Department for Automation, Biocybernetics and Robotics

Seznam napovedanih predavanj / List of planned seminars

(povzetki so pod seznamom / abstracts under the list)


2013 / 2014
sezona 2011/2012
sezona 2010/2011

Ni napovedanih predavanj. No planned seminars.
Pretekla predavanja / Past seminars - 2013 / 2014
19.12.2014: Bojan Nemec
Stabilnost ILC algoritmov v kontaktu z okoljem / Stability of ILC interacting with environment
Na seminarju bom predstavil analizo stabilnosti regulatorja za učenje ponavljajočih se gibanj v frekvenčnem prostoru. Analiziral bom vpliv filtriranja kompenzacijskega signala ter vpliv integralskega člena pri nalogah, ki vključujejo kontakte z okolico ter predlagal regulator, ki kompenzira te vplive.

Beside time lifted approach, the stability of an ILC can be analyzed also in frequency domain. Frequency domain analyses will be used to analyze the effect of filtering of the feed-forward signal generated by ILC and the effect of the integral term in learning algorithms interacting with the environment. Small modifications will be proposed in order to compensate the effect of the filtering and the effect of the integral loop in the control algorithm.

12.12.2014: Tadej Petrič
Sprotno učenje specifične dinamike naloge / Online Learning of Task-Specific Dynamics

In this seminar I will address the problem of accurate trajectory tracking while ensuring compliant robotic behavior without the use of experts and dynamical models. The proposed approach uses programming by demonstration to learn new task-related compliant movements composed of 1) position and 2) torque trajectories. The movements are encoded as a Compliant Movement Primitives, i.e., the position trajectory in the form of a dynamic movement primitive (DMP) and the torque trajectory as a linear combination of radial basis functions. By using the proposed control framework, the robot successfully accomplishes new tasks while being compliant and thus safer for humans sharing its workspace. The proposed approach was evaluated on a Kuka LWR robot.

12.12.2014: Barry Ridge
Učenje osnovnih funkcionalnih lastnosti objektov v robotskem sistemu / Learning Basic Object Affordances in a Robotic System

One of the fundamental enabling mechanisms of human and animal intelligence, and equally, one of the great challenges of modern day autonomous robotics is the ability to perceive and exploit environmental affordances. To recognise how you can interact with objects in the world, that is to recognise what they afford you, is to speak the language of cause and effect, and as with most languages, practice is one of the most important paths to understanding. This is clear from early childhood development. Through countless hours of motor babbling, children gain a wealth of experience from basic interactions with the world around them, and from there they are able to learn basic affordances and gradually more complex ones. Implementing such affordance learning capabilities in a robot, however, is no trivial matter. This is an inherently multi-disciplinary challenge, drawing on such fields as autonomous robotics, computer vision, machine learning, artificial intelligence, psychology, neuroscience, and others. In this thesis, we attempt to study the problem of affordance learning by embracing its multi-disciplinary nature. We use a real robotic system to perform experiments using household objects. Camera systems record images and video of these interactions from which computer vision algorithms extract interesting features. These features are used as data for a machine learning algorithm that was inspired in part by ideas from psychology and neuroscience. The learning algorithm is perhaps the main focal point of the work presented here. It is a self-supervised multi-view online learner that dynamically forms categories in one data view, or sensory modality, that are used to drive supervised learning in another. While useful in and of itself, the self-supervised learner can potentially benefit from certain augmentations, particularly over shorter training periods. To this end, we also propose two novel feature relevance determination methods that can be applied to the self-supervised learner. With regard to robotic experiments, we make use of two different robotic setups, each of which involves a robot arm operating in an experimental environment with a flat table surface, with camera systems pointing at the scene. Objects placed in the environment can be manipulated, generally pushed, by the arm, and the camera systems can record image and video data of the interaction. One of the camera systems in one of the setups is a stereo camera, and another in the other setup is an RGB-D sensor, thus allowing for the extraction of range data and 3-D point cloud data. In the thesis, we describe computer vision algorithms for extracting both salient object features from the static images and point cloud data, and effect features from the video data of the object in motion. A series of experiments are described that evaluate the proposed feature relevance algorithms, the self-supervised multi-view learning algorithm, and the application of these to real-world object push affordance learning problems using the robotic setups. Some surprising results emerge from these experiments and as well as those, under the conditions we present, our framework is shown to be able to autonomously discover object affordance categories in data, predict the affordance categories of novel objects and determine the most relevant object properties for discriminating between those categories.

