Differences

This shows you the differences between two versions of the page.

--- start [2021/03/31 12:52]
Matteo Saveriano
+++ start [2026/08/23 06:40] (current)
Justus Piater [Working With Us]
@@ Line 7: / Line 7: @@
 We seek to answer the question: //How can we enable robots to acquire the knowledge and understanding they require to interact sensibly with unstructured environments? //
-Our research addresses complete perception-action loops, from computer vision to grasping and manipulation, using reactive algorithms and/or cognitive models.  Much of our work uses machine learning to enable robots to synthesize and improve complex and robust sensorimotor behavior with experience. Related areas of interest include human-robot interaction, image and video analysis, and visual neuroscience.
+Our research addresses complete perception-action loops, from computer vision to grasping and manipulation, using reactive algorithms and/or cognitive models.  Much of our work uses machine learning to enable robots to synthesize and improve complex and robust sensorimotor behavior with experience.
 ===== Working With Us =====
   * Check our thesis topics for [[theses/start|Bachelor and Master students]].
-  * [[https://ellis.eu/|ELLIS]] PhD Program: [[https://ellis.eu/de/news/ellis-phd-program-call-for-applications|Call for Applications]]
+  * [[:jobs:#notice_for_non-eueea_prospective_master_students|Notice]] for non-[[https://en.wikipedia.org/wiki/EU|EU]]/[[https://en.wikipedia.org/wiki/European_Economic_Area|EEA]] prospective Master students
+  * We do not currently have any open PhD positions.
-<html><!--
-//Bachelor students://{{ ::mobile-manipulator-crop.jpg?nolink&100|}} **Want to create artificial intelligence for autonomous robots?** Want to join our [[@/uibk/piater/courses/RC_Poster2.pdf |interdisciplinary LFUI team]] to compete in the 2020 [[https://www.robocup.org/leagues/16|RoboCup@Work]] competition? Take the [[https://lfuonline.uibk.ac.at/public/lfuonline_lv.details?sem_id_in=19W&lvnr_id_in=703135|Introduction to Robotics]] course in the winter semester 2019-20!
---></html>
 <html>
@@ Line 22: / Line 20: @@
 </html>
+{{ ::iis_retreat_2024.jpg?direct&900 |}}\\
-{{::dsc_0052.jpg?640|Group Picture}}\\
+<html>
-Group picture taken at our retreat in Obergurgl
+<div style="text-align: center;">
+  <p>Group picture taken at our 2024 retreat at Meissner Haus.</p>
+</div>
+</html>
 ===== News =====
-<HTML><table><tr valign="top"><td>2021-03-24</td><td>Justus Piater gives an invited lecture <i>Machine Learning, Perception, And Abstract Concepts</i> at Invited lecture at Ontario Tech U, Canada, online. <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem0', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem0Abstract" style="display:none">With every spectacular achievement of a machine learning
+<HTML><table><tr valign="top"><td>2026-09-27</td><td>Justus Piater co-organizes the <a href="https://worldmodelworkshop.github.io/">RoBoWoMo:
-    system, the long-elusive AI breakthrough is popularly proclaimed
+    Bridging the Gap between Neural and Symbolic World Models for
-    to be just around the corner.  Most recent successes have been due
+    Robot Planning, Reasoning, and Action</a>.</td></tr><tr valign="top"><td>2026-05-20</td><td>Simon Haller-Seeber appears in the media: <a href="https://www.digital.tirol/page.cfm?vpath=news&amp;rnpageid=42902">
-    in large part to massive data and computation, in particular using
+        Digitale Bildung aktiv erleben: Der RoboCupJunior Austrian Open 2026 in Tirol</a>. (Interview im Rahmen des RoboCup Junior Austrian Open)</td></tr><tr valign="top"><td>2026-05-07</td><td>Justus Piater is a panelist at a public discussion on <a href="https://www.uibk.ac.at/de/medien-kommunikation/kommunikation/intern/">Invited
-    deep artificial neural networks.  But can artificial cognition
+    expert interview on “Human-robotics relations: What does the
-    really be achieved just by further scaling up existing
+    future hold?”</a>, Innsbruck. (Public event organized by the Department of Media, Society and Communication, Universität Innsbruck)</td></tr><tr valign="top"><td>2026-04-15</td><td>Justus Piater gives an invited talk <i>Some Latest Results in Robot Learning of Structure by
