Robotics Science and Systems, June 27, Berlin
Applications of robots are expanding at a fast rate and are expected to operate in less controllable and harder to model domains. Learning and adaptation becomes essential to deploy robots that continuously interact with the environment, acquire new data during operation and use them to improve its performance by developing new skills or improving and adapting its models.
How should a robot acquire and use this stream of data? How can it close the action-perception loop to efficiently learn models and acquire skills? Researchers in robotics, statistics and machine learning have answered these questions from different perspectives and setups: active learning, submodular optimization, exploration strategies, multi-armed bandits among many others. All such approaches provide ways for the robot to choose better data to learn, reducing the time and energy used while at the same time improving generalization capabilities.
The goal of this workshop is to show how formalisms developed in different communities can be applied in a multidisciplinary context as it is robotics research. It will bring together researchers to build bridges between these different perspectives and to exchange ideas about representations and methods for active learning in robotics. In addition to the classical exploration problem, the workshop will also explore connections with new trends such as using intrinsic motivation to model curiosity and drive exploration towards the acquisition of unknown skills or the development of active strategies for human-robot interaction in the context of co-working or learning from a human teacher.
Active Learning, Reinforcement Learning, Markov Decision Processes, Exploration/Exploitation, Intrinsic Motivation
9:00 Manuel Lopes, INRIA, Bordeaux
9:30 Maya Cakmak, Willow Garage
10:00 poster round 1: p1-p3,
10:30 break with coffee
11:00 Marc Toussaint, University of Sttutgart
11:30 Kevin Gurney, Sheffield University
12:00 Pieter Abeel, UC Berkeley
14:00 Andreas Krause, ETHZ
14:30 Jan Peters, TU Darmstadt
15:00 poster round 2 p4-p6
15:30 break with coffee
16:00 poster round 2 p7-p9
16:30 Fabio Ramos, ACFR Sidney
17:00 Oliver Brock, TU Berlin
17:30 wrap-up discussion
18:00 End of workshop
- Andreas Krause, ETHZ
- Jan Peters, Darmstadt TU
- Pieter Abeel, UC Berkeley
- Fabio Ramos, ACFR Sidney
- Oliver Brock, TU Berlin
- Kevin Gurney, Sheffield University
- Marc Toussaint, Sttutgart University
- Manuel Lopes, INRIA, Bordeaux
- Maya Cakmak,Willow Garage
- P1: Rudolph Triebel, Hugo Grimmett, Rohan Paul and Ingmar Posner. Introspective Active Learning for Scalable Semantic Mapping
- P2: Ruben Martinez-Cantin. BayesOpt: A Toolbox for Bayesian optimization with Robotics Applications
- P3: Daniel Levine and Jonathan How. Distributed Sensor Selection in Gaussian Trees with Nuisances
- P4: Riad Akrour, Michèle Sebag and Marc Schoenauer. On-board Robot Interactive Training
- P5: Matthew Luciw, Konstantin Lakhman, Mathis Richter, Sohrob Kazerounian and Yulia Sandamirskaya. Learning the Perceptual Conditions of Satisfaction of Elementary Behaviors
- P6: Gustavo Martins, Helder Coelho and Paulo Urbano. A Theoretical Model for Self-Motivated Agents That Learn
- P7: Stanimir Dragiev, Marc Toussaint and Michael Gienger. Tactile Exploration and Grasping Unknown Objects by Actively Learning Gaussian Process Implicit Shape Potentials
- P8: Guido Schillaci, Verena Vanessa Hafner and Bruno Lara. I Know How I Would Do It. Internal Simulations of Sensorimotor Experience
- P9: Sao Mai Nguyen and Pierre-Yves Oudeyer, Active Choice of Teachers, Learning Strategies and Goals for a Socially Guided Intrinsic Motivation Learner
Poster spotlights & dimensions
We solicit The dimensions of the poster board are 120cm x 160cm.
The spotlight presentation should be around 5 minutes.
Call for contributions (CLOSED):
We solicit contributed presentations in all areas of active learning applied in robotics including, but not limited to: exploration, reinforcement learning, active sensing and perception, intrinsic motivation, active manipulation, human-robot co-working. The workshop aims to foster discussion between the different active strategies. Therefore, we will accept already published materials as well as unpublished work. Contributions will be evaluated in terms of its relevance to the workshop topic. Accepted contributions will be given an oral presentation or poster (plus spotlight talk) at the workshop.
Paper Submission: May 1, 2013
Paper Notification: May 15, 2013
Workshop Dates: June 27, 2013
Paper format and submission
RSS format, max. 6 pages, please use the template available at:
Submission can be done at:
Sponsored by the German Research Foundation.