Grounding oriented design

Publication Type:
Issue Date:
Full metadata record
The symbol grounding problem[67] is a longstanding, poorly understood issue that has interested philosophers, computer scientists, and cognitive scientists alike. The grounding problem, in its various guises, refers to the task of creating meaningful representations for artificial agents. After more than 15 years of widespread debate and circular introspection of the so-called symbol grounding problem we seem none-the-wiser as to what constitutes being meaningful, and indeed grounded, for an agent[l28]. We argue, in the context of practical robotics, a grounded agent possesses a representation which faithfully reflects pertinent aspects of the world. In contrast, an ungrounded agent could be, for example, delusional or suffering from hallucinations ("false positives"), overly concerned with irrelevant things (e.g. the frame problem[93]), or incapable of reliably perceiving, recognising or anticipating relevant things in a timely manner ("false negatives"). While most grounding research concerns how to develop agents which can autonomously develop their own representations (i.e. autonomous grounding), the fact all robotic systems are grounded through human design on a case-by-case, ad-hoc bass has been overlooked. This thesis presents Grounding Oriented Design - a methodology for designing and grounding the "minds" of robotic agents. Grounding Oriented Design (or, alternatively Go-Design) is a vital first step towards the development of autonomous grounding capabilities through improved understanding of the processes by which human designers ground robot minds. Grounding Oriented Design offers guidelines and processes for iteratively decomposing a robot control problem into a set of collaborating skills, together with a notation for representing and documenting skill designs. Grounding Oriented Design consists of two main phases: basic-design which involves constructing a skill-architecture, and a detailed-design in which a skill-architecture is used to design the agent's representation and decision-making processes. A groundedness framework[l43] is used for describing and assessing the groundedness of either the complete system or of individual skills. Examples of the methodology's use and benefits are provided, while suggestions for future work are discussed.
Please use this identifier to cite or link to this item: