secondary task. The method is based on a null-space velocity control algorithm, which is essential for achieving good behaviour of a redundant robotic system. The effectiveness of the proposed system is demonstrated on skiing robot and leg robot for assuring stability and on two Kuka LWR robots for collision avoidance.

 

27.02.2012: Andrej Gams

Performing rhythmic tasks with robots: Frequency extraction, learning and generalization of movements

V seminarju bom predstavil dvonivojski sistem posnemanja gibanja s poudarkom na izločanju frekvenc. Sistem smo aplicirali na različne naloge in različne robote. Poleg moduliranja trajektorij sistem omogoča tudi generaliziranje med naučenimi gibi.

 

In the seminar I will present a two-layered system of movement imitation. The emphasis will be on frequency extraction. We applied the system to different tasks and different robots. besides the modulation of trajectories the system allows also generalization of movements.

 

 

09.02.2012: Aleš Ude

Learning Object Representations by Manipulation

I will present my work on learning object representations by manipulation.

The basic assumption of this work is that the robot has already succeeded to grasp an object and thus gained physical control over it. By manipulating an object, the robot can focus on the relevant part of the image, thus bypassing potential pitfalls of pure bottom-up attention and segmentation.

Thich leads to a reliable extraction of representations for object recognition. Our experimental results show that the acquired data is of sufficient quality to train a classifier for viewpoint-invariant 3-D object recognition.


26.01.2012: Andrej Gams

Dopolnjevanje posnemanja gibanja z refleksivnim vodenjem stabilnosti / Augmenting Movement Imitation with Reflexive Stability Behavior

Pri posnemanju gibanja sklepov more demonstrator paziti, da robot dejansko izvede želeno nalogo, ne pa da samo izvaja enak gib v sklepih. Razlike v lastnostih robota in človeka namreč lahko pripeljejo do povsem drugačnih rezultatov. Čeprav so gibi v sklepih enaki, lahko robot recimo seže drugam kot to človek želi. Pri počepanju lahko robot pade, namesto da bi izvedel počep. Predstavil bom metodo, ki omogoča posnemanje gibanja po sklepih, hkrati pa z višjo prioriteto poskrbi, da se robot pri posnemanju giba počepa ne prevrne. Posneman gib, prenesen po sklepih, je spremenjen samo takrat, ko bi se robot prevrnil, na refleksni način. Eksperimente smo izvedli na »domačem« robotu skakalcu.

When mapping joint movement to a robot the demonstrator must make sure that the task is transferred, not only joint movement. Due to different properties of a human and a robot, joint imitation can lead to different outcomes on a robot, for example it can reach somewhere else, or tip over when imitating squatting. I will present a method which allows imitating joint movement, but with a higher priority that makes sure the robot does not tip over when imitating squatting. The imitated movement is only is only changed when the robot would tip over, in a reflexive manner. The experiments were conducted on our “home” robot jumper.

12.01.2012: Mojca Amon

Obesity and erythropoietin

Obesity is characterized by chronic systemic inflammation and associated anemia. It is well known that hypoxia plays a key role in stimulating human erythropoietin production. The study investigating the effects of 10-day continuous normobaric hypoxic exposure on hematological responses will be the topic of Department Seminar.  Presentation will focus on the individual responses of normal-weight compared to overweight population with respect to hemoglobin and erythropoietin response after used hypoxic stimulus.

References:
Cheng et al. (2011). The relationship between obesity and hypoferraemia in adults: a systemic review. Obes Rev 13: 150-161.
Gunga et al. (2007). Erythropoietin regulations in human under different environmental and experimental conditions. Respir Physiol Neurobiol 158:287-297.

 

05.01.2012: Denis Forte

Moji eksperimenti na rokah KUKA / My experiments on KUKA robot

V svojem seminarju bom predstavil metodo ucenja in zdruzevanja razlicnih primitivov gibanja v realnem casu. Pristop se pricne s knjiznico primerov trajektorij za vsakega od primitovov gibanja, ki predstavlja osnovo za statisticno posplosevanje. Pokazal bom, da je s pretvorbo trajektorij v dinamicne sisteme mogoce preklapljati med razlicnimi primitivi gibanja v realnem casu. Na eksperimentu z robotsko roko KUKA bom pokazal, da je natancnost posplosenih gibov zadostna za nalogo seganja.

Predstavil bom tudi metodo sodelovanja ucenja s posnemanje in ojacenega ucenja v simulaciji. Cilj naloge je da se robot nauci nalivanja tekocine v kozarec z minimalnim stevilom zacetnih primerov in se s casom izpopolni. Ucenje s posnemanjem poda ojacenemu ucenju zacetni priblizek, ki ga ojaceno ucenje nato prilagodi, da je naloga uspesno opravljena. Naucen gib se nato shrani v bazo znanja.

