TOP
Search the Dagstuhl Website
Looking for information on the websites of the individual seminars? - Then please:
Not found what you are looking for? - Some of our services have separate websites, each with its own search option. Please check the following list:
Schloss Dagstuhl - LZI - Logo
Schloss Dagstuhl Services
Seminars
Within this website:
External resources:
  • DOOR (for registering your stay at Dagstuhl)
  • DOSA (for proposing future Dagstuhl Seminars or Dagstuhl Perspectives Workshops)
Publishing
Within this website:
External resources:
dblp
Within this website:
External resources:
  • the dblp Computer Science Bibliography


Dagstuhl Seminar 22391

Cognitive Robotics

( Sep 25 – Sep 30, 2022 )

(Click in the middle of the image to enlarge)

Permalink
Please use the following short url to reference this page: https://www.dagstuhl.de/22391

Organizers

Contact



Summary

Cognitive Robotics is concerned with endowing robots or software agents with higher level cognitive functions that involve reasoning, for example, about goals, perception, actions, the mental states of other agents, collaborative task execution, etc. This research agenda has historically been pursued by describing, in a language suitable for automated reasoning, enough of the properties of the robot, its abilities, and its environment, to permit it to make high-level decisions about how to act. Such properties were typically encoded by a human, but with recent advances in machine learning, many of these properties, and the determination of how to act, can be learned or adapted through experience. This in turn raises the question of how we can ensure that robots, or other intelligent agents, can be constructed in a manner that is compatible with human values and modes of interactions.

The Cognitive Robotics workshop series has been running since 1998 and includes a Dagstuhl Seminar held in 2010. While progress in Cognitive Robotics has undoubtedly been made over the past twenty years, it is fair to say that we are still far away from creating truly cognitive robots. In particular, the years since the previous Dagstuhl Seminar have seen tremendous progress in many areas that touch on the realisation of cognitive robots such as advances in human-robot interaction and machine learning.

This seminar featured sessions devoted to the following four themes:

Cognitive Robotics and KR: While knowledge representation and reasoning (KR) has played a role in robotic systems for many years, for example, by incorporating domain knowledge in the form of description logic-based ontologies or using automated planning systems for high-level robot control, obstacles remain, which prevent today's robots from benefiting from the true potential of KR. In this session we re-visited the state of the art of how KR is used in robotics and discussed challenges and possible benchmark problems that would demonstrate the need and benefit of KR techniques for cognitive robots. The session was organized by Michael Beetz, University of Bremen.

Verification of Cognitive Robots: Verification has been an active research area in formal methods for many years. It is also an important topic when it comes to cognitive robots, especially when it comes to achieving trustworthiness. However, the sheer complexity of the interplay between a robot's hard- and software components makes verification particularly challenging. In this session we discussed where we currently stand in terms of verifying cognitive robots and what challenges lie ahead. The session was organized by Michael Fisher, University of Manchester.

Human-robot Interaction and Robot Ethics: For cognitive robots to be useful in human environments, effective human-robot interaction (HRI) plays a crucial role. Besides the technological challenges such as multi-modal communication, ethical considerations have become more and more important. These range from robots observing norms and conventions to humans viewing robots as moral agents. In this session we discussed the many facets of robot ethics in the context of HRI and identified a number of future challenges and open problems. The session was organized by Matthias Scheutz, Tufts University.

Planning and Learning: While planning and learning have traditionally been separate research tracks in cognitive robotics, recent work has shown how action primitives that form the basis of planning can be learned from data without background knowledge, thus avoiding the need for hand-crafted solutions. In this sessions this work and related proposals were discussed and a roadmap with short- and long-term challenges was drawn up. The session was organized by Hector Geffner, ICREA and Universitat Pompeu Fabra, Spain. The format of the sessions varied and consisted of one or more plenary talks, plenary discussions and/or working groups. Working groups for all four themes discussed challenges and roadmaps for the future, and one representative of each group presented their findings on the last day of the seminar. Besides talks and discussions that centered around the four themes, the seminar also featured two invited talks by Luis Lamb, Universidade Federal Do Rio Grande Do Sul, on neurosymbolic AI and by Jan Peters, TU Darmstadt, on robot learning. In addition, a number of participants gave poster presentations on their research.

The organizers of the seminar wish to thank Schloss Dagstuhl for providing such an excellent environment for exchanging ideas on how to move the field of cognitive robotics forward.

Copyright Fredrik Heintz, Gerhard Lakemeyer, and Sheila McIlraith

Motivation

Cognitive Robotics is concerned with endowing robots or software agents with higher level cognitive functions that involve reasoning, for example, about goals, perception, actions, the mental states of other agents, collaborative task execution, etc. This research agenda has historically been pursued by describing, in a language suitable for automated reasoning, enough of the properties of the robot, its abilities, and its environment, to permit it to make high-level decisions about how to act. Such properties were typically encoded by a human, but with recent advances in machine learning, many of these properties, and the determination of how to act, can be learned or adapted through experience. This in turn raises the question of how we can ensure that robots, or other intelligent agents, can be constructed in a manner that is compatible with human values and modes of interactions.