-    machine-learning techniques?  I discuss examples of simple,
+    Interaction</i> at <a href="https://airov.at/2026/workshop/TactileRobotics.html">Austrian
-    perceptual problems that are easily solved by humans but very
+    Robotics Workshop</a>, Leoben. (Annual workshop of the GMAR; at AIRoV 2026)</td></tr><tr valign="top"><td>2026-04-08</td><td>Simon Haller-Seeber and Justus Piater co-organize the <a href="https://robocupjunior.at/">RCJ2026 - Robocup Junior Austrian Open: 08.-10.4.2026</a>.</td></tr><tr valign="top"><td>2026-04-08</td><td>Justus Piater gives an invited talk <i>Generative KI: Funktionsweise, Möglichkeiten und
-    difficult for today's machine learning methods.  These problems
+    Grenzen</i> at SchulleiterInnen-Tagung „IT-Sicherheit und KI-Einsatz in
-    reflect how humans conceptualize their world.  Their mastery is
+    der Schule“, Pädagogische Hochschule Tirol. (Veranstaltung für Schulleitungen der Bildungsdirektion für Tirol) <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem6', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem6Abstract" style="display:none">Generative KI-Systeme wie ChatGPT und Midjourney haben
-    thus likely to be an essential prerequisite for autonomous robots
+    die Welt im Sturm erobert. Wie können wir sie produktiv nutzen?
-    to attain higher levels of cognitive abilities.  To get there, a
+    Wie können wir mit den Herausforderungen umgehen, die durch sie
-    few core issues can be identified that should drive research in
+    entstehen? Ich werde diese Fragen von der technischen Seite her
-    cognitive robotics.</blockquote></td></tr><tr valign="top"><td>2021-01-28</td><td>Matteo Saveriano gives an invited talk <i>Making robots to learn from human observation?</i> at <a href="https://www.bmeia.gv.at/en/austrian-embassy-ottawa/news/events/detail/article/showcasing-young-austrian-scholars-and-scientists-an-original-seven-part-virtual-lectures-ser/">Showcasing Young Austrian Scholars and Scientists, Austrian cultural forum, Ottawa</a>, online.</td></tr><tr valign="top"><td>2021-01-14</td><td>Matteo Saveriano gives an invited talk <i>Hierarchical action decomposition and motion learning for the execution of manipulation tasks</i> at <a href="http://www.gmar.at/aktuell/">Hello Tyrol calling! Robotics Talk online, GMAR, Innsbruck</a>, online.</td></tr><tr valign="top"><td>2020-11-19</td><td>Justus Piater gives an invited talk <i>Machine Learning in Robotics</i> at <a href="https://baiome.org/">bAIome PI Talk, Center for
+    angehen und einen Eindruck davon vermitteln, wie diese Systeme
-    Biomedical AI, University Medical Center Hamburg-Eppendorf</a>, online. <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem4', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem4Abstract" style="display:none">Machine Learning increasingly equips robots with
+    funktionieren. Daraus ergeben sich ein realistisches Verständnis
-    learning capabilities and flexibility.  This will enable them to
+    für ihre Möglichkeiten und Grenzen sowie einige grundlegende
-    act purposefully in unstructured environments and to react to
+    Empfehlungen für den Umgang mit ihnen im Schulbetrieb.</blockquote></td></tr><tr valign="top"><td>2026-03-31</td><td>Samuele Tosatto gives an invited keynote <i>Accelerating Reinforcement Learning with Off-Policy Data: Promises, Pitfalls, and Future Directions</i> at <a href="https://rl4aa.github.io/RL4AA26/">Reinforcement Learning For Autonomous Accelerators 2026</a>, Liverpool. <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem7', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem7Abstract" style="display:none">Reinforcement learning is a promising technique for solving complex control problems in real-world physical systems, such as robotics, plasma stabilization, and particle accelerators. However, RL is often data-hungry, and its classic on-policy formulation is often inefficient, as it disallows data reuse, and unsafe, as it requires the agent to interact with the environment from scratch.
