Ni napovedanih predavanj. No planned seminars. Pretekla predavanja / Past seminars - 2012/2013 13.06.2013: Leon Žlajpah Izvedba nalog glede na prioriteto / 13h: Execution of prioritized tasks
Redundant robot manipulators, especially multi-arm robots or humanoid robots, have more degrees-of-freedom (DOFs) than those required to fulfill a task. Hence, they are expected to achieve several tasks simultaneously. Tasks can be positions of the robot end-effectors, positions of some other part on the robot body, or some constraining tasks like a desired robot posture, avoiding obstacles, joint limits and/or singular configurations, etc. The control problem for redundant robot manipulators doing more than one task can be formulated as the problem of finding a motion that will fulfill all the tasks simultaneously. Generally, this problem is addressed in the task-priority framework where the tasks are ordered by priority. We will present different task-priority redundancy resolution techniques to allow the specification of the primary task that is fulfilled with the higher priority with respect to a secondary tasks. One possible extension of general method to multiple task a successive method, where the motion of the lower-priority task was is the null-space of the next higher-primary task. A more correct formulation is to project a motion of a lower-priority task onto the null-space of the all higher-priority tasks, i.e. into the null space of the extended Jacobian. Another possibility is task augmentation where all tasks are fulfilled simultaneously. The main problems of these approaches are that near singular configurations excessive joint-space velocities (practically
infeasible) may result and that discontinuous joint-space solutions arise at singular configurations. These motivated the development of a new methods. One of them is to use the damped least-squares, which damps joint velocities near ill-conditioned configurations and a combination of extended Jacobian approach and the damped least-square method augmented with new factors for task priority selection. The effectiveness of these methods will be illustrated by simulation of highly redundant robots.
31.05.2013: Jan Babič / 13:00 Whole-body Compliant Dynamical Contacts for Humanoid Robotics
For complex systems, such as humanoid robots, to persist and act in natural environments like humans do, contacts and physical interaction become necessary and unavoidable. Robots may exploit predictable contacts to aid in goal achievement, as well as learn dynamics of contact in order to generalize over novel tasks and domains. Critically, robots should also be robust enough to cope with unpredictable contacts, via safe control mechanisms and compliance.
In the seminar I will introduce our new EU project titled Whole-body Compliant Dynamical Contacts for Humanoid Robotics (Codyco) that aims to advance the current control and cognitive understanding about robust, goal-directed whole-body motion interaction with multiple contacts. It will go beyond traditional approaches by proposing methodologies for performing coordinated interaction tasks with complex systems, by combining planning and compliance to deal with predictable and unpredictable events and contacts, and by validating theoretical advances in real-world interaction scenarios.
16.05.2013: Barry Ridge / 13:00 Action-Grounded Push Affordance Bootstrapping of Unknown Objects
When it comes to learning how to manipulate objects from experience with minimal prior knowledge, robots encounter significant challenges. When the objects are unknown to the robot, the lack of prior object models demands a robust feature descriptor such that the robot can reliably compare objects and the effects of their manipulation. In this talk, work that was recently submitted to IROS 2013 will be presented where, using an experimental platform that gathers 3-D data from the Kinect RGB-D sensor, as well as push action trajectories from a tracking system, we address these issues using an action-grounded 3-D feature descriptor. Rather than using pose-invariant visual features, as is often the case with object recognition, we ground the features of objects with respect to their manipulation, that is, by using shape features that describe the surface of an object relative to the push contact point and direction. Using this setup, object push affordance learning trials are performed by a human and both pre-push and post-push object features are gathered, as well as push action trajectories. A self-supervised multi-view online learning algorithm is employed to bootstrap both the discovery of affordance classes in the post-push view, as well as a discriminative model for predicting them in the pre-push view. Experimental results demonstrate the effectiveness of self-supervised class discovery, class prediction and feature relevance determination on a collection of unknown objects.
