The State Problem in Cognitive Robotics

Cognitive Robotics is a discipline whose origin lies in the problem of controlling robotic agents so as to make them accomplish all the required tasks. A crucial issue in this problem is that such control must not prevent the agent from behaving as autonomously as possible while operating in dynamic, partially unknown and unpredictable environments, without human assistance. In this context Cognitive Robotics addresses the design of an “embodied artificial intelligence”. This is turning out to be a major challenge, much harder to solve than the scientific community did reckon in the past, and demanding a great effort both in developing concepts, methodologies, tools and techniques, and in putting them all together. A main shared opinion is that the high-level control of autonomous robotic agents should be tackled by means of high cognitive functions, suitably designed or even learned by the agent itself. Many disciplines (Artificial Intelligence, Cognitive Sciences like Developmental and Behavioral Psychology, (Cognitive)Neuroscience etc.) provide insights in this respect. As to how cognitive functions should be obtained, and beyond it, many different, and sometimes opposite, approaches have been proposed. Some researchers think that such cognitive functions must not be directly taken into account, but will naturally arise from a suitably designed and tuned coupling of sensing and acting that can progressively grow in the complexity of their interactions. This view goes back to R. Brooks ([3], [4]) reactive architectures and has been being developed through behavioral ([22]) and other related approaches. What is peculiar of this approach is the lack of an explicit internal model representing the world, indeed the real world is taken as the ‘best model’ ([5]) for the agent. As opposite we find the model based approaches ([24]) that rely on the existence of such a representation, which is often symbolic and managed by means of logical languages and the related inference techniques. From the latter standpoint we focus our attention on one of these logical frameworks, the Situation Calculus ([20], [26], [32]), well known among those addressing the problem of reasoning about actions and change.