Events & News
Colloquium
| Category | Lecture |
| Date | Thursday, September 2, 2010 |
| Time | 11:00 am |
| Location | GS 906 |
| Details | Faculty Hire Reclassification 10:45am: Light refreshments in Gould-Simpson 9th Floor Atrium |
| Speaker | Ian Fasel |
| Title | Assistant Research Professor |
| Affiliation | Computer Science Department and SISTA - University of Arizona |
Research on perception, action, and cognition in the CACTUS Lab (Curious Agents: Cognition, Thinking, and Understanding Systems) at the University of Arizona.
Abstract: Just like humans, a cognitive robot has to process massive amounts of sensory data from a variety of noisy, limited range sensors, and use this information to make decisions in real-time. Ultimately, we aim for cognitive robots to be able to interact with human beings to perform everyday tasks, ranging from assistance with physical manipulation in the home or lab (like cleaning house, or fetching and preparing materials for experiments) to helping educate children, or assist in caring for the disabled. Thus for a cognitive robot, using this noisy sensor data to make sense of the "mental" world of human beings will be just as important as understanding the "external" world, i.e., the other physical objects that it encounters.
The goal of the CACTUS Lab is to develop techniques for a robot (or other autonomous agent) to infer hidden state and make intelligent, real-time decisions about the physical and social world given its noisy sensors, using tools from machine learning, probability and statistics, information theory, and stochastic optimal control. In this talk, I will focus on two projects ongoing at the UofA. The first is called CLIME: Concept learning through intrinsically motivated sensory-motor experience. In this project, we develop agents that use an information theoretic "curiosity" mechanism to actively select how to interact with objects in order to reduce their uncertainty about the environment as quickly as possible, while simultaneously learning about object "affordances", and discovering perceptual and functional object categories. The second project, in collaboration with researchers in the Department of Linguistics, focuses on communication with humans by improving our understanding of the physical mechanisms of human speech production. To this end we are developing a multi-modal database and analysis software to enable the study of the interaction of the lips, velum, tongue, and larynx (using simultaneously captured audio, ultrasound, video, nasal airflow, and electroglottography). I will focus on our recent techniques for performing automatic extraction of the tongue contour from ultrasound and real-time coding of dynamic tongue gestures. This work is important both to basic linguistic and cognitive sciences and to applied research such as automatic speech recognition, speech therapy, language documentation, and even language and music education. I will conclude with an overview of some of the other projects in the CACTUS Lab and provide a view of how these different lines of research mutually support each other toward the goal of creating intelligent, autonomous agents that humans can interact with using natural, intuitive forms of communication.