14.11.2014: Igor Mekjavić
/ In vivo retinal images for a non-invasive analysis of the microcirculation during hypoxia and unloading/inactivity

Astronauts will make use of planetary habitats during long-term manned space missions. Current habitat concepts foresee that the astronaut will have to deal with reduced gravity and hypoxic hypobaric environments. Exposure to these environmental conditions causes stress on several physiological systems. These effects need to be documented for risk assessment and for countermeasure development. We investigated cardiovascular effects during habitat simulation using non-invasive digital retinal fundus images. Retinal microvasculature shows homology with the microvascular beds found in the heart, lungs, and brain. It is proposed that retinal imaging is a convenient and relevant tool for monitoring systemic changes in the microcirculation. Dynamic microvascular changes were documented in subjects participating in a 21-day simulation of a planetary habitat that was maintained at an ambient equivalent of 4000 m altitude. Fourteen healthy male subjects participated in the 21-day cross-over experimental design to assess the effects of unloading/inactivity and hypoxia. Baseline retinal photographs were taken for both eyes. Repeated retinal photographs were taken during 21-day study periods and after recovery. Retinal vessel diameters were measured using a computer-based program and the results were summarized as central retinal arteriolar and venular equivalents (CRAE and CRVE, respectively). CRAE and CRVE have been shown to be sensitive parameters that reflect the impact of modifiable lifestyle and environmental cardiovascular disease risk factors on the microvasculature. We observed fast changes in the retinal microvasculature, which manifest themselves almost immediately when the volunteers were exposed to the planetary habitat test conditions. During normoxic bedrest, CRAE narrowed significantly and in a progressive way. The maximum effect was a 7-µm decrease. CRVE values were not affected. The hypoxic environment caused a significant increase in the CRAE of about 8 µm and a venous pooling, reflected in a 15-µm decrease in CRVE. No difference was observed between the ambulatory and unloading condition in hypoxia. CRAE and CRVE values returned to baseline values 24 h after the ending of the study in the three test conditions. We propose retinal imaging as a convenient and easy to use tool to investigate the impact planetary habitat conditions on the microvasculature.

17.10.2014: Leon Žlajpah
Uporaba simulacijskih orodij za analizo gibanja človeka / Analysis of Human Motion using Simulation Tools

To identify the biomechanical concepts that explain how the movements are made and how they can be modified it is necessary to analyze the human motion. We have conducted human behavioural studies for understanding human use of external contacts, where a person was standing on a platform. We measured the corrective movements of a person influenced by the perturbation caused by the platform. The measured data included the 3D position of markers attached to legs, body and arms, interaction forces in the handle, and ground reaction forces. We present the methodology used to obtain the joint motion and torques. First, we determined the joint angles, velocities, and accelerations using the markers data. Next, we computed the torques necessary for the measured motion, where the inverse dynamics of human body has to be solved using the calculated joint motion and measured external forces. To derive the model we used MATLAB/SimMechanics. Namely, SimMechanics provides the simulation mode where the motion is given as the input and the joint torques are then automatically calculated. The body model parameters needed in the simulation were determined by using standardized human body parameters (we present an application which we developed to determine the body model parameters). We gibve also some simulation results how the joint stiffness and different control strategies influence the corrective movements. Finally, the model generation im MATLAB/Simulink environment and in MuJoCo are described.