The Cognitive Robotics workshop series has been running since 1998 and includes a Dagstuhl Seminar held in 2010. While progress in Cognitive Robotics has undoubtedly been made over the past twenty years, it is fair to say that we are still far away from creating truly cognitive robots. In particular, the years since the previous Dagstuhl Seminar have seen tremendous progress in many areas that touch on the realisation of cognitive robots such as advances in symbolic and motion planning, human-robot interaction, understanding causality, and of course machine learning.

The focus of this Cognitive Robotics Dagstuhl Seminar will be on putting these and other pieces together through the thematic lens of Human-Compatible Trustworthy Cognitive Robotics. Robots are increasingly moving out of controlled manufacturing settings and into domains where they will be residing or interacting, directly or indirectly, with humans. The goal of developing cognitive robots that can live and work alongside humans will present a unifying theme that drives discussion and technical advances. Some of the topics the seminar will investigate include:

Hybrid reasoning: Identify and discuss challenges related to combining symbolic reasoning techniques with quantitative forms of reasoning so that these methods can be put to use in cognitive robotics in a human-compatible way.

Machine learning: Identify and discuss challenges related to principled ways of integrating machine learning and reasoning and the role of end-to-end learning when building robots that are deliberative and trustworthy, while having the capacity to learn.

Human Interaction: Successful interaction and communication requires sufficient understanding of both the surroundings and the other agents. This raises the question of which KR&R methods are needed to support this understanding in a cognitive robot, often combined with learning.

AI Ethics: To realize human-compatible trustworthy AI there are many key concepts that directly relate to cognitive robotics and where methods developed in this community could be of value to a broader community. We will identify key technical problems that contribute towards the development of cognitive robots that can reside and interact with humans in an ethically acceptable manner.

The seminar aims for both a road map of research topics in cognitive robotics for the next five to ten years and to stimulate more interdisciplinary work where researchers from diverse subfields of AI, including knowledge representation and reasoning, machine learning, human computer interactions, and other cognate subfields collaborate to pursue this shared research agenda.

Copyright Fredrik Heintz, Gerhard Lakemeyer, and Sheila McIlraith

Participants
On-site
  • Michael Beetz (Universität Bremen, DE) [dblp]
  • Mohamed Behery (RWTH Aachen, DE)
  • Jens Claßen (Roskilde University, DK) [dblp]
  • Anthony Cohn (University of Leeds, GB) [dblp]
  • Frank Dignum (University of Umeå, SE) [dblp]
  • Alexander Ferrein (Fachhochschule Aachen, DE) [dblp]
  • Michael Fisher (University of Manchester, GB) [dblp]
  • Hector Geffner (UPF - Barcelona, ES) [dblp]
  • Jasmin Grosinger (University of Örebro, SE)
  • Nick Hawes (University of Oxford, GB) [dblp]
  • Fredrik Heintz (Linköping University, SE) [dblp]
  • Till Hofmann (RWTH Aachen, DE) [dblp]
  • Mikhail Khodak (Carnegie Mellon University - Pittsburgh, US)
  • Sven Koenig (USC - Los Angeles, US) [dblp]
  • Gerhard Lakemeyer (RWTH Aachen, DE) [dblp]
  • Yves Lesperance (York University - Toronto, CA) [dblp]
  • Setareh Maghsudi (Universität Tübingen, DE) [dblp]
  • Cynthia Matuszek (University of Maryland, Baltimore County, US) [dblp]
  • Christian Muise (Queen's University - Kingston, CA) [dblp]
  • Bernhard Nebel (Universität Freiburg, DE) [dblp]
  • Tim Niemueller (Intrinsic Innovation - München, DE) [dblp]
  • Ron Petrick (Heriot-Watt University - Edinburgh, GB) [dblp]
  • Sebastian Sardiña (RMIT University - Melbourne, AU) [dblp]
  • Matthias Scheutz (Tufts University - Medford, US) [dblp]
  • Stefan Schiffer (RWTH Aachen University, DE) [dblp]
  • Maayan Shvo (University of Toronto, CA) [dblp]
  • Gerald Steinbauer-Wagner (TU Graz, AT) [dblp]
  • Brian C. Williams (MIT - Cambridge, US) [dblp]
Remote:
  • Sheila McIlraith (University of Toronto, CA) [dblp]

Related Seminars
  • Dagstuhl Seminar 10081: Cognitive Robotics (2010-02-21 - 2010-02-26) (Details)

Classification
  • Artificial Intelligence
  • Robotics

Keywords
  • cognitive robotics
  • knowledge representation and reasoning
  • machine learning
  • cognitive science