Event Location: NSH 3002

Abstract: As we reach out to explore the universe time, distance, and danger compel us to use robotic proxies to conduct exploration missions. Limitations on or the absence of communications with our robotic explorers demand science autonomy. While exploring robots will encounter stimuli that are valuable for completing mission objectives. Robotic explorers will have to identify interesting stimuli for investigation and rank the worth of those stimuli with respect to the mission objective. The premise of the proposed work is that novelty of sensory stimuli is the best director of exploratory behaviour. For the purposes of this research novelty is a combination both surprise – stimuli that are statistically unlikely – and habituation – the process of becoming accustomed to sensory input.

The use of novelty as a guide for exploration allows robotic explorers to operate unsupervised in the long term. The proposed research presents a biologically grounded framework for exploration in the context of science autonomy. The biological models employed are parallelizable, allowing the exploitation of parallel hardware, and are amenable to an on-line setting, allowing a robotic explorer to learn about the world while navigating through it. The results of preliminary work in the identification of interesting stimuli will also be presented. Finally a plan for the development of the novelty detection algorithms with be expounded with respect to incorporating interest detection algorithms.

Committee:David Wettergreen, Chair

Jeff Schneider

Dean Pomerleau

David Thompson, Jet Propulsion Laboratory