Workshop: Beyond Robot Grasping - Modern Approaches for Learning Dynamic Manipulation

Quick Facts

We are preparing a special issue of the Autonomous Robots journal on Modern Approaches for Dextrous Manipulation. Take a look at the call for papers here!!

Abstract

The field of robot grasping and manipulation is reaching an important milestone. In recent years, we have seen various robots that can reliably perform basic grasps on unknown objects in unstructured environments. However, these robots are still far away from being capable of human-level manipulation skills such as in-hand or bimanual manipulations of objects, interactions with non-rigid objects, and multi-object tasks such as stacking and tool-usage. As such advanced manipulations involve interacting with uncertain real-world environments, they pose major problems for current approaches and traditional methods that depend on accurate models of the robot and its surrounding.

This workshop focuses on how modern machine learning techniques and sensor technologies can help robots go beyond basic grasping abilities towards advanced manipulation skills. In recent years, innovations from these fields have played a pivotal role in developing state-of-the-art approaches for robot grasping under uncertainty. By further refining and improving upon these approaches, we strive towards creating robots that are capable of complex manipulation skills in real-world environments.

The workshop is a core GeRT event supported by EUCog III and PASCAL2.

Format

The workshop will consist of presentations, posters, and panel discussions. Topics to be addressed include, but are not limited to:

• What is the state-of-the-art in robot learning of manipulations?
• How can we benefit from recent results in machine learning, e.g., structured learning, Gaussian Processes, Conditional Random Fields, etc?
• How can robots make use of reinforcement learning, or other self-improvement methods, to adapt to changing environments and tasks?
• How can robots learn to handle ambiguous sensory signals?
• How can robots model uncertainty in their surroundings and their actions?
• Which representations can leverage the acquisition of complete multimodal models of the environment?
• How can robots perform bimanual actions that are synchronized?
• How can robots determine optimal actions on non-rigid objects?
• How can robots learn to robustly detect the salient events in manipulation tasks, e.g. when objects make and break contact?
• What is the state of the art in robot hand technology?
• How much can we reliably learn from simulations?
• How can apprenticeship learning help to overcome the correspondence problem?
• How can robots remove and place complex objects in cluttered environments?
• How can we model finger synergies over longer action sequences?
• How can human task knowldege be efficiently transferred to robots?
• How can task-relevant features of objects be estimated?
• How can robots efficiently generalize a task from only a few human demonstrations?
• How can a robot represent compound objects; e.g. objects stacked on a tray or a bottle and a cap?
• How can the effects of actions be represented in a general form?
• What prior knowledge can a robot be expected to have?
• What are the key challenges and can we decide on benchmark tasks that allow us to measure and compare progress in this field?
• Which datasets and code components can be shared, to allow researchers to compare their approaches and build upon each other's work?

Confirmed Invited Speakers

Peter Allen (Columbia University) [ Slides]
Christoph Borst (DLR - German Aerospace Center)
Oliver Brock (Technical University Berlin) [Slides]
Aaron Dollar (Yale University)
Alberto Rodriquez (Carnegie Mellon University)
Jan Babic (Jožef Stefan Institute) [Slides]
Anis Sahbani (Université Pierre et Marie Curie)
Ashutosh Saxena (Cornell University)

Call for Posters

The field of robot grasping and manipulation is reaching an important milestone. In recent years, we have seen various robots that can reliably perform basic grasps on unknown objects in unstructured environments. However, these robots are still far away from being capable of human-level manipulation skills such as in-hand or bimanual manipulations of objects, interactions with non-rigid objects, and multi-object tasks such as stacking and tool-usage. As such advanced manipulations involve interacting with uncertain real-world environments, they pose major problems for current approaches and traditional methods that depend on accurate models of the robot and its surrounding.

This workshop focuses on how modern machine learning techniques and sensor technologies can help robots go beyond basic grasping abilities towards advanced manipulation skills. In recent years, innovations from these fields have played a pivotal role in developing state-of-the-art approaches for robot grasping under uncertainty. By further refining and improving upon these approaches, we strive towards creating robots that are capable of complex manipulation skills in real-world environments.

Will be about recent success as well as the next steps in robot grasp and manipulation. As an outcome of this workshop we expect the participating researchers to identify and address important challenges, techniques, and benchmarks that are needed to realize complex robot manipulation skills.