17.10.2014: Roman Hribar
Stabilno posnemanje gibanja celotnega telesa z robotom HOAP in Kinect-om V2 / Stable body motion imitation on the HOAP, Kinect V2

I will talk about on-line full body imitation with a humanoid robot, based on prioritized task control. The method allows for real-time simultaneous control of balance and transfer of motion from a human demonstrator to the robot. Furthermore, self-collision avoidance is included in the control loop. I will give a detailed description of all steps of the algorithm. The method was implemented in SL simulation software as well as on humanoid robot HOAP, while human motion was capture using Kinect camera. At the end I will tell something about Kinect tracking and how to use Kinect V2 Server aplication.

24.06.2014: Aleš Ude
Orientacija in dinamicni generatorji gibov v kartezičnem prostoru / Orientation in Cartesian Space Dynamic Movement Primitives

Dynamic movement primitives (DMPs) were pro- posed as an efficient way for learning and control of complex robot behaviors. They can be used to represent point-to-point and periodic movements and can be applied in Cartesian or in joint space. One problem that arises when DMPs are used to define control policies in Cartesian space is that there exists no minimal, singularity-free representation of orientation. In this paper we show how dynamic movement primitives can be defined for non minimal, singularity free representations of orientation, such as rotation matrices and quaternions. All of the advantages of DMPs, including ease of learning, the ability to include coupling terms, and scale and temporal invariance, can be adopted in our formulation. We have also proposed a new phase stopping mechanism to ensure full movement reproduction in case of perturbations.

17.06.2014: Jan Babič
Posturalna kontrola z uporabo nosilnih kontaktov roke / Supportive hand contacts in reactive postural control

There are many everyday situations in which a supportive hand contact is required for an individual to counteract various postural perturbations. In the seminar I will discuss how we emulated situations when balance of an individual is challenged and examined functional role of supportive hand contact where balance of an individual was perturbed by translational perturbations of the support surface. I will elaborate on the effects of handle location, perturbation direction and perturbation intensity on the postural control and the forces generated in the handle. I will conclude the seminar with a preliminary robotic study where we aim to employ our biological findings and equip humanoid robots with the ability to use supportive hand contacts in the same manner as humans do.

12.06.2014: Aljaž Kramberger
Primerjava strategij in konceptov učenja trajektorij s človeško demonstracijo za peg-in-hole akcijo. / A comparison of exception strategies and force based learning-by-demonstration concepts for peg-in-hole action.
V sklopu seminarja bo predstavljeno moje dosedanje raziskovalno delo. V uvodu bodo na kratko predstavljeni rezultati magistrskega dela. Predstavitev raziskovalnega dela bo temeljila predvsem na raziskovalnem delu narejenem v okviru Intellact projekta. Nanaša se predvsem na izboljševanje natančnosti in zanesljivosti peg-in-hole akcije. Predstavljeni bodo rezultati implementiranih strategij baziranih na zajetih silah in pozicijah. Prav tako se bom osredotočil na zadnje raziskovalno delo. Ki se je nanašalo, na primerjavo treh različnih konceptov zajemanja trajektorij s človeško demonstracijo. Prestavljeni bodo rezultati raziskave in načrti za prihodnje delo.

For this seminar I will present my research work. In the introduction I will briefly present the results of my master’s thesis. The main research work will be based primarily on research done in the context of Intellact project. It refers mainly to improve the accuracy and reliability of the peg-in-hole action. For that reason force and position based evaluation strategies were implemented and tested. Furthermore I will present the results from the last research. The research was mainly a comparison of three different learning-by-demonstration concepts. In conclusion the future research plans will be presented.

29.05.2014: Anton Ružić
Značilnosti zajemanja in obdelave posnetkov v proizvodnem okolju z visokimi temperaturami / Characteristics of visual information acquisition and processing in a high temperature production environment

24.04.2014: Rok Vuga
Verjetnostni semantični modeli za predstavitev in pridobivanje manipulacijskih nalog / Probabilistic semantic models for manipulation action representation and extraction

In this seminar I will present a novel framework to represent manipulation actions and their recognition and segmentation based on semantic object relations. We propose to probabilistically model the occurrence of semantic events over the duration of an action. The resulting models are highly action dependent and can be used to provide probabilistic similarity scores for newly observed action sequences. Furthermore, we show how to couple the presented framework, which encodes higher level semantics of an action, with other available lower level data, such as motion trajectories. We demonstrate the applicability of the approach on a problem of top down action extraction from observation.

