APERoC 2012
first international workshop on
Action for Perception: Endowing Robots with Cognition

Motivation

As robots are slowly but decisively moving from controlled environments to unpredictable human-inhabited places, they are facing an increasing need for understanding their surroundings and for adapting their actions to an ever growing range of situations. To fulfill this need, we believe that actions and the assessment of their sensory consequences should get a more central role in cognitive robotics.

Although a more central role of action has been advocated already twodecades ago within behavior-based robotics, e.g., [1-2], many current robotic systems are still based on the classical sense-think-act view on cognition. Such a system often consists of sophisticated perception mechanisms that turn sensory observations into internal representations (sense). Based on these representations, the system deliberates about an action plan (think), and executes this plan (act). Although this classical view has led to many interesting techniques and applications, it misses some essential parts of cognition, namely that perception not only precedes action, but that action also precedes perception. Through the environment,actions have immediate consequences on the sensory observations. By using actions, a robotic system can influence its own sensory experience, thereby simplifying many perceptual tasks, and allowing to learn from the sensorimotor experience.

Different from the behavior-based robotics view, we do not propose a central role for action to circumvent the pitfalls of internal representations, but because we believe that actions are crucial in perception and the understanding of the world. Moreover, we believe that actions play a key role in developing more complex cognitive capabilities. As a weak claim, we pose that action is important to enable or simplify perceptual tasks, as in the sense of active perception [3,4], but as a stronger claim, we also pose that without an emphasis on action, a system cannot truly understand and act in the world since it does not understand the sensorimotor relations that govern the system's interaction with the world. And it is the understanding of these relations that enable the system to choose actions in order to behave coherently. To use the example of O'Regan \cite{ORegan01}:a system can only truly understand what asponge is if it can experience the sponginess by squeezing the object and observing the sensory consequences. Having this understanding allows the system to grasp and use the sponge correctly.

An important aspect is that these sensorimotor relations are specific to the system, its morphology, and the interaction with its environment. It is therefore crucial for the system to learn about the relations from its own experience.

Judging from current ongoing European projects, such as eSMCs (extending sensorimotor contingencies to cognition) and XPERIENCE, we feel that there is a growing interest in the robotic community to address these topics. With this workshop, we provide a platform to bring interested researches together.

As an outcome, we expect that the attendees will contribute to an overview of the latest trends in robot cognition, specifically in sensorimotor integration, seen under a variety of perspectives. We aspire in a constructive osmosis of ideas that will enrich the currentstate-of-the-art of robotic systems that autonomously understand and use the sensorimotor relations governing the system's interaction with the world.

References

  1. R. Brooks, Intelligence without representation, Artificial Intelligence, vol. 47, no. 1-3, pp. 139--159, 1991.
  2. R. C. Arkin, Behavior-Based Robotics. Massachusetts Institute of Technology, 1998.
  3. J. J. Gibson, The Ecological Approach to Visual Perception. Boston: Houghton-Mifflin, 1979.
  4. D. H. Ballard, Animate vision, Artificial Intelligence, vol. 48, pp. 57--86, 1991.
  5. J. O'Regan and A. Noe, A sensorimotor account of vision and visual consciousness, Behavioral and Brain Sciences, vol. 24, no. 5, pp. 939--1031, 2001.
Lazaros Nalpantidis, Renaud Detry, Gert Kootstra,
Alexander Maye, Justus Piater, Tamim Asfour, Danica Kragic