Program

08.30 - 09.00 Introduction
09.00 - 09.30 Invited talk: Anis Sahbani
09.30 - 09.45 Poster Spotlight Presentations
09.45 - 10.15 Invited talk: Peter Allen
10.15 - 10.30 Poster Spotlight Presentations
10.30 - 11.00 Coffee break and poster session
11.00 - 11.30 Invited talk: Christoph Borst
11.30 - 11.45 Poster Spotlight Presentations
11.45 - 12.15 Invited Talk: Jan Babic
12.15 - 12.30 Poster Spotlight Presentations
12.30 - 14.00 Lunch break
14.00 - 14.30 Invited talk: Aaron Dollar
14.30 - 14.45 Poster Spotlight Presentations
14.45 - 15.15 Invited talk: Alberto Rodriquez
15.15 - 15.30 Poster Spotlight Presentations
15.50 - 16.50 Coffee break and poster session
16.50 - 17.20 Invited talk: Ashutosh Saxena
17.20 - 17.50 Invited talk: Oliver Brock
17.50 - 18.00 Final comments

Accepted Papers

• Benjamin Balaguer. Learning and Optimizing Bimanual Regrasping Behaviors. [pdf]
• Erik Berger. Cooperative Human-Robot Manipulation Tasks. [pdf]
• Oliver Kroemer. Generalizing Manipulations using Vision Kernels. [pdf]
• Herke van Hoof. Maximally Informative Interaction Learning for Scene Exploration. [pdf]
• Qiang Li. Grasp Point Optimization for Unknown Object Manipulation in Hand Task. [pdf]
• Robert Paolini. A Data-Driven Statistical Framework for Post-Grasp Manipulation. [pdf]
• Mario Prats. Specification of Physical Interaction through Vision and Force-based Demonstration. [pdf]
• Tarek El-Gaaly. Multi-Modal RGBD Sensors for Object Grasping and Manipulation. [pdf]
• Mario Prats. On the Frontiers of Mobile Manipulation: the Challenge of Autonomous Underwater Intervention. [pdf]
• Jose A. Bernabe. Contact detection and location from robot and object tracking on RGB-D images. [pdf]
• Claudio Zito. Exploratory reach-to-grasp trajectories for uncertain object poses. [pdf]
• Tekin Mericli. Experience Guided Achievable Push Plan Generation for Passive Mobile Objects. [pdf]
• Susanne Petsch. Path Configuration for Abstractly Represented Tasks with Respect to Efficient Control. [pdf]
• Lars-Peter Ellekilde. Robust Peg-In-Hole Manipulation Motivated by a Human Tele-Operating Strategy. [pdf]
• Javier Gonzalez-Quijano. A Human-Based Genetic Algorithm Applied to the Problem of Learning in-Hand Manipulation Tasks. [pdf]
• Yasemin Bekiroglu. Learning Task- and Touch-based Grasping. [pdf]
• Sahar El-Khoury. Teaching Robots to Grasp through a User Friendly Interface. [pdf]
• Baris Özyer. Momentum Transfer From Preshape to Grasping. [pdf]
• Satoshi Makita. Geometrical Constraint in Grasping. [pdf]
• Ryuta Ozawa. Robust, Cheap and Dexterous Manipulation. [pdf]
• Lars Karlsson.Progress and Challenges in Planning for a Two-Arm Robot. [pdf]
• Bidan Huang. Learning a Real-Time Grasping Strategy. [pdf]
• External Force Estimation for Textile Grasp Detection. Adria Colome. [pdf]
• Alexander Herzog. Template-Based Exploration of Grasp Selection. [pdf]
• Filipe Veiga. Towards Bayesian Grasp Optimization with Wrench Space Analysis [pdf]

Important Dates

August 6st - Deadline of submission (Deadline Extension!!)
August 10th - Notification of Acceptance

Submissions

Extended abstracts (1 pages) will be reviewed by the program committee members on the basis of relevance, significance, and clarity. Accepted contributions will be presented as posters but particularly exciting work may be considered for talks. Submissions should be formatted according to the conference templates and submitted via email to beyond.grasping.iros2012@googlemail.com

Organizers

Heni Ben Amor, Technische Universitaet Darmstadt
Ashutosh Saxena, Cornell University
Oliver Kroemer, Technische Universitaet Darmstadt
Jan Peters, Technische Universitaet Darmstadt and Max Planck Institute for Intelligent Systems

Location and More Information

The most up-to-date information about the workshop can be found on the IROS 2012 webpage.