ICAR 2015 Workshop on Robot Learning
Bottom-up and top-down development of robot skills
Istanbul, Turkey, July 31st, 2015
With the recent advances in machine learning and computational power, robots now can acquire
impressive skills via learning. However, these learned skills are often isolated and limited to particular
levels of behavior hierarchy with no or very little interaction. In fact, the state of the art approaches in
developmental robotics can be broadly divided into two: those approaches that emphasize the
emergence of sensorimotor and cognitive knowledge through developmental progression in a bottomup
fashion, and others that assume existence of a high-level knowledge for complex reasoning, and
try to either ground this knowledge in the continuous sensorimotor world of the or guide the
sensorimotor learning with this knowledge robot in a top-down fashion. Human learning is a never
ending process that involves variety of skills represented at different sensorimotor and cognitive levels
that interact both horizontally and vertically. To capture the dynamics of horizontal (e.g. transfer of
knowledge from one task to another) and vertical (e.g. scaffolding provided by the higher level and
pattern formation at the lower level) interaction among learned skills, we need a paradigm shift on how
we design the learning systems of our robots to create a bootstrapping effect to accelerate learning.
Motivation and Objectives
The aim of this workshop is to bring experts and young researchers who study different aspects of
learning in robotics, and to create an environment for cross-fertilization with the aim of investigating
how we can create synergies so that bottom-up and top-down learning can interact to create a
bootstrapping effect in different levels of sensorimotor and cognitive development.
First of all, our invited speakers who work in different aspects of robot learning, such as visuomotor learning, symbol and language acquisition, and multi-modal learning, will present their research, along with an overview of the current state-of-the-art in the respective fields. Second, we will encourage our speakers to explicitly link their learning work to others'. We will motivate our speakers to comment on how bottom-up and top-down processes can be coupled in order to accelerate learning at both levels. While all sorts of interaction between learning systems in different fronts are important, we plan to explicitly address the following questions:
- How can robots form categories, concepts, logical-rules; acquire the skill of predicting effects of self and other's actions based on the formed structures; and use these abilities for high-level reasoning?
- How can robot's learning be scaffolded with direct teaching, teaching by demonstration, teaching by imitation using different modalities and cues?
- How the high-level abstract knowledge of the robot, innate or learned from experience, can regulate robot's exploration in developing further concepts and learning complex sensorimotor capabilities?
We believe that the question of how to create interaction mechanisms among different learning
components of a robot to create a bootstrapping affect in learning and skill acquisition will be central in
creating robots with human-like cognitive abilities. With this workshop, we expect to plant the initial
seeds for a research community to investigate and develop such interaction mechanisms.
The topics that are indicative but by no means exhaustive are as follows:
- Computational approaches to the study of development and learning
- Scaffolding in development
- Human assisted learning
- Machine learning techniques for robot learning and development
- Imitation learning
- Cognitive and perceptual development
- Natural and artificial intelligence
- Universal intelligence
- Embodied cognition
- Exploration and learning in animals and robots
- Interactive learning
- Social and emotional development in humans and robots
- Theory of mind
- Language acquisition
- Skill acquisition
- Curiosity and intrinsic motivation
- Dynamical systems
- Attention mechanisms in development
- Symbol grounding
- Concept formation
- Developmental disorders
- Affordance learning
- Human-robot interaction
- Sensorimotor learning and development
- Biologically inspired robotics
- Emre Ugur, Innsbruck University, Austria.
- Lorenzo Jamone, Instituto Superior Tecnico, Portugal.
- Yukie Nagai, Osaka University, Japan.
- Erhan Oztop, Ozyegin University, Turkey.