15.05.2013: Miha Deniša 13:00 Iskanje in sinteza novih robotskih gibanj v hierarhični bazi vzorčnih trajektorij - Zagovor teme doktorske disertacije / 13:00 Discovery and synthesis of new robot control policies through search in a hierarchical database of example movements - Ph.D. thesis proposal Predlog teme obravnava pridobivanje novih robotskih gibanj preko človeških demonstracij.
Razvili bomo metodo, ki bo manjše število raznolikih vzorčnih gibov zapisala v hierarhično bazo in preko nje omogočala sintezo in izvajanje nalog v spremenjenih okoliščinah. Hierarhična baza vzorčnih gibov bo oblike dvojiškega drevesa z grafi prehodov na vsakem nivoju. Grafi prehodov bodo opisali možne prehode med posameznimi vozlišči. Vsak nivo baze bo predstavljal vse vzorčne gibe na različnih nivojih razdrobljenosti. Preko hierarhičnega iskanja delnih poti in interpolacije, bomo omogočili sintezo novih, še ne demonstriranih, gibov. To večjo bazo bomo združili z metodami statističnega posploševanja in jih tako še nadaljnjo prilagodili različnim okoliščinam. Poleg tega bomo zagotovili tudi izvedbo gibov na robotu. Preizkusili bomo še uporabo hierarhične baze za razpoznavo človeških gibov in sintezo primernih kooperacijskih gibov robota.
Metodo bomo ocenili z več eksperimenti in z učenjem različnih nalog.
The talk will sadly be in Slovenian, as this will be a "test" presentation before my Ph.D. thesis proposal defense on Friday.
10.04.2013: Luka Peternel Učenje robotskega gibanja pri interakciji z okoljem in človekom / Robot learning for dynamical interaction with environment and human
11. 4. ob 13h.
A common goal in humanoid robotics is to make human-like robots and integrate them into the human environment, where they should work alongside human as assistants, caretakers and companions. The human environment is unstructured and filled with uncertainties. If the robots are to work and perform tasks in human environment they will have to be adaptable and be able to dynamically interact with human and human environment. In the seminar I will present robot learning method which exploits human sensorimotor learning and ability to control various tools by adapting the tool in their body schema. The human is placed in robot control loop and learns how to operate the robot. The human knowledge is then used to build an autonomous controller which then replaces the human operator. The feasibility of this method is shown by experiments concerning robot interaction with uncertain environment (human, cooperative tool use).
05.04.2013: Denis Forte Uporaba Gaussove regresije v sodelovanju s spodbujevalnim ucenjem z namenom samostojnejsega ucenja robotov / Using Gaussian process regression with reinforcement learning to make robot learning more autonomous Predstavil bom metodo ucenja robotov, ki zdruzuje ucenje z demonstracijo in spodbujevalno ucenje. Ideja je, da robotu posredujemo karseda malo trening podatkov, in tako lahko ucenje poteka bolj samostojno. Najprej je bilo zajetih nekaj primerov izvajanja naloge na robovih delovnega prostora. Nato je bilo uporabljeno statisticno posplosevanje za izracun sorazmerno dobrega zacetnega priblizka glede na trenutno situacijo okolja. Kot naslednji korak je bilo uporabljeno spodbujevalno ucenje za izboljsavo zacetnega priblizka v nekaj korakih, dokler ni robot zadovoljivo opravil naloge. Vsak naucen gib se nato shrani v bazo podatkov, da so nadaljnje ocene statisticnega posplosevanja natancnejse, in da bi posledicno spodbujevalno ucenje dalo hitrejse rezultate. Primerjal sem, kako hitro je ucni prostor dodobra raziskan z uporabo Gaussove regresije (kot metode za
posplosevanje) v sodelovanju s spodbujevalnim ucenjem PoWER, ter koliko se izboljsa podobnost med ucnimi podatki, ko zraven uporabljamo se spodbujevalno ucenje PI2. Pristopi so bili preizkuseni z ucenjem natakanja na robotski roki »KUKA Light-Weight Robot« s sedmimi prostostnimi stopnjami.