17.04.2014: Nejc Likar
Identifikacija kontakta na podlagi meritve zunanjih navorov / Seminar room at 10:00 - External Joint Torque based Estimation of Contact Information
Seminar bo predstavil metodo identifikacije kontakta, oziroma kontakne sile in lokacije kontakta. Identifikacijski algoritem temelji na meritvi navorov, kateri so posledica zunanje sile. Navor, kot posledica zunanje sile lahko povzročijo različne kontaktne situacije (različne sile, lokacije), zato je sama identifikacija kontaktne sile in lokacije kontakta, kompleksen nelinearen problem. Z uporabo nekaterih poenostavitev je mogoče zmanjšati kompleksnost problema in ga rešiti z predlagano metodo, ki temelji na nelinearni omejeni optimizaciji. Rezultati simulacij in eksperimentov kažejo možnost identifikacije kontakta na celotni strukturi robota, s pomočjo meritve zunanjih navorov, brez uporabe dodatnih senzorjev.

In the seminar, a method for estimating the contact information i.e. the contact force and the contact location, is presented. The estimation algorithm is based on the measurement of the part of joint torques caused by the external force. The torque produced by the contact force, may arise from a wide variety of contact situations, which makes this a complex nonlinear problem. Using some assumptions, the complexity of the problem is reduced and solved by the proposed method, which is based on nonlinear constrained optimization. The simulation and experimental results show that the proposed approach allows estimation of contact forces by using only joint torque sensors without any additional external sensory systems for detection of contacts along the robot body structure.

03.04.2014: Robert Bevec
Sejna soba, 13:00 - Učenje predstavitev objektov s pomočjo robotskih manipulacij / The seminar room, 13:00 - Learning of Object Representations Through Robotic Manipulation
Avtonomni robotski sistemi, ki delujejo v naravnih okoljih, se neprestano srečujejo z novimi objekti. Za takšne robote je pomembno, da znajo pridobivati nove predstavitve predmetov, ki jih lahko kasneje uporabijo pri prepoznavanju in manipulaciji objektov. Predlagamo metodo robustnega učenja predstavitev novih predmetov in uspešnega razpoznavanja že znanih predmetov za humanoidne robote. Metoda mora delovati v nestrukturiranem okolju, kjer robot avtonomno, ali v sodelovanju s človekom, zgradi predstavitve neznanih predmetov, ko o objektih nima na voljo nobenega predznanja. Ker je segmentacija izključno na osnovi vida težka, bomo uporabili robotove manipulacijske sposobnosti, da v prizoru pridobimo dodatne informacije za učenje ali razpoznavanje objektov. V umetnem sistemu želimo imitirati človeško sposobnost zaznavanja predmetov v širokem vidnem polju, hkrati pa imeti visoko gostoto točk za natančno analizo predmeta. To storimo z uporabo dveh kamer v vsakem očesu, ki imata različni goriščni razdalji. Širokokotni kameri predstavljata periferni vid in pokrijeta veliko področje, ozkokotni kameri pa prikazujeta manjši del prizora v visoki ločljivosti. Znanje o objektih moramo predstaviti v obliki, ki omogoča uspešno razpoznavanje velike zbirke objektov. Seminar bo v slovenščini.

The seminar will be conducted in Slovene, since it's a practice run for my thesis proposal presentation coming up. The presentation will last 20 minutes and every one is still welcome to join. Below you can find the summary in Slovene and English. Autonomous robots that operate in unstructured environments must be able to seamlessly expand their knowledge base. To identify and manipulate previously unknown objects, a robot should be able to acquire new object knowledge. We propose a robust method of object learning and recognition for humanoid robots. The method must work in unstructured environments, where the robot autonomously, or aided by a human teacher, builds object representations, when no prior information about the objects or the environment is available. Since perception of objects from single images has proved very difficult, we take advantage of the robot's manipulation capabilities to induce changes in the scene and extract additional information from the scene to make the task more feasible. We also want the artificial system to mimic the human capability of detecting objects in the peripheral view, while having a high acuity in the foveal view. A simple imitation of this biological system can be accomplished using two cameras per eye with different focal lengths. This enables the robot to obtain high resolution images of objects with one camera pair, while still being able to perceive events in the wide angle peripheral view with the other pair. Knowledge about objects must be described with representations suitable for reliable recognition in a large database.

