Special corner on “cognitive robotics”

Abstract

Cognitive Robotics (CR) is a theoretical and practicalapproach to implement cognitive abilities for reasoning,perception and action on technical robots. It aims at robotsthat cognize their worlds and can reason, e.g., about goals,beliefs, actions, when to perceive and what to look for, thecognitive states of other agents, or collaborative task exe-cution. This research line originated as a counterpart towork in classical robotics and automation that focuses onengineering tasks like sensory processing, path planning,manipulator design and control, usually with prepro-grammed, task-specific models. Reminiscent of the rise ofclassical Artificial Intelligence, which started out from‘‘heuristic programming’’ (cf. Michael 1972), CR startedout as a form of ‘‘behavior programming’’ but hasemphasized high-level primitives that rest upon principlesof human-like perception and action, and which is based oninternal models (e.g., representations) going beyond fixedbehavior-based architectures.In this sense, cognitive robots embody the behavior ofintelligent ‘‘Cartesian agents’’ in the physical world (or avirtual world, in the case of simulated CR). But this callsfor more than putting A.I. reasoners on robotic devices.While traditional cognitive modeling approaches haveassumed complex symbolic representations as a meansfor capturing the world, CR can be seen as an implemen-tation of minimal robust representationalist models ofintelligence in actual embodied, technical devices. Theseminimal models and systems result from an amalgamationof classical representationalist Cognitive Science approa-ches and ‘‘new’’ robotics or dynamical systems approaches(Clark and Grush 1999). In this way, CR creates anopportunity for studying how Cognitive Science and A.I.accounts can (and must) be deeply grounded in real-worldphysical embodiment and situatedness, and—the other wayaround—how robots facing real-world problems can beendowed with necessary, robust cognitive skills.With this special corner, we wish to further strengthenthe interaction between Cognitive Science, ArtificialIntelligence and Robotics as it figures fruitfully in the fieldof CR. From our point of view, CR has only begun toreveal its full potential for fertilizing work in the theoretical,technical, and empirical sciences targeting the under-standing of cognition. Today’s cognitive robots are stillrather limited, and many challenging topics lie ahead:• Understanding the ‘‘deep grounding’’ of cognitiveabilities in technical bodies that are not vehicles butparts of the cognitive processes. This addresses theintersection of embodied cognition and robotics and hasled, e.g., to work on compliant robots or ‘‘morpholog-ical computation’’ (Pfeifer et al. 2006), where con-straints of biomimetic bodies provide a lot of‘‘thinking’’ for the mind, but which is still to arrive atthe integration of higher cognitive skills with artificialbodies.• Exploring and exploiting further the role of learning,from learning motor skills (Rolf et al. 2010)toaprincipled approach for the acquisition of cognitiveabilities as in the subfield of ‘‘cognitive developmentalrobotics’’ (Asada et al. 2001; Cangelosi et al. 2010).• Incorporating language and communication bystrengthening the links between linguistics, artificial