I will present methodology for robot learning that combines learning by demonstration and reinforcement learning. The idea is to provide as few training data as needed, so the robot can learn more autonomously.
First a few training movements were obtained that solved the given task in some specific situations. Then statistical generalization was applied to compute relatively good initial approximation of new situation inside learning space. As the next step the reinforcement learning was used to refine the approximation in few steps so the robot could accomplish the task correctly. Every learned movement was then stored in the training base, so that additional approximations of different situations could be estimated more accurately and that reinforcement learning could get faster results. I compared how fast the learning space is fairly revealed by using Gaussian process regression (as a generalization method) in cooperation with reinforcement learning PoWER and also, how much better the similarity of training data is when using reinforcement learning PI2 together with PoWER. Approaches were tested by learning of pouring action with
7 DOF KUKA Light-Weight Robot arm.
21.02.2013: Tadej Debevec Pozitivni in negativni učinki hipoksije, vadbe in hipoksične vadbe / Hypoxia, exercise & hypoxic exercise: Good or Bad? Tako hipoksija, kot športna vadba močno vplivata na človeški organizem. Osnovna tema seminarja bodo potencialni negativni učinki, predvsem z vidika oksidativnega stresa, dolgotrajne neprekinjene izpostavitve hipoksiji, v kombinaciji z ali brez srednje-intenzive vzdržljivostne vadbe. Predstavljeni bodo preliminarni rezultati raziskave HEC, ki je bila izvedena leta 2012 v Olimpijskem športnem centru v Planici.
Both hypoxia and exercise have significant effects on human organism. Although the majority of up-to date research work on hypoxia and exercise aimed to establish potential benefits for performance/weight loss, this talk will focus on the possible deleterious effects of continuous hypoxic exposure combined with moderate intensity exercise training or hypoxia exposure only. Preliminary results from the Hypoxic exercise confinement (HEC) study performed in 2012 @ Planica Olympic sports centre will be presented with a special reference to oxidative stress.
13:00: I'm going to present my work during my research stay at German Aerospace Center (DLR) where my main focus was on the bipedal walking (without thinking). The principle behind bipedal walking is up to now far from being completely understood. Different biological inspired models have been used to identify key features of biped locomotion. Among them is the Spring Loaded Inverted Pendulum (SLIP), which allows generation of stable limit cycles and human like contact force profiles. However, this approach has some limitations: it is highly parameter dependent; one cannot easy select desired walking velocity. Therefore, to further exploit the mechanical structures, we proposed a building block, which includes movement prototypes with corresponding dynamics implemented as an extended version of dynamical systems (DMPs). The outputs of these building blocks are the desired trajectories and the corresponding feed-forward torques.
25.01.2013: dr. Luka Šušeteršič, IJS / The bibliographic method used to evaluate work of researchers
Dr. Luka Šušteršič, IJS will explain the bibliographic method used to evaluate our work. This information is extremely important for us, as we are continuously evaluated individually, and as a group, based on this scoring method. When we apply for promotion at IJS, Dr. Šušteršič evaluates our bibliography, and provides his analysis to the IJS Scientific Committee. In preparation for the evaluation of our Programme Group later this year, it is extremely important that all your references are included in your bibliography, and that they are properly scored.
25.01.2013: Robert Križnar and Polona Kramer, Alpina d.d.
Historically, our department has had a very good and close relationship with the R&D group at Alpina d.d. (a shoe manufacturing company). We have provided expertise in the automation of the manufacturing process (Leon Žlajpah and Bojan Nemec), and have been involved in the development of currently used military and police footwear (Igor Mekjavic). The aim of this meeting is to review our previous work (a brief review will be provided by Leon, Bojan & Igor), and to learn of Alpina’s current R&D focus, with the view of establishing some new areas of collaboration. The presentation will be share by the Mr. Križnar, who currently heads the R&D at Alpina, and Mrs. Polona Kramer, who is currently conducting part of her doctoral woek in our department.
