Intrinsically Motivated Open-Ended Learning in Autonomous Robots

102.3K

views

58

authors

19

articles

Intrinsically Motivated Open-Ended Learning in Autonomous Robots

102.3K

views

58

authors

19

articles

Editorial

17 January 2020

Editorial: Intrinsically Motivated Open-Ended Learning in Autonomous Robots

Vieri Giuliano Santucci

,

Pierre-Yves Oudeyer

,

Andrew Barto

and

Gianluca Baldassarre

5,768 views

21 citations

Editors

4

Vieri Giuliano Santucci

National Research Council (CNR)

Pierre-Yves Oudeyer

Institut National de Recherche en Informatique et en Automatique (INRIA)

Andrew Barto

University of Massachusetts Amherst

Gianluca Baldassarre

National Research Council (CNR)

Impact

About

This Research Topic aims to present state-of-the-art research on open-ended learning in autonomous robots driven by intrinsic motivations, and surrounds the information presented at the upcoming third international workshop on Intrinsically Motivated Open-ended Learning (IMOL2017), which will be held in Rome, Italy, October 4-6: http://www.imol-conf.org/.

Artificial Intelligence (AI) is improving at an impressive pace. Sophisticated robots and powerful algorithms able to perform increasingly complex tasks are being developed every year. They are not only able to perform complex and lengthy tasks, but they can also discover better, or totally new, ways to support human activity and problem solving. Impressive new technologies, such as self-driving cars and sophisticated image and speech recognition systems have achieved significant breakthroughs, and their impact on human activities is expected to exponentially grow over the near future.

Regardless of the actual complexity of their behavior, the autonomy and versatility of current artificial agents are still limited compared to what biological agents are capable of. This lack of autonomy in present robots and systems prevents them from fully succeeding in interacting with realistic environments where they have to face situations that are unknown at design-time, where learning needs to be multi-task, incremental, and online, or when new goals/tasks have to be discovered and solved autonomously.

Over the last decade, intrinsically motivated learning (sometimes called “curiosity-driven learning”)
has been studied by many researchers as an approach to autonomous lifelong learning in machines. Intrinsic motivations are inspired by human ability to discover how to produce “interesting” effects on the environment driven by self-generated motivational signals not related to specific tasks or instructions. The research in this field aims to develop agents that acquire, under the guidance of intrinsic motivations, repertoires of diverse skills that are likely to become useful later when specific tasks need to be performed.

Advancing our knowledge in this direction is currently at the forefront of AI research. Indeed, the
most impressive AI outcomes, especially those based on deep neural networks, mainly involve
perception: notwithstanding the initial successes of approaches such as deep reinforcement learning and end-to-end robot control, the capability and autonomy of intelligent artificial agents on the side of action and motivation are still in their infancies. Research on intrinsically motivated open-ended learning now has an unprecedented opportunity to substantially advance the state-of-the-art.

Progress can be made in the direction of autonomous formation of goals, development of
parameterized skills able to solve multiple tasks and to transfer/generalize knowledge between
them, and composition and organization of multiple goals and skills to form hierarchies able to
solve more complex problems.

The goal of this Research Topic is to bring together researchers who are dealing with different
issues on intrinsic motivations and open-ended learning, including but not limited to:
- Autonomous robots lifelong learning
- Multi-task reinforcement learning
- Deep reinforcement learning
- Intrinsic motivations
- Curriculum learning
- Goal self-generation
- Multiple task solution and parameterized skills
- Neural/probabilistic representations and abstractions
- Architectures for open-ended learning
- Goal-based skill learning
- Knowledge transfer and avoidance of catastrophic forgetting
- Compositionality and chunking
- Hierarchies of goals and skills
- Mitigating risks of real-world deployment of open-ended learning systems

Download ebook

PDF

EPUB

Share

Editors

Vieri Giuliano Santucci

National Research Council (CNR)

Pierre-Yves Oudeyer

Institut National de Recherche en Informatique et en Automatique (INRIA)

Andrew Barto

University of Massachusetts Amherst

Gianluca Baldassarre

National Research Council (CNR)

Impact

102,309 Total views

80,722 Article views

15,901 Article downloads

5,686 Topic views

Published In

Frontiers in Neurorobotics

Frontiers in Robotics and AI

Computational Intelligence in Robotics, Humanoid Robotics

About Frontiers Research Topics

With their unique mixes of varied contributions from Original Research to Review Articles, Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author.

Suggest a topic