From Game Characters to Game Companions: Practice, Anticipation, and Embodiment

Speaker: Guy Hoffman
Host: Prof. Gerard Medioni Abstract:
Aiming to make computer game agents more engaging and longer-lasting, I propose the notion of "game companions", based on three notions: (a) viewing game play as a collaborative activity, highly coordinated with a human player; (b) modeling practice to improve the agents' skills over time; and (c) moving game characters into the physical world, in the form of personal robots. Framed within an overall methodology that optimizes for engagement instead of efficiency, this talk introduces computational models for two aspects of practice: anticipation and embodiment, and their implementation on game agents and collaborative robots. First, I present an extension to Markov Decision Processes, modeling anticipation for agents collaborating with humans. I discuss the implementation of the model on a game character, which adapts to a human collaborator through repetition. I demonstrate the effect of anticipation on repetitive teamwork by showing this agent to be more efficient and fluent, and to seem significantly more committed and contributing to the team, as rated by human players. Then, I extend this discrete state/action model to a cognitively plausible embodied framework operating on continuous sensory input and performing real-world motor activity. Based on recent neuro- psychological findings supporting an embodied view of cognition, I model practice as priming perceptual simulation and affecting Hebbian learning. Implemented on two robots, I show this framework to have significant effects on both the efficiency of a human-robot team, and on human subjects' perception of the robot's intelligence, fluency, and even gender. With our aims set on long-term engaging interaction with artificial agents, these systems' capacity to physically practice with humans may hold a key to the kind of efficient and satisfying performance that humans are accustomed to from each other. Biography:
Guy Hoffman is a postdoctoral research associate at the Georgia Institute of Technology Center for Music Technology. Prior, he was a postdoctoral associate at MIT. Hoffman holds a Ph.D. from the MIT Media Lab in the field of Human-Robot Interaction and an M.Sc. in Computer Science from Tel Aviv University. His research investigates practice, anticipation, and joint action between humans and artificial agents, with the aim of designing agents that display more fluent behavior with their human partners. Other research interests include theater and musical performance robots as well as non-anthropomorphic agent design. His robot AUR won the 2007 IEEE International Robot Design Competition; he was animation and software lead on the 2008 World Expo Digital Water Pavilion, one of TIME magazine's "Best Inventions of 2007"; his research has been covered in the international press, including the New York Times, Süddeutsche Zeitung, and Science; Hoffman has held several senior research and product development positions in the Israeli software technology industry; and his work in animation, data visualization, architecture, theater, and new media art has been internationally published and exhibited.

Location: Seaver Science Library (SSL) - 150

Audiences: Everyone Is Invited

Contact: CS Colloquia