 

In my seminar I will present a methodology to learn and integrate different movement primitives in real-time. The approach starts from a library of example trajectories for each primitive movement, which serves as a basis for statistical generalization. I will show that by converting the initial trajectory data into dynamic systems, we can switch to a new movement primitive within a real-time sensory feedback loop. Experimentally we also show that the accuracy of the generalized movements is sufficient to realize tasks such as feed-forward grasping.

I will also present a methodology of cooperation of learning by imitation and reinforcement learning in simulation. The goal of the task is that robot learns how to pour a liquid in the glass with 

minimal number of initial examples and improves through time.  

Imitation learning gives initial approximation to reinforcement learning, which then adjust the pouring move to accomplish the task correctly. Learned move is then saved in the database.

 

22.12.2011: David Schiebener

Autonomous Object Learning for Humanoid Robots

This presentation will focus on the newest developments concerning the autonomous learning of household objects for later recognition.

 

15.12.2011: Igor Kovač

Affordable automation and robot aided reconfiguration 

Unpredictable markets, increased product customization and quest for competitive advantages are challenges facing manufacturing enterprises now and in the foreseeable future. Frequent changes in product, production technologies and manufacturing system are evident today with their significant implementation costs. In the presentation the idea of affordable automation with robot aided reconfiguration will be presented.

 

08.12.2011: Shawnda Morrison

COLD PHYSIOLOGY AND CIVD IN ALPINISTS (presentation)

Freezing cold injuries can be categorised along a spectrum of injury, from the least severe "frostnip", to more serious "trenchfoot" and finally to extreme cold injury "frostbite".  When the hands or feet become very cold, a phenomenon called "cold-induced vasodilation" or, CIVD is often observed.  This involves a large inflow of blood to the periphery and is thought by some researchers to provide a protective effect against cold injury for those tissues.  However, very little is known about the mechanisms of this CIVD response, how it differs between people and also between hands and feet.  This talk will discuss briefly the physiology behind freezing cold injuries, with examples provided by military, industrial and sporting contexts.  The discussion will conclude with new research which has been recently collected in our lab investigating elite Slovenian alpinists, some of whom suffered freezing cold injuries to their hands or feet requiring amputation, compared to alpinists who suffered no injuries. Are there differences in how the alpinists hands and feet respond after being exposed to a cold stress? Are there differences between the injured and non-injured fingers and toes within the injured alpinist population?  These data will contribute to the understanding of how people recover after a freezing cold injury, and whether those people have a higher incidence of being re-injured with repeated exposure to the cold.

 

25.11.2011: Mitja Babič

ON THE DESIGN OF DRIVER ELECTRONICS FOR DIELECTRIC ELASTOMER APPLICATIONS
In the seminar I will present the problematic related to the design of driver electronic for dielectric elastomer actuators. DC high voltages in the range of kilovolts and DC currents in the range of milliamperes are necessary to properly energize dielectric elastomer actuators. Two novel activation strategies will be presented. The first electronic driver derives from the flyback converter topology and it is able of delivering to the dielectric elastomer actuator middle-frequency, current-pulse trains dependent on the duty-cycle value. The second driving circuit features two high-voltage opto-couplers in a high-voltage push-pull configuration. The push-pull configuration is useful for driving strictly capacitive loads, therefore also dielectric elastomer actuators, and exhibits high efficiency. Finally, the agonist-antagonist conically-shaped dielectric elastomer linear actuator with the appropriate driver and controller is presented. Experimental results are provided to validate the two drivers.

 

10.11.2011: Anton Ružič

System design of a special purpose manipulator

I will present some activities carried out as part of an overall system design of a special purpose robot manipulator, aimed for automation of a cleaning task.

Design activities included identification of the target task requirements, overall design choices, synthesis of the cleaning technology and synthesis of the manipulator structure.

 

20.10.2011: Tadej Debevec

Uporaba normobarične hipoksije in hiperoksije za izboljšanje sposobnosti na nižini in/ali višini

The use of normobaric hypoxia and hyperoxia for the enhancement of sea level and/or altitude exercise performance


Seminar bo ponovitev zagovora moje doktorske disertacije pretekli petek. Predstavil bom štiri raziskave, ki sem jih vključil v doktorsko delo. Seminar bo v angleškem jeziku.
The seminar will be a "replay" of the dissertation defense that took place last Friday. I will present the four studies included in my dissertation.