-    unforeseen events.  People can teach them intuitively to perform
+Off-policy reinforcement learning offers a more appealing paradigm by enabling the reuse of historical data and the utilization of safe, external behavior sources (such as human operator logs). However, this flexibility comes at a cost: off-policy learning introduces significant theoretical instabilities. In this talk, we will analyze some fundamental difficulties in off-policy reinforcement learning, both in value and policy learning, explore the algorithmic landscape that tames them, and see the future direction in which the field is moving. </blockquote></td></tr><tr valign="top"><td>2025-11-19</td><td>Justus Piater gives an invited talk <i>Structural Understanding – The Grand Challenge of Robot
-    tasks instead of having to program them in painstaking ways.
+    Learning</i> at <a href="https://elliit.se/news-and-events/focus-period-lund-2025/symposium/">ELLIIT Focus Period Symposium: Robot Learning</a>, Lund University. (The ELLIIT Focus Period Symposium is the highlight of the five-week focus period, during which young international scholars, ELLIIT researchers and other well-established international academics gather in Lund to work together on joint research challenges.) <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem8', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem8Abstract" style="display:none">AI has made great progress in recent years, and the
-    Robots can learn from experience and can improve their behavior
+    sophistication of robots has been rising with costs falling. Yet,
-    over time.  In this talk I will give an overview of methods,
+    the capabilities of AI-enabled robots are not keeping pace. I
-    opportunities, and challenges of machine learning in
+    argue that this is due to a lack of structural understanding by
-    robotics.</blockquote></td></tr><tr valign="top"><td>2020-11-20</td><td>Simon Haller-Seeber and Patrick Lamprecht present a show <i>Explainable AI: A sneak peek into the Black-Box</i> at <a href="https://youtube.com/watch?v=4NVfPwdLnCg&amp;t=17m43s">Science Slam</a>, online.</td></tr>
+    current AI systems. I will discuss several lines of research in my
+    lab that seek to enable robots to generalize better and learn
-    <tr valign="top"><td>2020-10-22/23</td><td>Matteo Saveriano, Erwan Renaudo, Antonio Rodríguez-Sánchez, and Justus Piater organize the </i> <a href="https://iis.uibk.ac.at/conferences/hfr2020/"> 13th International Workshop on Human-Friendly Robotics (HFR 2020)</a>, Innsbruck (online).</td></tr>
+    faster thanks to explicit notions of structure.</blockquote></td></tr><tr valign="top"><td>2025-09-18</td><td>Samuele Tosatto gives an invited keynote <i>Where are all the intelligent robots? A quest for efficiency in reinforcement learning</i> at <a href="https://sarl-plus.github.io/RL-Bootcamp2025/">Reinforcement Learning Bootcamp 2025</a>, Salzburg. <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem9', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem9Abstract" style="display:none">As artificial intelligence reshapes our digital world at a breathtaking pace,
+    a curious question arises: Where are all the real-world, intelligent robots?
-    <tr valign="top"><td>2020-09-30</td><td>Erwan Renaudo contributes a talk <i>ROSSINI: RobOt kidS deSIgn thiNkIng</i> at <a href="https://www.springer.com/gp/book/9783030674106">Robotics in
+    While we have mastered the generation of text, images, and video by leveraging vast web-scale datasets,
-    Education 2020</a>, online.</td></tr><tr valign="top"><td>2020-06-22</td><td>Justus Piater gives an invited talk <i>Conditional Neural Movement Primitives</i> at <a href="http://www.gdr-isis.fr/index.php?page=reunion&amp;idreunion=424">GdR
+    robotics still faces a fundamental data bottleneck. Reinforcement learning (RL) offers a compelling solution,
-    ISIS Réunion Apprentissage et Robotique</a>, online. <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem7', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem7Abstract" style="display:none">Conditional Neural Movement Primitives (CNMP) constitute
+    enabling agents to learn autonomously by collecting their own experience.