18.01.2013: Aleš Ude / Humanoid and Cognitive Robotics Laboratory
Humanoid and Cognitive Robotics Lab (HCR) was established in June 2012. In my presentation I'm going to explain the organizational and financial structure of the lab. The main part of the presentation will focus on the latest research results achieved in the projects currently going on in the lab (Robots bootstrapped through learning from experience, Intelligent observation and execution of actions and manipulations, Learning, analysis, and detection of motion in the framework of a hierarchical visual architecture). I'm going to conclude with some upcoming projects (Learning and execution of action categories, Accelerated development of autonomous behaviors for humanoid robots) and project proposals.
19.12.2012: Nejc Likar / Obstacle avoidance for dual-arm redundant manipulators
The presentation will talk about obstacle avoidance as a control problem on the kinematic level for dual arm redundant manipulators. There are numerous approaches for solving the obstacle avoidance problem, but all are used for single robot. However there are some approaches concerning also dual robots, but require remodeling of the system. The method of chaining serial mechanisms serially, allowed us to model the system of two KukaLWR robots as one kinematic chain. With this method it is possible to use single robot obstacle avoidance algorithms on the dual arm cooperative system.
13.12.2012: Igor Kovač / Sustainable production with machinery system integration
German Federal Ministry of Education and Research started within the "Hightech-Strategy 2020" frame a research offensive called "Secure Company Location by Adaptive Production Systems" (Standortsicherung durch Wandlungsfähige Produktionssysteme). The sustainable production in manufacturing is characterised with a fact that the only constant is change (Hoda A. ElMaraghy). So the focus will be, to present some machinery system integration ideas, how this change could be managed efficiently.
06.12.2012: Igor Mekjavić / Human thermoregulatory function during exercise and diving.
Cannon (1929) used the regulation of body temperature as an example of a homeostatic system, comprising mechanisms acting simultaneously or in succession to maintain constancy of body temperature, and thus contribute to the homeostasis of the internal environment. Although it was established that sensory information from the skin and core regions contributes to the autonomic thermoregulatory responses of heat production, heat loss and heat retention, the manner in which these were regulated was unclear. The invention of stabilised feedback in electrical engineering (Black, 1934), also termed negative feedback, prompted Cannon’s associate Rosenbleuth, together with Wiener and Bigelow (1943) to introduce the concept of purpose controlled feedback in physiological systems. Their seminal paper lay the foundation for the field of cybernetics, and suggested that negative feedback control could be used to model physiological systems. The mutually inhibiting activity of the hypothalamic thermoregulatory centres was thus modelled as a negative feedback control system, in which the effector responses of heat production and heat loss were initiated in proportion to the displacement of body temperature from a fixed reference value, or set-point. Such rigid control of a quality and quantity of the internal environment are not in line with the concept of homeostasis, which resulted in the development of the adjustable set-point concept. The characteristics of the thermoafferent information and thermoregulatory effector responses have been well documented, but the manner in which the former is integrated centrally to give rise to the latter remains unresolved. The reciprocal cross inhibition theory of thermoafferent information eliminates the need for a set-point, but unlikely represents the manner in which homeostasis is achieved. Human and animal experimental evidence have now confirmed that core temperature is not maintained at a set-point, but rather within a band of core temperatures, the interthreshold zone, bound by the thresholds of sweating and shivering onset, where vasomotor activity predominates. These autonomic responses are activated by thermal factors (core and skin temperature), but can also be modified by nonthermal factors. Studies investigating autonomic and behavioural thermoregulatory responses during exercise and diving, have established the effect of nonthermal factors associated with these activities, such as dehydration, inert gas narcosis, hypercapnia, hyperoxia, hydrostatic pressure, etc. on autonomic and behavioural temperature regulation. This research has demonstrated that nonthermal factors have different patterns of action on the heat loss and heat production pathways, which has led to a re-examination of existing concepts of mammalian temperature regulation. Similar to the core temperature interthreshold zone, a skin temperature thermal comfort zone exists, bound by the thresholds for thermal discomfort. A neuronal model will be presented, incorporating separate, but interacting sensor-to-effector pathways. It will incorporate the contribution of thermal and nonthermal factors to the autonomic and behavioural regulation of body temperature.
