IROS’12 Workshop on

Active Semantic Perception (ASP’12)

www.activeperception.org


2nd Edition:

“Context and Semantic Prior in Active Perception”


Abstract


The problem of semantic perception and mapping received a lot of attention in the recent years, with many approaches proposed and workshops devoted to the topic. At the same time, in realistic complex systems operating in large-scale environments, the perception of semantic cues must be considered an active process. Robots need the ability to seek semantic cues beyond sensory horizon, exploit them to generate active behavior and do it robustly and efficiently. To this end, this workshop focuses on active exploration and search for semantic cues and their exploitation for generating a more efficient and informed behavior.

The workshop is a second edition of the successful event organized during IROS’11 in San Francisco which attracted considerable attention.  The theme of this year’s edition builds upon the last year’s introduction to active perception and focuses on the problem of acquisition and exploitation of contextual information as well as general world knowledge. It will bring together researchers in such fields as robotics, computer vision and neuropsychology in order to exchange ideas and discuss opportunities in using context and semantic prior for building active robotic systems.

List of Topics

This edition of the annual workshop series focuses on acquisition and use of contextual information and semantic world prior in active perception. Topics of interest include, but are not necessarily limited to:
  • Extracting semantic information from visual sensors, 3D sensors and other modalities
  • Object detection and categorization in complex scenes
  • Semantic mapping, place categorization and scene understanding
  • Combining heterogeneous sources of semantic information
  • Acquisition of semantic priors through analysis of large-scale databases
  • Exploiting context for object, scene and place understanding
  • Using semantic knowledge for driving active perception and efficient robot behavior
  • Planning for active perception and exploration
  • Trading exploration and exploitation of semantic cues
  • Closing the perception action loop with semantic information
  • Object search using vision or other sensory modalities
  • Multi-robot scenarios in active semantic perception

Motivation

Semantic perception has seen a lot of progress recently, with many new and interesting techniques being developed in parallel by different research groups. Moreover, with the advent of inexpensive and accurate RGBD sensors, there has been an explosion of new approaches utilizing the potential of 3D information. At the same time, we have witnessed an increased interest in integrated robotic systems operating in unstructured human environments and applications of assistive and service robots in the real-world.

By exploiting the dependencies between objects, places and actions, the robot’s understanding of the world can be greatly enhanced and its behavior can be planned more efficiently. At the same time, in large-scale environments, the perception and processing of semantic cues should by itself be an active process. It cannot be assumed that all the required information is available within the immediate sensory reach of the robot and an exhaustive search over the whole environment becomes unfeasible in large-scale scenarios. Robots need the ability to seek semantic cues beyond sensory horizon, do it robustly and efficiently. To this end, this workshop goes beyond pure semantic perception or mapping. We focus on active exploration and search for semantic cues and their exploitation for generating a more efficient and informed behavior.

The theme of this workshop builds upon the last year’s introduction to active perception and focuses on the problem of acquisition and exploitation of contextual information as well as general world knowledge. Years of research in such areas as human perception, cognition and neuroscience showed the importance of context in object categorization, scene understanding, exploration of large-scale spaces etc.. Many of the insights have already been exploited by computer vision and robotics researchers for the purpose of building more robust and efficient systems. Furthermore, the availability of huge amounts of information stored in Internet databases, social and photo sharing services or through crowdsourcing opened new possibilities for building robotic systems equipped with a large body of general world knowledge. This edition of the workshop aims at bringing together researchers in such fields as robotics, computer vision and neuropsychology in order to exchange ideas and discuss the opportunities in using context and semantic prior for building active robotic systems.