-    a novel framework for robot programming by demonstration based on
+    However, the path to autonomy is often blocked by the staggering inefficiency of current RL algorithms,
-    Conditional Neural Processes (CNP).  Like Bayesian methods such as
+    which can require millions of trials to master simple tasks. This talk embarks on a quest to tackle this
-    Gaussian Processes (GP), CNP learn how target distributions depend
+    efficiency problem head-on. I will argue that a crucial step toward unlocking the potential of
-    on data, and can be conditioned on specific data points to infer
+    robot learning lies in a two-pronged approach: first, by developing statistically efficient
-    new target distributions at test time.  Unlike GP that are
+    algorithms that can reuse data by leveraging sound off-policy techniques, and second,
-    expensive to train and scale poorly to high dimensions, CNP are
+    by designing better action representations for physical, real-world agents.
-    neural networks and are trained by gradient descent.  CNMP
+    By making our algorithms more efficient and refining their core hypotheses,
-    leverage CNP to represent motion trajectories that can be
+    we can accelerate the journey toward real embodied artificial intelligence.</blockquote></td></tr><tr valign="top"><td>2025-07-15</td><td>Justus Piater gives an invited keynote <i>Making robots learn to perceive and act with
-    conditioned, at test time, on task paramters such as goal
+    understanding</i> at <a href="https://www.uibk.ac.at/en/congress/multibody2025/">The
-    locations, via points, and/or force readings.  Moreover, CNMP are
+th ECCOMAS Thematic Conference on Multibody Dynamics</a>, Innsbruck. <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem10', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem10Abstract" style="display:none">The flexibility and robustness of current robots is
-    conditioned on sensor readings during execution, resulting in
+    limited by their lack of understanding of their environment. For
-    robust, reactive behavior.  This talk will present an overview of
+    this reason, most robots operate in controlled environments.
-    how CNMP work and how they can be used in various robot
+    Machine learning can circumvent modeling problems but introduces
-    applications.</blockquote></td></tr><tr valign="top"><td>2020-06-03</td><td>Justus Piater appears in the media: Wie der Roboter denken lernt.</td></tr><tr valign="top"><td>2020-01-29</td><td>Justus Piater gives an invited talk <i>Digital Science</i> at <a href="https://www.graduateacademy.uni-heidelberg.de/karriere/veranstaltungsreihen.html">Vortragsreihe
+    new problems of generalizing from examples. How can robots acquire
-    „Primers for Predocs – Strategien für eine erfolgreiche
+    understanding (of structure, function, causality, etc.) that
-    Promotion“</a>, Universität Heidelberg. <span class="actions"><a href="javascript:void(0)" onclick="showHide('newsitem10', 'Abstract')">[Abstract]</a></span><blockquote id="newsitem10Abstract" style="display:none">Massive availability of data and computing power are
+    allows them to generalize from sparse experience? Motivated by
-    promoting data-driven methods in all areas of science and
+    shortcomings of current machine-learning methods, I will argue
-    technology.  I will describe how the University of Innsbruck
+    that &quot;understanding&quot; is a meaningful notion in AI that reaches
-    supports this via its new Digital Science Center, and will give a
+    beyond prediction and control. I will discuss examples of our
-    flavor of machine learning for data analysis.</blockquote></td></tr><tr valign="top"><td>2020-01-20</td><td>Joanna Chimiak-Opoka, Carina König, and Justus Piater appear in the media: <a href="https://www.uibk.ac.at/newsroom/ergaenzung-digital-science-erfolgreich-gestartet.html.de">Ergänzung Digital Science erfolgreich gestartet – UIBK Newsroom</a>.</td></tr></table></HTML>
+    recent work on learning visual relational concepts, extrapolation
+    of learned movements beyond the training distribution, learning of
+    symbolic concepts and rules, and structure-driven skill learning
+    from sensorimotor experience. Our long-term objective is to
+    improve abstraction, generalization, robustness, and ultimately
+    explainability of robot perception and action.</blockquote></td></tr></table></HTML>
 [[news|Older News]]
@@ Line 89: / Line 94: @@
 Austria
-**How to find us:** See the [[http://informatik.uibk.ac.at/how-to-reach-us/|directions]].
+**How to find us:** See the [[https://www.uibk.ac.at/informatik/kontakt/anfahrt.html.en|directions]].
 **Legal Notice:** See the [[:impressum|Impress and Privacy Notice]].

User Tools

Site Tools

Differences

Page Tools