Abstract. The goal of this work is to evaluate the application of DMPs to solve assembly tasks like peg-in-hole. In this context, DMPs need to be combined with force-torque feedback control. We propose a scheme for PiH with LbD (Learning by Demonstration) and adaptation using iterative learning. During the learning (trajectory execution) we modify the actual trajectory by an offset according to the stiffness control law. The proposed approach has been evaluated experimentally on Kuka lightweight robot arm.
11.10.2012: Andrej Gams Spreminjanje primitivov gibanja za dvoročne naloge in interakcijo z okoljem / Modulation of Motor Primitives for Bimanual Tasks and Interaction with the Environment
The framework of dynamic movement primitives allows the generation of discrete and periodic trajectories, which can be modulated in various aspects. In the seminar I will present a modulation approach which extends the framework to allow interaction with objects and the environment. The algorithm enables the coupling of independently executed robotic trajectories and thus simplifies the execution of bimanual and cooperative asks. In a few iterations the proposed algorithm learns the necessary coupling term to modify the trajectory in accordance to the desired position or external force. The strengths of the algorithm, which fits in the scope of the iterative learning control algorithms, are shown in bimanual or two-agent obstacle avoidance tasks, where no higher level cognitive reasoning or planning are required.
Results of simulated and real-world experiments on two KUKA LWR robots will be presented.
04.10.2012: Rok Vuga / Task Primitives in Learning by Demonstration
The main topic of my presentation today will be robot learning. When we humans want to learn a new task, the best thing to do is ask someone to demonstrate you how it is done. Observing this demonstration we are able to efficiently extract the required motor primitives and use this knowledge to accomplish the task by ourselves. Robots, however, can only be “taught” new things by tedious programming work. In my presentation I am going to talk about how we are going to change that. The learning process can be split into two stages: primitive extraction and task execution. For both of these processes I will present you our state of the art implementation on practical experiments.
20.09.2012: Katelyn Marsden / Cerebral blood flow dysregulation: a possible mechanism of symptom exacerbation in concussion?
Cerebral blood flow is tightly regulated to ensure optimal cognitive function by adequately delivering oxygen and other nutrients to the brain. There are several regulating mechanisms that work together and independently to ensure cerebral blood flow is well maintained to match brain activity and the needs of the cerebral tissue. However, in cases of disease or trauma, these tightly regulating mechanisms can be blunted or abolished altogether, creating cerebral instability and injury to the brain. In the case of mild traumatic brain injury, or more simply termed, a concussion, the trauma translated through the brain is quite substantial and has some severe acute and long-term consequences. However, the mechanisms behind the manifestation of symptoms, such as headaches, dizziness, and nausea have yet to be elucidated. This presentation will present a general introduction to the mechanisms of cerebral blood flow regulation and its critical role in concussion pathophysiology. I will also discuss how cerebrovascular research will help create better clinical practices and further our understanding of this complex injury.
Sleep apnoea is characterised by repeated episodes of breathing cessation during sleep resulting in periods of low oxygen (ie. hypoxia) and sleep disruption. This abnormal breathing during sleep can lead to accelerated cardiovascular disease and serious problems for the heart and brain. Typically, an oscillatory pattern is observed; changes in tidal volume can then develop into periodic breathing, and eventually, episodes of central sleep apnoea. Central sleep apnoea (CSA), involves a dysfunction of ventilatory control in the central nervous system (i.e. loss of ventilatory effort). Adverse health outcomes, including high blood pressure, heart attacks, and strokes, are extremely common in people with both obstructive and CSA. Interestingly, ascent to high altitude (>3000 m) in newcomers often leads to unstable breathing during both wakefulness and sleep. The time-course for recovery to normal breathing patterns upon return from high-altitude is not known.
This presentation will give an introduction to the general mechanisms of central sleep apnoea, the possible interaction between low-activity (bedrest) and hypoxia on the development of unstable breathing patterns during sleep, and directions for future research.