06.03.2014: Adam McDonnell
Simulacija planetarnih habitatov: učinki na psihološko stanje / Ground based facilities, bed rest, planetary habitats and psychology

This seminar will address measuring the affects space travel has on humans and how to measure this while still on earth. Travelling to space has been and is currently limited a very fortunate few which renders it difficult to carry out scientific experiments and draw real conclusions. The logistics, cost and time schedule makes it nigh on impossible to investigate all pertinent questions during any single sojourn. The most straightforward way to address this issue is to create facilities on earth which replicate to a certain extent the microgravity and confinement of space travel. In doing so, a larger number of subjects can be studied in a controlled environment who are not charged with the running of a space station or with piloting a space craft, therefore their time can be made available to experimentation. One such facility has in some measure been fashioned from the Olympic Sports Centre, Planica. During the course of my PhD studies we have investigated the combined affects of microgravity (bed rest) and hypoxia on males for 10 and 21 days and also females or 10 days. Usually these types of studies focus on the physiological (cardiovascular, musculoskeletal and metabolical) changes associated with microgravity (bed rest), however, an important factor sometimes overlocked is the psychological well being of the participants. I would like to present to you some of the psychological results we have obtained during these studies.

27.02.2014: Miha Deniša
Iskanje in sinteza podajnih gibov v bazi vzorčnih trajektorij / Synthesizing Compliant Movements by Searching a Database of Example Tasks

The talk will address the problem of generating new compliant reaching movements by searching a structured database of example trajectories. The proposed control framework is a multi-step process, where in the first step a human tutor teaches the robot how to perform a set of example reaching movements. In the second step, the recorded motion trajectories are executed with different velocities using a high gain feedback controller, for the purpose of learning a corresponding torque control signals of the executed behavior. The commanded torques are measured and saved together with the trajectory data. Once these data are acquired, they are 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 in the graph. The proposed approach can construct a complete representation for newly discovered movement primitives, including the torque feedforward trajectory. The task is executed using a low gain feedback controller and the associated feedforward torque signal. This ensures sufficient tracking accuracy and at the same time compliant behavior, which allows smooth interaction with the environment and is safe for cooperative task with humans. The usefulness of the proposed method was shown on a Kuka LWR robot.

06.02.2014: Tarsi Bali
/ Planetary Habitat Simulation (PlanHab)

Seminar room, 13:00 In future planetary habitats the partial pressure of oxygen will be lower than in atmospheric air and prolonged exposure to low gravity will result in deconditioning of physiological systems. The aim of PlanHab was to investigate how prolonged exposure to both hypoxia and reduced gravity will affect human health. In tomorrow’s presentation the purpose and methodology of particular projects of PlanHab such as the ‘’Autonomic Temperature Regulation Study’’, ‘’the muscle strength and body composition measurements’’ will be presented as well as part of the experimental results.