Program

 09:00 – 09:10 Introduction (by the workshop organizers)
 09:10 – 09:30 Dennis Stampfer, Matthias Lutz and Christian Schlegel Informed Active Perception with an Eye-in-hand Camera for Multi Modal Object Recognition
 09:30 – 10:00 Invited talk: Ingmar Posner What is Where? And Why Robots Care. Exploiting Robust Decision-Making, Planning and Exploration for Robot Perception
 10:00 – 10:30 COFFEE BREAK
 10:30 – 11:00 Invited talk: Matthew Walter and Stefanie Tellex Acquiring Rich Models of Objects and Space Through Vision and Natural Language
 11:00 – 11:20 Vladyslav Usenko, Florian Seidel, Zoltan-Csaba Marton, Dejan Pangercic and Michael Beetz Furniture Classification using WWW CAD Models
 11:20 – 11:40 Christian A. Mueller, Paul G. Plöger and Matthew S. Roscoe
Towards Scalable 3D Object Shape Categorization
 11:40 – 12:00 Sorin Mihai Grigorescu, Dejan Pangercic and Michael Beetz
2D-3D Collaborative Tracking (23CT): Towards Stable Robotic Manipulation
 12:00 – 14:30 LUNCH BREAK
 14:30 – 15:00 Invited talk: Patric Jensfelt
 14:00 – 14:30 Invited talk: Ashutosh Saxena
 15:00 – 15:20 Grace Tsai and Benjamin Kuipers Toward visual semantic modeling of the local environment for an indoor navigating robot
 15:20 – 15:40 Alexander J. B. Trevor, Akansel Cosgun, Jayasree Kumar and Henrik I. Christensen
Interactive Map Labeling for Service Robots
 15:40 – 16:00 Ayoung Kim and Ryan Eustice Next-Best-View Visual SLAM for Bounded-Error Area Coverage
16:00 – 16:30  COFFEE BREAK
 16:30 – 17:00 Invited talk: Carlo Ciliberto, Sean Ryan Fanello, Fabrizio Smeraldi, Lorenzo Natale and Giorgio Metta
Visual Object Recognition by Self-supervision and Human Robot Interaction Robotics
 17:00 - 17:30 Panel discussion and closing remarks (by the workshop organizers)

Important Dates
  • Submission deadline: ** DEADLINE EXTENDED DUE TO DEMAND ** 28 August
  • Notification of acceptance: 7 September
  • Camera ready: 24 September
  • Workshop date: 7 October

Setup

The workshop will consist of a mix of invited speakers, 2 or 3 paper presentations and a poster session. All submissions should be in PDF format using the IROS template. Papers should be 4-6 pages in length, optionally accompanied by a video. 

All submissions will be reviewed for relevance to the topic of the workshop as well as scientific merit. Submissions be made through the easychair system following this link: http://www.easychair.org/conferences/?conf=irosasp12

Please contact the organizers at irosasp12@easychair.org for questions/ideas etc.

Presenters and Program Committee

Invited talks are going to be given by (in alphabetical order):
  • Patric Jensfelt, Royal Institute of Technology KTH, Stockholm, Sweden. Use of prior semantic knowledge for object search in large-scale environments.
  • Giorgio Metta, Italian Institute of Technology, Genoa, Italy Biological and developmental perspective on active perception.
  • Ingmar Posner, Oxford University, UK Using semantic knowledge for exploration in large scale spaces.
  • Ashutosh Saxena, Cornell University, NY, USA Scene understanding and its applications in personal robotics.
  • Matthew Walter and Stefanie Tellex, MIT, Boston, USA Using semantic knowledge for symbol grounding in navigation and manipulation.

In addition to invited talks, we accept high quality submissions that are both peer reviewed and selected by the workshop’s program committee. It is formed by:
  • Wolfram Burgard, University of Freiburg, Germany
  • Tom Duckett, University of Lincoln, UK
  • Jana Kosecka, George Mason University, USA
  • Andreas Nuechter, Jacobs University Bremen, Germany
  • Dejan Pangercic, Technische Universität München, Germany
  • Alessandro Saffiotti, Örebro University, Sweden
  • Marcus Vincze, Vienna University of Technology, Austria

Organizers

  • Alper Aydemir Centre for Autonomous Systems, KTH Royal Institute of Technology, Teknikringen 14, 11428 Stockholm, Sweden Phone: +46-8-790-6725 E-mail: aydemir@kth.se
  • Dirk Holz Autonomous Intelligent Systems Group, University of Bonn, Friedrich-Ebert-Allee 144, 53113 Bonn, Germany Phone: +49-228-73-4434 E-mail: holz@ais.uni-bonn.de
  • Radu Bogdan Rusu Willow Garage / Stanford University, 68 Willow Road, Menlo Park, CA 94025, USA Phone: +1-650-475-2700 E-mail: rusu@willowgarage.com
  • Andrzej Pronobis Centre for Autonomous Systems, KTH Royal Institute of Technology, Teknikringen 14, 11428 Stockholm, Sweden Phone: +46-8-790-6724 E-mail: pronobis@csc.kth.se