Active Learning in Robotics: Exploration Strategies in Complex Environments

Summary and Objectives

Robotics has achieved tremendous results: Robots can drive cars, precisely build goods at industrial scale, and perform surgery. The design of agents that can learn such complex skills by themselves is still in its infancy. One bottleneck is the lack of labeled data. Datasets are often limited in size and coverage, since they need experiments on robots or expensive simulations. Thus an important capability of every robot in complex environments is to gather new data.

To do this efficiently, intelligent robots need to be able to identify which task-relevant information is still missing. Only this ability enables agents to actively work towards a better understanding of the environment, the agent's state, and the task at hand, which will eventually lead to better performance.

Active learning is an approach to gain this ability. Within this paradigm the agent chooses the next datapoint to achieve the best learning result with as few data points as possible. Such strategies can shorten the process of data gathering significantly and also lead to the important information more quickly. In robotics, active learning approaches can be applied on various levels, from low-level tasks such as motor control learning to higher level reasoning tasks.

In this workshop, we want to discuss the state of the art of active learning in robotics, but also address important questions that are still open. The topics discussed in this workshop include (but are not limited to):

• What representation of knowledge allows efficient reasoning about it?
• What are successful strategies based on these representations?
• How can we generalize or transfer experiences to decrease the need of data?
• How to explore safely without damaging the robot and its environment?
• How can existing approaches be `scaled up' to large, continuous real world domains?

Intended Audience

The workshop is intended for all scientists conducting research in robotics or machine learning facing the problem of little data in robotics and those interested in exploration strategies for robots in novel situations.

Confirmed Speakers

We are happy to announce the following speakers have confirmed their participation in the workshop:

• Andreas Krause (ETH Zürich)
• Ruben Martinez-Cantin (Centro Universitario de la Defensa)
• Jivko Sinapov (University of Texas at Austin)
• Clement Moulin-Frier (Flowers Laboratory, INRIA Bordeaux)

Contributed Posters

The following contributed posters were accepted:

• Nicholas Kirk: Towards Learning Object Affordance Priors from Technical Texts
• Chang Wang, Koen V. Hindriks and Robert Babuska: Active Affordance Learning in Continuous State and Action Spaces
• Daisuke Tanaka, Takamitsu Matsubara, and Kenji Sugimoto: Optimal Control Approach for Active Exploratory Action Design with Object Manifold Learning
• Christian Daniel, Malte Viering, Jan Metz, Oliver Kroemer, Jan Peters: Learning Reward Functions for Efficient Robot Learning using Bayesian Optimization

Schedule

The tentative workshop schedule is as follows:

Organizers

• Johannes Kulick, University of Stuttgart
• Herke van Hoof, Technische Universität Darmstadt
• Marc Toussaint, University of Stuttgart
• Jan Peters, Technische Universität Darmstadt

Acknowledgement

The workshop receives funding from the DFG (German Science Foundation) within the priority programm "Autonomous Learning" (SPP 1527). The workshop organizers acknowledge support from the European Community’s Seventh Framework Programmes (FP7-ICT-2013-10) under grant agreement 610878 (3rdHand), 610967 (TACMAN), and 270327 (CompLACS).