23.01.2014: Denis Forte
Sejna soba ob 13h: Posploševanje knjižnic robotskih gibov s statističnimi metodami
Na seminarju bom predstavil končno verzijo prezentacije za doktorat in še nekaj nedavnih rezultatov s področja pospešenega učenja robotov. Glavni problem današnjih robotov je, da se še ne znajo odzivati na spremenljive razmere v nestrukturiranih okoljih. Ugodno bi bilo, da bi robotom pokazali opravljanje naloge v različnih okoliščinah, nato pa bi le-ti s posploševanjem vzorčnih gibov znali oceniti pravilno izvedbo naloge v podobnih razmerah ali pa celo izpopolnili svoje gibanje, da bi bila naloga kar najbolje opravljena. Pri našem delu, smo robotu najprej pokazali nekaj primerov pravilne izvedene naloge, ter takšno robotsko gibanje zapisali v DMP-je z ustreznim številom baznih funkcij. Vzorčne gibe smo povezali s pripadajočimi značilkami naloge, ki jih imenujemo tudi vzorčne iskalne točke, in tako ustvarili bazo vzorčnih podatkov. Podatke v takšni obliki lahko posredujemo predlagani metodi statističnega posploševanja, ki ob novi situaciji (oz. novi iskalni točki) ter zadostnem številu podatkov generira aproksimirano trajektorijo, ki uspešno reši nalogo. V primeru, da število vzorčnih podatkov ne zadostuje za korektno izvedbo naloge, pa aproksimiran gib podamo spodbujevanemu učenju, ki le-tega izboljšuje toliko časa, dokler ni naloga ustrezno opravljena, nato pa ga lahko shranimo v obstoječo bazo podatkov. Z večanjem baze podatkov statistično posploševanje vse natančneje ocenjuje gibe, dokler na neki stopnji spodbujevano učenje ni več potrebno. V delu smo predlagali metodo spodbujevanega učenja, ki na podlagi nekaj vzorčnih gibov zmanjšuje dimenzijo prostora s pomočjo statističnega modela, ki preslika iskalne točke v parametre giba. Spodbujevano učenje tako večino iskanja opravi v prostoru iskalnih točk, ki imajo precej nižjo dimenzijo kot izhodni podatki, ki predstavljajo trajektorijo.

16.01.2014: Barry Ridge
/ Staged Self-Supervised Learning of Basic Object Push Affordances

Seminar room at 13:00 Continuous learning of object affordances in a cognitive robot is a challenging problem whose solution arguably requires a developmental approach. In this talk we describe a scenario where a robotic system interacts with household objects by pushing them using a robot arm while observing the scene with its vision systems, and must incrementally learn, without explicit supervision, both the resulting object effect semantic categories that emerge from these interactions as well as a discriminative model for predicting them from object properties. Through the use of feature relevance mechanisms, the learning algorithm can determine that certain features are more relevant for class discovery and prediction than others, thus, we take a look at the idea of staged developmental learning, where learning proceeds in stages where less relevant features are discarded in order to refine the learning process.

20.12.2013: Bojan Nemec
Sejna soba, 13:00 - Estimacija poze smučarja s pomočjo GPS tehnologije / Seminar room, 13:00 - Skier's Posture Estimation Using Real Time Kinematics Gnss Measurements
V seminarju obravnavamo problem, kako estimirati pozo smučarja, pri čemer imamo na voljo samo trajektorijo vratu, ki jo izmerimo s pomočjo GPS tehnologije. Predlagali bomo dva pistopa : a) z uporabo inverznega nihala in b) s pomočjo nevronskih mrež.

In this talk we deal with the problem how to estimate the skier's pose, when only one point of the skier's body is available. This point is measured using real time GNSS technology. We will propose two approaches: a) using an inverted pendulum and b) using neural networks.

12.12.2013: Bojan Nemec
Pospešitev izvajanja demonstriranih nalog z roboti / Velocity adaptation for self-improvement of skills learned from user demonstrations
Sejna soba ob 13:00 Povzetek-V prispevku rešujemo problem, kako povečati hitrost nalog, ki vključujejo fizično interakcijo z okoljem in kjer smo nalogo naučili s kinestetičnim vodenjem. Naš pristop ne potrebuje modela okolja ali modela robota, ker hitrost izvajanja prilagaja iterativno. Pri tem uporabljamo dva pristopa - Iterativno učenje (ILC) in spodbujevano učenje (RL). Pristop smo priskusili na dvoročnem robotskem sistemu KUKA LWR, kjer smo sestavljali napravo za okraševanje tort.

Seminar room at 13:00 Abstract—We address the problem of how to increase the speed of movements that occur in contact with the environment, where the initial movements were acquired by kinesthetic guiding. We take into account dynamic capabilities and constrains of both the robot and the environment. This leads to a modified, non-uniformly accelerated motion. To enable the non-uniform modulation of the movement policy, we encode the initial control policy using an extended formulation of dynamic movement primitives. The initial policy is improved using feedback error adaptation, ILC-based learning or reinforcement learning. We propose a new policy learning algorithm which takes into account intermediate rewards during the policy learning. The proposed approach was experimentally evaluated on a bimanual kitchen task, where the robot, composed of two KUKA LWR arms, had to assemble a cake decoration tool.

05.12.2013: Luka Peternel
Metoda za co-adaptacijo človeka in eksoskeleta pri izvedbi periodičnih nalog / Human-exoskeleton co-adaption method for periodic tasks
Sejna soba ob 13h: Tema seminarja je nova metoda za vodenje obtelesnih robotskih mehanizmov (eksoskeletov). Ta metoda temelji na človekovi povratni informaciji pridobljeni iz mišične aktivnosti. Mišično aktivnost smo merili s pomočjo elektromiografije (EMG). Za vsak sklep dobimo mišično aktivacijo iz pripadajočih antagonističnih mišičnih skupin. Ta povratna informacija potem predstavlja signal za oblikovanje trajektorije gibanja za robotski sklep. Trajektorije smo zapisali s pomočjo Dinamičnih Generatorjev Gibanja (DMP). Učenje trajektorij je potekalo preko Lokalno Utežene Regresije. Obstoječi DMP sistem smo modificirali tako, da trajektorije niso bile direktno naučene, ampak je so se postopoma posodabljale v posamezni fazi. Predhodno omenjen signal povratne informacije smo uporabili kot signal za posodabljanje trajektorij. Ker smo delali z periodičnimi nalogami, smo uporabili Adaptivne Oscilatorje za izločanje faze in frekvence naloge, ter ju posredovali trajektorijam. Predlagani sistem omogoča, da eksoskelet postopno prilagaja navore v svojih sklepih tako, da sledi človekovemu obnašanju, in mu pomaga pri izvedbi danih nalog.

Seminar room, 13:00 The topic of seminar is the novel method for exoskeleton control. The method is based on biofeedback obtained from human muscle activity. The muscle activity is measured by the means of electromyography (EMG). We get the muscle activity feedback signal for each joint from corresponding groups of antagonistic muscles. The feedback signal is then used to shape and reshape the motion trajectory of exoskeleton joint. The trajectory is encoded with Dynamical Movement Primitives (DMP). The learning of the trajectory is done by Locally Weighted Regression. The existing DMP system was modified in a way that the trajectory is not directly learned but gradually updated by phase. We use the aforementioned feedback signal to update the trajectories. Since we work with periodic task we use Adaptive Frequency Oscillators to extract the task phase and frequency from the human biofeedback and feed it to the DMPs. The proposed system makes the exoskeleton to gradually adapt its joint to torques to follow the human behaviour and assist him/her in the given tasks.

28.11.2013: Andrej Gams
/ Rich Periodic Motor Skills on Humanoid Robots

Seminar room, today at 13:00 In this talk I will present the work of the second half of my stay at EPFL. Just as their discrete counterparts, periodic or rhythmic dynamic motion primitives allow easily modulated and robust motion generation, but for periodic tasks. I will present an approach for modulating periodic dynamic movement primitives based on force feedback, allowing for rich motor behavior and skills. The combination of feedback and learned feed-forward terms allows to fully adapt the motions of a robot in order to achieve a desired force interaction with the environment. To demonstrate the approach, I will show results of simulated and real world experiments on a compliant humanoid robot COMAN.

26.11.2013: Shawnda Morrison
Bivanje v ekstremnih okoljih / 11:00: Moving in Extreme Environments- IJS Research Summary 2011-2013

This seminar update will focus on the human physiology research I have been involved in over the past 2.5 years with Dr. Mekjavic's team in our department. I will outline the rationale for the various research projects, including the study design, methods and preliminary findings and / or publications resulting from this research programme. This talk will sequentially move through three main areas of research: projects investigating Freezing Cold Injuries, Behavioural Thermoregulation and Advances in Sleep Physiology.